[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"portal-settings:stajic:es":3,"public-menus:all":38,"post:openai-agents-api-vs-agents-sdk-vs-responses-api-what-should-you-build-on-in-2026:es":205,"related:post:openai-agents-api-vs-agents-sdk-vs-responses-api-what-should-you-build-on-in-2026:es:1":1590},{"statusCode":4,"data":5,"message":37},200,{"tenantId":6,"lang":7,"defaultLang":8,"siteUrl":9,"contactEmail":10,"brandName":11,"logoUrl":12,"siteName":11,"siteDescription":13,"ogImage":10,"robotsIndex":14,"socialLinks":10,"reservedSlugs":10,"seoPolicy":15},"stajic","es","de","https:\u002F\u002Fstajic.de",null,"Stajic Platform","\u002FLogo_Planet.svg","Stajic Portal",true,{"branding":16,"relatedContent":17,"crossDomainLinks":18},{"logoUrl":12},{"enabled":14},[19,22,25,28,31,34],{"url":20,"label":21,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Ffigure.rocks","figure.rocks",{"url":23,"label":24,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Floving.rocks","loving.rocks",{"url":26,"label":27,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Fbazify.com","bazify.com",{"url":29,"label":30,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Fbazify.de","bazify.de",{"url":32,"label":33,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Fbazify.at","bazify.at",{"url":35,"label":36,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Fbazify.ba","bazify.ba","Portal settings resolved",[39,45],{"id":40,"name":41,"location":42,"isActive":14,"isDefault":43,"items":44},1,"main-navigation","header",false,[],{"id":46,"name":47,"location":48,"isActive":14,"isDefault":14,"items":49},4,"main-menu","sidebar",[50,66,79,93,103,118,133],{"id":51,"title":52,"url":60,"target":61,"icon":62,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":64,"portfolioId":10,"children":65},"item-18",{"de":53,"en":54,"es":55,"fr":56,"it":54,"ru":57,"sr":58,"zh":59},"Startseite","Home","Inicio","Accueil","Главная","Почетна","首页","\u002Ffull-stack-web-developer-munich-performance-seo-and-maintainable-builds","_self","i-lucide-home","page",111,[],{"id":67,"title":68,"url":75,"target":61,"icon":76,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":77,"portfolioId":10,"children":78},"item-22",{"de":69,"en":69,"es":70,"fr":69,"it":71,"ru":72,"sr":73,"zh":74},"Vision","Visión","Visione","Видение","Визија","想象","\u002Fueber-uns-webdesign-muenchen-webaplikation","i-lucide-eye",113,[],{"id":80,"title":81,"url":89,"target":61,"icon":90,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":91,"portfolioId":10,"children":92},"item-19",{"de":82,"en":83,"es":84,"fr":83,"it":85,"ru":86,"sr":87,"zh":88},"Leistungen","Services","Servicios","Servizi","Услуги","Услуге","服务","\u002Fservices-dienstleistungen-muenchen","i-lucide-wrench",116,[],{"id":94,"title":95,"url":99,"target":61,"icon":100,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":101,"portfolioId":10,"children":102},"item-23",{"de":96,"en":96,"es":96,"fr":96,"it":96,"ru":97,"sr":97,"zh":98},"Blog","Блог","博客","\u002Fblog","i-lucide-book-open",112,[],{"id":104,"title":105,"url":114,"target":61,"icon":115,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":116,"portfolioId":10,"children":117},"item-32",{"de":106,"en":107,"es":108,"fr":109,"it":110,"ru":111,"sr":112,"zh":113},"Neue Technologien","New Technologies","Nuevas tecnologías","Nouvelles technologies","Nuove tecnologie","Новые технологии","Нове технологије","新技术！","\u002Fneue-webtechnologien","i-lucide-sparkles",122,[],{"id":119,"title":120,"url":129,"target":61,"icon":130,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":131,"portfolioId":10,"children":132},"item-20",{"de":121,"en":122,"es":123,"fr":124,"it":125,"ru":126,"sr":127,"zh":128},"Kontakt","Contact us!","Contacto","Contact","Contatto","Контакт","Контактирајте нас","联系我们！","\u002Fcontact","i-lucide-mail",115,[],{"id":134,"title":135,"url":144,"target":61,"icon":145,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":146,"portfolioId":10,"children":147},"item-21",{"de":136,"en":137,"es":138,"fr":139,"it":140,"ru":141,"sr":142,"zh":143},"Unsere Arbeit","Our Work","Nuestro trabajo","Nos réalisations","I nostri lavori","Наши работы","Наши радови","文件夹","\u002Fportfolio","i-lucide-briefcase",114,[148,161,175,181,193],{"id":149,"title":150,"url":144,"target":61,"icon":159,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":146,"portfolioId":10,"children":160},"item-24",{"de":151,"en":152,"es":153,"fr":154,"it":155,"ru":156,"sr":157,"zh":158},"Alle Projekte","All Projects","Todos los proyectos","Tous les projets","Tutti i progetti","Все проекты","Сви пројекти","所有项目","i-lucide-grid-3x3",[],{"id":162,"title":163,"url":171,"target":61,"icon":172,"isActive":14,"type":173,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":10,"portfolioId":10,"children":174},"item-29",{"de":164,"en":165,"es":166,"fr":167,"it":168,"ru":169,"sr":170,"zh":143},"Local Roots, Global Reach","Local Roots - Global Reach","Empresa local ","Entreprise locale","Azienda locale","Местная компания","Локално предузеће глобално тржиште","\u002Fportfolio\u002Flocal-roots-global-reach-communication-media-systems-for-modern-business","i-lucide-folder","custom",[],{"id":176,"title":177,"url":179,"target":61,"icon":172,"isActive":14,"type":173,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":10,"portfolioId":10,"children":180},"item-28",{"de":178,"en":178,"es":178,"fr":178,"it":178,"ru":178,"sr":178,"zh":178},"Solr Suggester","\u002Fportfolio\u002Fsolr-fuzzy-suggester-und-solr-infix-suggester-abfrage-ueber-ajax-und-filterung",[],{"id":182,"title":183,"url":191,"target":61,"icon":172,"isActive":14,"type":173,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":10,"portfolioId":10,"children":192},"item-27",{"de":184,"en":185,"es":186,"fr":187,"it":188,"ru":189,"sr":190,"zh":185},"Firmenwebseite SEO","Company Website SEO","Sitio web corporativo SEO","Site web d’entreprise SEO","Sito web aziendale SEO","Корпоративный сайт SEO","Пословна веб-страница SEO","\u002Fportfolio\u002Fseo-sem-branding-mobile-webseite-muenchen",[],{"id":194,"title":195,"url":203,"target":61,"icon":172,"isActive":14,"type":173,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":10,"portfolioId":10,"children":204},"item-31",{"de":196,"en":197,"es":198,"fr":199,"it":200,"ru":201,"sr":202,"zh":197},"Digitalisierungsportal","Digitalization Portal","Portal de digitalización","Portail de numérisation","Portale di digitalizzazione","Портал цифровизации","Портал за дигитализацију","\u002Fportfolio\u002Fdigitalisierungsportal-archiv-museum-bibliothek-ead-lido-mets-mods",[],{"statusCode":4,"data":206,"message":1589},{"id":207,"title":208,"slug":209,"content":210,"contentJson":211,"excerpt":799,"featuredImage":800,"featuredImageAlt":801,"featuredImageCaption":10,"featuredImageTitle":10,"featuredImageCopyright":10,"featuredImageAuthor":10,"featuredImageSourceUrl":10,"featuredImageLicense":10,"featuredImageIsAiGenerated":43,"status":802,"publishedAt":803,"createdAt":804,"updatedAt":805,"seoLocalePaths":806,"categories":815,"author":827,"translations":832},"473","OpenAI Agents API vs Agents SDK vs Responses API: ¿Sobre qué deberías construir en 2026?","openai-agents-api-vs-agents-sdk-vs-responses-api-what-should-you-build-on-in-2026","\u003Cnav class=\"editorjs-toc\" data-editorjs-toc=\"true\" aria-label=\"Índice\">\u003Cstrong class=\"editorjs-toc__title\">Índice\u003C\u002Fstrong>\u003Col class=\"editorjs-toc__list editorjs-toc__list--depth-0\">\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-6\" class=\"editorjs-toc__link\">La arquitectura ha cambiado: elija un límite de runtime, no una biblioteca\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-9\" class=\"editorjs-toc__link\">Comparativa resumida\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-11\" class=\"editorjs-toc__link\">1. Agents API: harness gestionado, agente duradero en la nube\u003C\u002Fa>\u003Col class=\"editorjs-toc__list editorjs-toc__list--depth-1\">\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-15\" class=\"editorjs-toc__link\">Qué se obtiene con la Agents API\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-17\" class=\"editorjs-toc__link\">Qué sigue bajo su control\u003C\u002Fa>\u003C\u002Fli>\u003C\u002Fol>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-19\" class=\"editorjs-toc__link\">2. API Responses: controle el bucle, use las primitivas de la plataforma\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-23\" class=\"editorjs-toc__link\">3. Agents SDK: con soporte, pero ya no es la vía predeterminada hacia el futuro\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-28\" class=\"editorjs-toc__link\">Dónde encaja el Codex SDK\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-31\" class=\"editorjs-toc__link\">La prueba de propiedad del entorno de ejecución\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-34\" class=\"editorjs-toc__link\">Un árbol de decisión para nuevos sistemas\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-36\" class=\"editorjs-toc__link\">Qué no debería motivar la decisión\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-38\" class=\"editorjs-toc__link\">La migración es un cambio arquitectónico, no un renombramiento de importaciones\u003C\u002Fa>\u003Col class=\"editorjs-toc__list editorjs-toc__list--depth-1\">\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-41\" class=\"editorjs-toc__link\">El inventario de migración\u003C\u002Fa>\u003C\u002Fli>\u003C\u002Fol>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-43\" class=\"editorjs-toc__link\">La beta pública cambia el modelo de riesgo\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-46\" class=\"editorjs-toc__link\">Una arquitectura predeterminada práctica\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-50\" class=\"editorjs-toc__link\">¿Qué cambiaría esta respuesta?\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-53\" class=\"editorjs-toc__link\">Limitaciones\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-56\" class=\"editorjs-toc__link\">Conclusión\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-59\" class=\"editorjs-toc__link\">Preguntas frecuentes\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-61\" class=\"editorjs-toc__link\">Glosario\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-63\" class=\"editorjs-toc__link\">Fuentes principales y lecturas complementarias\u003C\u002Fa>\u003C\u002Fli>\u003C\u002Fol>\u003C\u002Fnav>\n\u003Cp>El stack de agentes de OpenAI cambió sustancialmente en septiembre de 2026. La nueva Agents API introdujo un harness de Codex gestionado para agentes duraderos en la nube, mientras que el Agents SDK anterior pasó a una fase de mantenimiento con funcionalidad completa. La Responses API sigue siendo la superficie de nivel inferior para aplicaciones que requieren llamadas directas al modelo o gestionar el bucle del agente por sí mismas. No son tres wrappers intercambiables sobre lo mismo: sitúan el límite del runtime en lugares distintos.\u003C\u002Fp>\n\u003Caside class=\"editorjs-callout editorjs-callout--info my-6 rounded-xl border p-5 border-blue-300 bg-blue-50 dark:border-blue-900 dark:bg-blue-950\u002F20\" role=\"note\">\u003Cstrong class=\"block mb-2 text-gray-900 dark:text-gray-100\">Respuesta directa\u003C\u002Fstrong>\u003Cdiv class=\"text-gray-700 dark:text-gray-200\"><strong>Para una nueva aplicación de agentes nativa de OpenAI en septiembre de 2026, comience con la Agents API a menos que tenga una razón concreta para controlar una mayor parte del runtime.<\u002Fstrong> Utilice la Responses API cuando su aplicación deba gestionar el bucle, la estrategia de estado y la orquestación. Conserve el Agents SDK para aplicaciones existentes o brechas a corto plazo donde la Agents API aún no admita una capacidad requerida. Si desea el harness de Codex en sí pero necesita ejecutarlo en infraestructura operada por usted, evalúe el Codex SDK en lugar de forzar al Agents SDK a asumir esa función.\u003C\u002Fdiv>\u003C\u002Faside>\n\u003Caside class=\"editorjs-callout editorjs-callout--warning my-6 rounded-xl border p-5 border-amber-300 bg-amber-50 dark:border-amber-900 dark:bg-amber-950\u002F20\" role=\"note\">\u003Cstrong class=\"block mb-2 text-gray-900 dark:text-gray-100\">Decisión de arquitectura sensible al tiempo\u003C\u002Fstrong>\u003Cdiv class=\"text-gray-700 dark:text-gray-200\">Este artículo refleja el estado de la plataforma de OpenAI al <strong>25 de septiembre de 2026<\u002Fstrong>. La Agents API se lanzó en beta pública el 10 de septiembre de 2026. Las superficies en beta, la cobertura de funciones, los precios y las directrices de migración pueden cambiar; verifique la documentación actual antes de comprometer una arquitectura de producción.\u003C\u002Fdiv>\u003C\u002Faside>\n\u003Caside class=\"editorjs-callout editorjs-callout--note my-6 rounded-xl border p-5 border-gray-300 bg-gray-50 dark:border-gray-700 dark:bg-gray-900\u002F40\" role=\"note\">\u003Cstrong class=\"block mb-2 text-gray-900 dark:text-gray-100\">El modelo de decisión utilizado aquí\u003C\u002Fstrong>\u003Cdiv class=\"text-gray-700 dark:text-gray-200\">La prueba de control del runtime (Runtime Ownership Test) que figura a continuación es un modelo práctico de arquitectura, no una taxonomía de productos de OpenAI. Plantea qué partes del runtime del agente debe controlar su equipo frente a qué partes desea que opere OpenAI.\u003C\u002Fdiv>\u003C\u002Faside>\n\u003Ch2 id=\"section-6\">La arquitectura ha cambiado: elija un límite de runtime, no una biblioteca\u003C\u002Fh2>\n\u003Cp>La decisión importante ya no es simplemente: «¿Qué SDK debo instalar?». Se trata de quién controla el harness, el bucle del agente, el estado duradero de la sesión, la compactación del contexto, la recuperación, el entorno de ejecución y el ciclo de vida de la aplicación.\u003C\u002Fp>\n\u003Cp>La descripción general actual de Agents de OpenAI hace explícito este límite. La Agents API ejecuta un harness de Codex alojado y gestiona la orquestación además del estado de sesión duradero. La Responses API proporciona respuestas del modelo y capacidades alojadas mientras su aplicación gestiona el bucle del agente circundante. El Agents SDK ejecuta el bucle en su aplicación y ahora está en estado de funcionalidad completa (feature complete) en lugar de ser la vía a seguir para nuevas características importantes de agentes.\u003C\u002Fp>\n\u003Ch2 id=\"section-9\">Comparativa resumida\u003C\u002Fh2>\n\u003Cdiv class=\"overflow-x-auto\">\u003Ctable class=\"w-full border-collapse\">\u003Cthead>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">Opción\u003C\u002Fth>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">Mejor opción en 2026\u003C\u002Fth>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">¿Quién gestiona el bucle del agente?\u003C\u002Fth>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">Responsabilidad sobre la sesión \u002F contexto\u003C\u002Fth>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">Estado estratégico\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Agents API\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Nuevos agentes duraderos nativos de OpenAI\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Harness de Codex gestionado por OpenAI\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">OpenAI gestiona las sesiones, la orquestación, la compactación y la recuperación\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Punto de partida recomendado para nuevas aplicaciones de agentes; beta pública\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Responses API\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Integraciones directas de modelos y runtimes de agentes personalizados\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Su aplicación\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Usted elige el encadenamiento de respuestas, Conversations, el almacenamiento y la lógica del bucle\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Primitiva principal de la API; recomendada sobre Chat Completions para nuevos proyectos\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Agents SDK\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Aplicaciones existentes con el SDK o brechas temporales de funcionalidad\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Su aplicación a través del ejecutor del SDK\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Su aplicación opera el despliegue, el almacenamiento y el comportamiento en runtime\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Funcionalidad completa; se mantienen el soporte y la compatibilidad, no se prevén nuevas funciones principales\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Codex SDK\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Harness de Codex en infraestructura operada por usted\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Harness de Codex en su entorno\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Usted opera el alojamiento y el ciclo de vida del harness\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Opción independiente cuando se desea el harness sin el runtime alojado de la Agents API\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003C\u002Fdiv>\n\u003Ch2 id=\"section-11\">1. Agents API: harness gestionado, agente duradero en la nube\u003C\u002Fh2>\n\u003Cp>La Agents API expone el harness de Codex a través de un servicio gestionado por OpenAI. OpenAI gestiona las sesiones, la orquestación, la compactación del contexto y la recuperación. Su aplicación sigue proporcionando herramientas y eligiendo el entorno de ejecución.\u003C\u002Fp>\n\u003Cp>Esa última distinción es importante. «Agente gestionado» no significa necesariamente que «todo el cómputo se ejecute dentro de OpenAI». La arquitectura de la Agents API admite no tener entorno, un entorno alojado por OpenAI o un entorno autohospedado conectado al harness gestionado. Con un entorno autohospedado, su aplicación asume el aprovisionamiento, la reconexión, el apagado y los archivos persistentes, mientras que el harness permanece gestionado.\u003C\u002Fp>\n\u003Cp>Por lo tanto, la Agents API es un servicio de runtime, no un simple formato de solicitud. Las sesiones pueden persistir, transmitir el progreso en tiempo real, recibir tareas adicionales, usar herramientas, trabajar con archivos y recuperarse a lo largo de trabajos de ejecución prolongada.\u003C\u002Fp>\n\u003Ch3 id=\"section-15\">Qué se obtiene con la Agents API\u003C\u002Fh3>\n\u003Cul>\u003Cli>Un harness de Codex gestionado en lugar de construir y operar el bucle principal del agente por su cuenta.\u003C\u002Fli>\u003Cli>Sesiones duraderas para tareas que abarcan múltiples turnos y procesos de larga duración.\u003C\u002Fli>\u003Cli>Orquestación gestionada, compactación de contexto y recuperación.\u003C\u002Fli>\u003Cli>Entornos de ejecución alojados en OpenAI o autohospedados, según los requisitos de la carga de trabajo.\u003C\u002Fli>\u003Cli>Streaming y webhooks para el seguimiento del progreso y eventos del ciclo de vida.\u003C\u002Fli>\u003Cli>Una dirección de plataforma que OpenAI recomienda explícitamente para nuevas aplicaciones de agentes.\u003C\u002Fli>\u003C\u002Ful>\n\u003Ch3 id=\"section-17\">Qué sigue bajo su control\u003C\u002Fh3>\n\u003Cul>\u003Cli>Su producto y servidor de aplicaciones.\u003C\u002Fli>\u003Cli>Implementaciones de herramientas de funciones y lógica de negocio.\u003C\u002Fli>\u003Cli>Decisiones de autorización y políticas relativas a sus propios sistemas.\u003C\u002Fli>\u003Cli>El ciclo de vida del entorno de ejecución al optar por cómputo autohospedado.\u003C\u002Fli>\u003Cli>Evaluación, criterios de aceptación, barreras de seguridad (guardrails) específicas del dominio y la decisión sobre qué tiene permitido hacer el agente.\u003C\u002Fli>\u003C\u002Ful>\n\u003Ch2 id=\"section-19\">2. API Responses: controle el bucle, use las primitivas de la plataforma\u003C\u002Fh2>\n\u003Cp>La API Responses es la opción de menor nivel cuando desea las capacidades de modelos y herramientas de OpenAI sin delegar el entorno de ejecución general del agente. OpenAI describe Responses como la primitiva de API recomendada para nuevos proyectos y como una evolución de Chat Completions con herramientas integradas, opciones de estado multiturno, entrada multimodal y uso agéntico de herramientas.\u003C\u002Fp>\n\u003Cp>Una solicitud a Responses puede invocar herramientas por sí misma, pero su aplicación sigue siendo responsable del flujo de trabajo general cuando crea un agente a su alrededor. Eso significa que su código decide cómo persistir el estado de la aplicación, cuándo continuar, cómo recuperarse, cómo coordinar especialistas, cómo compactar historiales largos y cómo representar el trabajo reanudable.\u003C\u002Fp>\n\u003Cp>Esto no es intrínsecamente inferior. Es el límite adecuado cuando el comportamiento del agente debe estar profundamente integrado en la lógica de la aplicación existente, cuando necesita un modelo de estado personalizado o cuando un entorno de ejecución administrado ocultaría un control que realmente necesita.\u003C\u002Fp>\n\u003Ch2 id=\"section-23\">3. Agents SDK: con soporte, pero ya no es la vía predeterminada hacia el futuro\u003C\u002Fh2>\n\u003Cp>El Agents SDK sigue siendo un marco de código abierto para ejecutar flujos de trabajo de agentes en su aplicación. Proporciona definiciones de agentes, herramientas, transferencias (handoffs), barreras de seguridad (guardrails), trazado, sesiones y el bucle de ejecución en TypeScript y Python.\u003C\u002Fp>\n\u003Cp>Pero su estatus estratégico cambió. OpenAI ahora califica el Agents SDK como completo en características (feature complete): el mantenimiento, las correcciones de seguridad, las correcciones de errores críticos y el trabajo de compatibilidad continúan, pero no se planean nuevas características importantes. OpenAI recomienda la API Agents para nuevas aplicaciones.\u003C\u002Fp>\n\u003Cp>Eso no significa que una aplicación existente construida con el SDK deba reescribirse de inmediato. Significa que la arquitectura debe dejar de asumir que el SDK es donde llegarán las próximas capacidades importantes del entorno de ejecución de agentes.\u003C\u002Fp>\n\u003Caside class=\"editorjs-callout editorjs-callout--warning my-6 rounded-xl border p-5 border-amber-300 bg-amber-50 dark:border-amber-900 dark:bg-amber-950\u002F20\" role=\"note\">\u003Cstrong class=\"block mb-2 text-gray-900 dark:text-gray-100\">No interprete «feature complete» como «en desuso»\u003C\u002Fstrong>\u003Cdiv class=\"text-gray-700 dark:text-gray-200\">OpenAI no ha calificado el Agents SDK como sin soporte. Las aplicaciones existentes pueden seguir utilizándolo, y el mantenimiento, la seguridad, las correcciones de errores críticos y el trabajo de compatibilidad continúan. El cambio importante es que no se planean nuevas características principales y las nuevas aplicaciones se orientan hacia la API Agents.\u003C\u002Fdiv>\u003C\u002Faside>\n\u003Ch2 id=\"section-28\">Dónde encaja el Codex SDK\u003C\u002Fh2>\n\u003Cp>La elección actual no es una simple bifurcación de tres vías. La descripción general del entorno de ejecución de OpenAI incluye el Codex SDK como la opción para ejecutar el entorno de Codex en la infraestructura que usted opera. Eso es arquitectónicamente diferente tanto de la API Agents alojada como del Agents SDK.\u003C\u002Fp>\n\u003Cp>Si su requisito real es «Quiero el entorno de Codex, pero necesito operarlo yo mismo», el Codex SDK es la superficie a evaluar. Si su requisito es «Quiero controlar el bucle alrededor de las llamadas al modelo», evalúe Responses. Si su requisito es «Ya tengo una aplicación funcional con Agents SDK», el SDK existente puede seguir siendo válido mientras planifica en función de su estado de mantenimiento.\u003C\u002Fp>\n\u003Ch2 id=\"section-31\">La prueba de propiedad del entorno de ejecución\u003C\u002Fh2>\n\u003Cp>Una decisión de arquitectura útil comienza por determinar qué debe poseer su equipo. Califique cada requisito como: debe controlar, prefiere controlar o prefiere administrado.\u003C\u002Fp>\n\u003Csection class=\"editorjs-comparison my-6\">\u003Ch3 class=\"mb-3 text-lg font-semibold\">Prueba de propiedad del entorno de ejecución\u003C\u002Fh3>\u003Cdiv class=\"overflow-x-auto\">\u003Ctable class=\"w-full border-collapse\">\u003Cthead>\u003Ctr>\u003Cth class=\"border border-gray-300 bg-gray-50 px-4 py-3 text-left dark:border-gray-700 dark:bg-gray-900\">\u003C\u002Fth>\u003Cth class=\"border border-gray-300 bg-gray-50 px-4 py-3 text-left font-semibold dark:border-gray-700 dark:bg-gray-900\">Decisión\u003C\u002Fth>\u003Cth class=\"border border-gray-300 bg-gray-50 px-4 py-3 text-left font-semibold dark:border-gray-700 dark:bg-gray-900\">Si prefiere administrado\u003C\u002Fth>\u003Cth class=\"border border-gray-300 bg-gray-50 px-4 py-3 text-left font-semibold dark:border-gray-700 dark:bg-gray-900\">Si requiere control\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-3 text-left font-semibold dark:border-gray-700\">Bucle del agente\u003C\u002Fth>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-3 text-left font-semibold dark:border-gray-700\">Sesiones duraderas\u003C\u002Fth>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-3 text-left font-semibold dark:border-gray-700\">Entorno de ejecución (harness)\u003C\u002Fth>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-3 text-left font-semibold dark:border-gray-700\">Entorno de ejecución\u003C\u002Fth>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-3 text-left font-semibold dark:border-gray-700\">Semántica de orquestación\u003C\u002Fth>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-3 text-left font-semibold dark:border-gray-700\">Flexibilidad de proveedor \u002F transporte\u003C\u002Fth>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-3 text-left font-semibold dark:border-gray-700\">Carga operativa\u003C\u002Fth>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003C\u002Fdiv>\u003C\u002Fsection>\n\u003Ch2 id=\"section-34\">Un árbol de decisión para nuevos sistemas\u003C\u002Fh2>\n\u003Csection class=\"editorjs-process my-6\">\u003Ch3 class=\"mb-3 text-lg font-semibold\">Elija el entorno de ejecución según el límite de control\u003C\u002Fh3>\u003Cdiv class=\"grid grid-cols-1 md:grid-cols-2 xl:grid-cols-3 gap-4\">\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">1\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">1. ¿Es esta una nueva aplicación de agente?\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">Si la respuesta es no, no migre únicamente porque existe una superficie más nueva. Evalúe primero las limitaciones reales de la aplicación actual.\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">2\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">2. ¿Desea un entorno administrado de agentes de larga ejecución?\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">Si la respuesta es sí, comience con la API Agents como la vía recomendada por OpenAI para nuevas aplicaciones de agentes.\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">3\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">3. ¿Necesita el entorno de Codex pero debe operarlo usted mismo?\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">Evalúe el Codex SDK en lugar de reconstruir el comportamiento del entorno sobre el Agents SDK.\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">4\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">4. ¿Necesita controlar el bucle del agente y el modelo de estado?\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">Use la API Responses como la primitiva de plataforma de menor nivel y construya el bucle a su alrededor.\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">5\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">5. ¿Ya está utilizando el Agents SDK?\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">Continúe si cumple con sus requisitos; no se planean nuevas características importantes, por lo que debe considerar la futura adopción de la plataforma como una decisión explícita en su hoja de ruta.\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">6\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">6. ¿Falta alguna capacidad requerida en la API Agents?\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">OpenAI permite explícitamente el Agents SDK como una opción a corto plazo para nuevas aplicaciones que necesitan capacidades aún no admitidas.\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">7\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">7. Valide con una prueba de concepto (spike) representativa de producción\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">Pruebe herramientas, entorno, aprobaciones, latencia, observabilidad, recuperación ante fallos y ciclo de vida antes de fijar la arquitectura definitiva.\u003C\u002Fdiv>\u003C\u002Fdiv>\u003C\u002Fdiv>\u003C\u002Fsection>\n\u003Ch2 id=\"section-36\">Qué no debería motivar la decisión\u003C\u002Fh2>\n\u003Cdiv class=\"overflow-x-auto\">\u003Ctable class=\"w-full border-collapse\">\u003Cthead>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">Regla de decisión débil\u003C\u002Fth>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">Por qué falla\u003C\u002Fth>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">Mejor pregunta\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">“La API más nueva debe ser la mejor.”\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Lo más nuevo puede ser estratégicamente preferible y, aun así, carecer de una capacidad que necesitas.\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">¿Qué responsabilidades del entorno de ejecución deben ser gestionadas versus pertenecer a la aplicación?\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">“Ya conocemos el SDK.”\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">La familiaridad del equipo puede preservar una arquitectura cuya hoja de ruta ha cambiado.\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">¿Cuál es el coste de quedarse frente a migrar durante el próximo ciclo de producto?\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">“Gestionado significa sin infraestructura.”\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">La Agents API todavía puede usar entornos autohospedados y tu aplicación sigue siendo propietaria de la lógica de producto.\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">¿Qué capa de infraestructura se está delegando realmente?\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">“Responses es solo para llamadas simples.”\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Responses proporciona herramientas integradas y primitivas con estado; puede ser la base de bucles de agentes personalizados.\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">¿Necesitamos que la plataforma sea dueña del arnés, o solo de las primitivas de modelos\u002Fherramientas?\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">“Que esté completa a nivel de funciones significa que debemos migrar ahora.”\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">El SDK sigue manteniéndose para las aplicaciones existentes.\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">¿Qué requisito futuro concreto se ve bloqueado por quedarse?\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003C\u002Fdiv>\n\u003Ch2 id=\"section-38\">La migración es un cambio arquitectónico, no un renombramiento de importaciones\u003C\u002Fh2>\n\u003Cp>Pasar del Agents SDK a la Agents API cambia la titularidad. En el SDK, el bucle se ejecuta en tu aplicación. En la Agents API, OpenAI ejecuta el arnés y la sesión, mientras que tu aplicación se integra a través de tareas, eventos, herramientas y límites de entorno.\u003C\u002Fp>\n\u003Cp>Por lo tanto, un plan de migración real debe mapear el estado de la sesión, la orquestación personalizada, los traspasos (handoffs), la ejecución de herramientas, las aprobaciones, el almacenamiento, el rastreo, los reintentos, la recuperación ante fallos, el ciclo de vida del entorno y cualquier abstracción específica del proveedor. El volumen de código puede disminuir mientras los supuestos operativos cambian.\u003C\u002Fp>\n\u003Ch3 id=\"section-41\">El inventario de migración\u003C\u002Fh3>\n\u003Cul>\u003Cli>Definiciones de agentes y titularidad de las instrucciones.\u003C\u002Fli>\u003Cli>Definiciones de herramientas y dónde se ejecuta cada herramienta.\u003C\u002Fli>\u003Cli>Traspasos (handoffs), patrones de gestor\u002Fespecialista y comportamiento de subagentes.\u003C\u002Fli>\u003Cli>Identificadores de sesión, estado de la conversación, capacidad de reanudación y retención del historial.\u003C\u002Fli>\u003Cli>Aprobaciones humanas y semántica de interrupción.\u003C\u002Fli>\u003Cli>Lógica personalizada de recorte o compactación de contexto.\u003C\u002Fli>\u003Cli>Rastreo (tracing), evaluaciones, observabilidad y depuración en producción.\u003C\u002Fli>\u003Cli>Archivos autohospedados, contenedores, acceso a redes privadas u otras dependencias de ejecución.\u003C\u002Fli>\u003Cli>Abstracción de proveedores o dependencias de modelos ajenos a OpenAI.\u003C\u002Fli>\u003Cli>Supuestos de reintento, tiempo de espera, idempotencia, recuperación y ciclo de vida.\u003C\u002Fli>\u003C\u002Ful>\n\u003Ch2 id=\"section-43\">La beta pública cambia el modelo de riesgo\u003C\u002Fh2>\n\u003Cp>La Agents API es la dirección recomendada para nuevas aplicaciones de agentes, pero también está en beta pública. Esos hechos no son contradictorios. La dirección estratégica responde a “¿hacia dónde va la plataforma?”. El estado beta responde a “¿cuánto cambio operativo y de interfaz debo presupuestar?”.\u003C\u002Fp>\n\u003Cp>Para los sistemas en producción, aísla la integración detrás de un límite de la aplicación. Mantén el estado del dominio, los permisos, los datos de auditoría y las reglas de negocio fuera de los objetos de sesión específicos del proveedor siempre que sea posible. Esto facilita absorber la evolución de la API sin convertir el entorno de ejecución del agente en la fuente de verdad de todo tu producto.\u003C\u002Fp>\n\u003Ch2 id=\"section-46\">Una arquitectura predeterminada práctica\u003C\u002Fh2>\n\u003Cp>Para muchas aplicaciones nativas de OpenAI nuevas, un valor predeterminado razonable para 2026 es: Agents API para el arnés gestionado y la sesión duradera, servicios de dominio y autorización propios de la aplicación, herramientas de función explícitas para acciones comerciales, y ejecución alojada en OpenAI o autohospedada según los requisitos de datos y cómputo.\u003C\u002Fp>\n\u003Cp>Eso mantiene la potencia del entorno de ejecución del agente sin convertirlo en el propietario de la verdad del negocio. La aplicación sigue decidiendo qué puede hacer un usuario, qué datos son autoritativos, qué acciones requieren aprobación y cómo se validan los resultados.\u003C\u002Fp>\n\u003Caside class=\"editorjs-referral my-6\">\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fes\u002Fblog\u002Fai-agent-reliability-why-the-final-answer-is-not-enough\" class=\"flex flex-col sm:flex-row gap-4 rounded-xl border border-gray-200 dark:border-gray-700 p-4 transition hover:border-primary-500\">\u003Cdiv class=\"min-w-0 flex-1\">\u003Cstrong class=\"block text-lg text-gray-900 dark:text-gray-100\">Fiabilidad de los agentes de IA: por qué la respuesta final no es suficiente\u003C\u002Fstrong>\u003Cp class=\"mt-2 text-sm text-gray-600 dark:text-gray-300\">Un arnés gestionado reduce la carga operativa, pero la fiabilidad aún requiere una ejecución trazable, criterios de aceptación explícitos y evaluación más allá de la respuesta final.\u003C\u002Fp>\u003Cspan class=\"mt-3 inline-flex text-sm font-medium text-primary-600 dark:text-primary-400\">Leer el artículo sobre fiabilidad →\u003C\u002Fspan>\u003C\u002Fdiv>\u003C\u002Fa>\u003C\u002Faside>\n\u003Ch2 id=\"section-50\">¿Qué cambiaría esta respuesta?\u003C\u002Fh2>\n\u003Cp>La recomendación cambia si la Agents API añade o elimina capacidades, sale de la beta con contratos diferentes, cambia los límites de entorno o de precios, o introduce herramientas de migración que reduzcan las diferencias de titularidad. También cambia si tu aplicación depende de la portabilidad de proveedores, semánticas de orquestación personalizadas, ejecución exclusivamente local o una capacidad que el arnés alojado no pueda admitir.\u003C\u002Fp>\n\u003Cp>Para una aplicación existente de Agents SDK, la respuesta también varía en función del coste de migración. Si el sistema es estable, está bien evaluado y no se ve bloqueado por el estado de funciones completas del SDK, la migración inmediata puede generar más riesgos que valor. Si la hoja de ruta del producto depende de capacidades que solo llegan a la Agents API, retrasar la migración puede generar un tipo diferente de deuda.\u003C\u002Fp>\n\u003Ch2 id=\"section-53\">Limitaciones\u003C\u002Fh2>\n\u003Cp>Esta comparación se centra en la titularidad del entorno de ejecución y la dirección de la plataforma declarada por OpenAI. No evalúa la latencia, la calidad ni el coste total para una carga de trabajo específica. Esas propiedades dependen de la elección del modelo, el uso de herramientas, el entorno, la duración de la tarea, el almacenamiento en caché, el uso de sandboxes y la arquitectura de la aplicación.\u003C\u002Fp>\n\u003Cp>La API de Agents también es lo suficientemente nueva como para que la experiencia en producción aún se esté acumulando. Por lo tanto, un diseño debe validarse con cargas de trabajo representativas en lugar de elegirse únicamente a partir del posicionamiento del producto.\u003C\u002Fp>\n\u003Ch2 id=\"section-56\">Conclusión\u003C\u002Fh2>\n\u003Cp>La decisión sobre agentes de OpenAI en 2026 trata fundamentalmente sobre la propiedad del entorno de ejecución (runtime). La API de Agents significa que OpenAI asume una mayor parte del arnés y de la maquinaria de sesiones duraderas. Responses implica que su aplicación es propietaria del bucle en torno a las primitivas de la plataforma. El Agents SDK sigue siendo válido para los sistemas existentes, pero ya no es el destino predeterminado para las nuevas características importantes del runtime de agentes.\u003C\u002Fp>\n\u003Cp>Para una aplicación nueva, siga la dirección de la plataforma a menos que un requisito real lo empuje hacia una capa inferior de la pila. Comience con la API de Agents, pase a Responses cuando necesite controlar el bucle, evalúe Codex SDK cuando necesite el arnés en su propia infraestructura y conserve el Agents SDK donde las inversiones existentes o las brechas temporales de capacidades lo justifiquen.\u003C\u002Fp>\n\u003Ch2 id=\"section-59\">Preguntas frecuentes\u003C\u002Fh2>\n\u003Csection class=\"editorjs-faq my-6 rounded-xl border border-gray-200 p-5 dark:border-gray-700\">\u003Ch3 class=\"mb-3 text-lg font-semibold\">Opciones de runtime para agentes de OpenAI en 2026\u003C\u002Fh3>\u003Cdiv id=\"faq1\" class=\"border-t border-gray-200 py-4 first:border-t-0 dark:border-gray-700\">\u003Ch4 class=\"font-semibold text-gray-900 dark:text-gray-100\">¿Debería usar la API de Agents de OpenAI o el Agents SDK para un nuevo proyecto?\u003C\u002Fh4>\u003Cdiv class=\"mt-2 text-gray-600 dark:text-gray-300\">Actualmente, OpenAI recomienda la API de Agents para nuevas aplicaciones de agentes. El Agents SDK cuenta con todas las funciones completas y sigue teniendo soporte para aplicaciones existentes, continuando con las labores de mantenimiento, seguridad, corrección de errores críticos y compatibilidad.\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv id=\"faq2\" class=\"border-t border-gray-200 py-4 first:border-t-0 dark:border-gray-700\">\u003Ch4 class=\"font-semibold text-gray-900 dark:text-gray-100\">¿Está en desuso el Agents SDK?\u003C\u002Fh4>\u003Cdiv class=\"mt-2 text-gray-600 dark:text-gray-300\">OpenAI lo describe como completo a nivel de funciones (feature complete), no como sin soporte. No se prevén nuevas características importantes, pero continúan el mantenimiento, los parches de seguridad, la corrección de errores críticos y el trabajo de compatibilidad.\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv id=\"faq3\" class=\"border-t border-gray-200 py-4 first:border-t-0 dark:border-gray-700\">\u003Ch4 class=\"font-semibold text-gray-900 dark:text-gray-100\">¿Cuándo debería usar la API de Responses en lugar de la API de Agents?\u003C\u002Fh4>\u003Cdiv class=\"mt-2 text-gray-600 dark:text-gray-300\">Utilice Responses cuando su aplicación deba controlar el bucle de agentes, la estrategia de estado, la orquestación y la lógica de continuación mientras sigue utilizando los modelos y las herramientas de la plataforma de OpenAI.\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv id=\"faq4\" class=\"border-t border-gray-200 py-4 first:border-t-0 dark:border-gray-700\">\u003Ch4 class=\"font-semibold text-gray-900 dark:text-gray-100\">¿Requiere la API de Agents capacidad de cómputo alojada en OpenAI?\u003C\u002Fh4>\u003Cdiv class=\"mt-2 text-gray-600 dark:text-gray-300\">No. La API de Agents puede no utilizar ningún entorno de ejecución, un entorno alojado en OpenAI o un entorno autoalojado conectado al arnés gestionado.\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv id=\"faq5\" class=\"border-t border-gray-200 py-4 first:border-t-0 dark:border-gray-700\">\u003Ch4 class=\"font-semibold text-gray-900 dark:text-gray-100\">¿Dónde encaja el Codex SDK?\u003C\u002Fh4>\u003Cdiv class=\"mt-2 text-gray-600 dark:text-gray-300\">OpenAI posiciona el Codex SDK para ejecutar el arnés de Codex en una infraestructura operada por usted. Es la opción relevante cuando desea el arnés pero no el runtime alojado de la API de Agents.\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv id=\"faq6\" class=\"border-t border-gray-200 py-4 first:border-t-0 dark:border-gray-700\">\u003Ch4 class=\"font-semibold text-gray-900 dark:text-gray-100\">¿Debería migrar inmediatamente una aplicación existente basada en Agents SDK?\u003C\u002Fh4>\u003Cdiv class=\"mt-2 text-gray-600 dark:text-gray-300\">No de forma automática. Evalúe si el sistema actual se ve limitado por el estado de funciones completas del SDK, si las capacidades requeridas existen en la API de Agents y si el beneficio de la migración supera el cambio operativo y arquitectónico.\u003C\u002Fdiv>\u003C\u002Fdiv>\u003C\u002Fsection>\n\u003Ch2 id=\"section-61\">Glosario\u003C\u002Fh2>\n\u003Csection class=\"editorjs-glossary my-6 rounded-xl border border-gray-200 dark:border-gray-700 p-5\">\u003Ch3 class=\"mb-3 text-lg font-semibold\">Términos clave del runtime\u003C\u002Fh3>\u003Cdl>\u003Cdiv id=\"harness\" class=\"border-t border-gray-200 dark:border-gray-700 py-3 first:border-t-0\">\u003Cdt class=\"font-semibold text-gray-900 dark:text-gray-100\">Arnés (Harness)\u003C\u002Fdt>\u003Cdd class=\"mt-1 text-gray-600 dark:text-gray-300\">El bucle de ejecución y la maquinaria de soporte que coordina las llamadas al modelo, herramientas, contexto, sesiones y la ejecución del agente.\u003C\u002Fdd>\u003C\u002Fdiv>\u003Cdiv id=\"agents-api\" class=\"border-t border-gray-200 dark:border-gray-700 py-3 first:border-t-0\">\u003Cdt class=\"font-semibold text-gray-900 dark:text-gray-100\">API de Agents\u003C\u002Fdt>\u003Cdd class=\"mt-1 text-gray-600 dark:text-gray-300\">La API gestionada de OpenAI para agentes en la nube duraderos que utiliza un arnés de Codex alojado.\u003C\u002Fdd>\u003C\u002Fdiv>\u003Cdiv id=\"responses-api\" class=\"border-t border-gray-200 dark:border-gray-700 py-3 first:border-t-0\">\u003Cdt class=\"font-semibold text-gray-900 dark:text-gray-100\">API de Responses\u003C\u002Fdt>\u003Cdd class=\"mt-1 text-gray-600 dark:text-gray-300\">La primitiva de API de bajo nivel de OpenAI para respuestas del modelo, herramientas alojadas e interacciones con estado alrededor de las cuales las aplicaciones pueden construir su propio bucle de agentes.\u003C\u002Fdd>\u003C\u002Fdiv>\u003Cdiv id=\"agents-sdk\" class=\"border-t border-gray-200 dark:border-gray-700 py-3 first:border-t-0\">\u003Cdt class=\"font-semibold text-gray-900 dark:text-gray-100\">Agents SDK\u003C\u002Fdt>\u003Cdd class=\"mt-1 text-gray-600 dark:text-gray-300\">El framework de código abierto de OpenAI para ejecutar flujos de trabajo de agentes en el código de la aplicación; con características completas a partir de septiembre de 2026.\u003C\u002Fdd>\u003C\u002Fdiv>\u003Cdiv id=\"codex-sdk\" class=\"border-t border-gray-200 dark:border-gray-700 py-3 first:border-t-0\">\u003Cdt class=\"font-semibold text-gray-900 dark:text-gray-100\">Codex SDK\u003C\u002Fdt>\u003Cdd class=\"mt-1 text-gray-600 dark:text-gray-300\">La opción de runtime que OpenAI enumera para ejecutar el arnés de Codex en infraestructura operada por usted.\u003C\u002Fdd>\u003C\u002Fdiv>\u003Cdiv id=\"runtime-ownership\" class=\"border-t border-gray-200 dark:border-gray-700 py-3 first:border-t-0\">\u003Cdt class=\"font-semibold text-gray-900 dark:text-gray-100\">Propiedad del runtime\u003C\u002Fdt>\u003Cdd class=\"mt-1 text-gray-600 dark:text-gray-300\">El límite arquitectónico que describe qué partes del bucle del agente, el estado de la sesión, el entorno de ejecución y el ciclo de vida son operados por la plataforma frente a la aplicación.\u003C\u002Fdd>\u003C\u002Fdiv>\u003C\u002Fdl>\u003C\u002Fsection>\n\u003Ch2 id=\"section-63\">Fuentes principales y lecturas complementarias\u003C\u002Fh2>\n\u003Ca href=\"https:\u002F\u002Fopenai.com\u002Findex\u002Fintroducing-the-agents-api\u002F\" target=\"_blank\" rel=\"noopener noreferrer\" class=\"editorjs-link-tool block border border-gray-200 dark:border-gray-700 rounded-lg p-4 transition text-gray-900 dark:text-gray-100 hover:border-primary-500 hover:bg-primary-50 dark:hover:bg-gray-900 hover:text-gray-900 dark:hover:text-gray-100\">\u003Cstrong class=\"block font-semibold\">OpenAI — Presentación de la API de Agents\u003C\u002Fstrong>\u003Cp class=\"text-sm text-gray-600 dark:text-gray-400\">Anuncio del lanzamiento de la API de Agents el 10 de septiembre de 2026, que describe el arnés gestionado de Codex y la versión beta pública.\u003C\u002Fp>\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fagents\" target=\"_blank\" rel=\"noopener noreferrer\" class=\"editorjs-link-tool block border border-gray-200 dark:border-gray-700 rounded-lg p-4 transition text-gray-900 dark:text-gray-100 hover:border-primary-500 hover:bg-primary-50 dark:hover:bg-gray-900 hover:text-gray-900 dark:hover:text-gray-100\">\u003Cstrong class=\"block font-semibold\">OpenAI — Descripción general del runtime de agentes\u003C\u002Fstrong>\u003Cp class=\"text-sm text-gray-600 dark:text-gray-400\">Comparación actual de la API de Agents, Codex SDK y la API de Responses, incluyendo el estado de soporte del Agents SDK.\u003C\u002Fp>\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fagents-api\u002Foverview\" target=\"_blank\" rel=\"noopener noreferrer\" class=\"editorjs-link-tool block border border-gray-200 dark:border-gray-700 rounded-lg p-4 transition text-gray-900 dark:text-gray-100 hover:border-primary-500 hover:bg-primary-50 dark:hover:bg-gray-900 hover:text-gray-900 dark:hover:text-gray-100\">\u003Cstrong class=\"block font-semibold\">OpenAI — Descripción general de la API de Agents\u003C\u002Fstrong>\u003Cp class=\"text-sm text-gray-600 dark:text-gray-400\">Documentación sobre agentes duraderos en la nube, sesiones, orquestación, compactación de contexto, recuperación y opciones de entorno.\u003C\u002Fp>\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fagents-api\u002Farchitecture\" target=\"_blank\" rel=\"noopener noreferrer\" class=\"editorjs-link-tool block border border-gray-200 dark:border-gray-700 rounded-lg p-4 transition text-gray-900 dark:text-gray-100 hover:border-primary-500 hover:bg-primary-50 dark:hover:bg-gray-900 hover:text-gray-900 dark:hover:text-gray-100\">\u003Cstrong class=\"block font-semibold\">OpenAI — Arquitectura de la API de Agents\u003C\u002Fstrong>\u003Cp class=\"text-sm text-gray-600 dark:text-gray-400\">Límite arquitectónico entre el arnés alojado, el servidor de aplicaciones y los entornos de ejecución sin entorno\u002Falojados en OpenAI\u002Fautoalojados.\u003C\u002Fp>\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fagents\u002Fsdk\" target=\"_blank\" rel=\"noopener noreferrer\" class=\"editorjs-link-tool block border border-gray-200 dark:border-gray-700 rounded-lg p-4 transition text-gray-900 dark:text-gray-100 hover:border-primary-500 hover:bg-primary-50 dark:hover:bg-gray-900 hover:text-gray-900 dark:hover:text-gray-100\">\u003Cstrong class=\"block font-semibold\">OpenAI — Agents SDK\u003C\u002Fstrong>\u003Cp class=\"text-sm text-gray-600 dark:text-gray-400\">Aviso actual de soporte para el Agents SDK y explicación del bucle de agente controlado por la aplicación.\u003C\u002Fp>\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fmigrate-to-responses\" target=\"_blank\" rel=\"noopener noreferrer\" class=\"editorjs-link-tool block border border-gray-200 dark:border-gray-700 rounded-lg p-4 transition text-gray-900 dark:text-gray-100 hover:border-primary-500 hover:bg-primary-50 dark:hover:bg-gray-900 hover:text-gray-900 dark:hover:text-gray-100\">\u003Cstrong class=\"block font-semibold\">OpenAI — Migración a la API de Responses\u003C\u002Fstrong>\u003Cp class=\"text-sm text-gray-600 dark:text-gray-400\">Posicionamiento actual de la API de Responses, herramientas integradas, contexto con estado y primitivas de agentes.\u003C\u002Fp>\u003C\u002Fa>\n\u003Cnav class=\"editorjs-toc\" data-editorjs-toc=\"true\" aria-label=\"Contenidos\">\u003Cstrong class=\"editorjs-toc__title\">Contenidos\u003C\u002Fstrong>\u003Col class=\"editorjs-toc__list editorjs-toc__list--depth-0\">\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-6\" class=\"editorjs-toc__link\">La arquitectura ha cambiado: elija un límite de runtime, no una biblioteca\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-9\" class=\"editorjs-toc__link\">Comparativa resumida\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-11\" class=\"editorjs-toc__link\">1. Agents API: harness gestionado, agente duradero en la nube\u003C\u002Fa>\u003Col class=\"editorjs-toc__list editorjs-toc__list--depth-1\">\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-15\" class=\"editorjs-toc__link\">Qué se obtiene con la Agents API\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-17\" class=\"editorjs-toc__link\">Qué sigue bajo su control\u003C\u002Fa>\u003C\u002Fli>\u003C\u002Fol>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-19\" class=\"editorjs-toc__link\">2. API Responses: controle el bucle, use las primitivas de la plataforma\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-23\" class=\"editorjs-toc__link\">3. Agents SDK: con soporte, pero ya no es la vía predeterminada hacia el futuro\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-28\" class=\"editorjs-toc__link\">Dónde encaja el Codex SDK\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-31\" class=\"editorjs-toc__link\">La prueba de propiedad del entorno de ejecución\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-34\" class=\"editorjs-toc__link\">Un árbol de decisión para nuevos sistemas\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-36\" class=\"editorjs-toc__link\">Qué no debería motivar la decisión\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-38\" class=\"editorjs-toc__link\">La migración es un cambio arquitectónico, no un renombramiento de importaciones\u003C\u002Fa>\u003Col class=\"editorjs-toc__list editorjs-toc__list--depth-1\">\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-41\" class=\"editorjs-toc__link\">El inventario de migración\u003C\u002Fa>\u003C\u002Fli>\u003C\u002Fol>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-43\" class=\"editorjs-toc__link\">La beta pública cambia el modelo de riesgo\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-46\" class=\"editorjs-toc__link\">Una arquitectura predeterminada práctica\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-50\" class=\"editorjs-toc__link\">¿Qué cambiaría esta respuesta?\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-53\" class=\"editorjs-toc__link\">Limitaciones\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-56\" class=\"editorjs-toc__link\">Conclusión\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-59\" class=\"editorjs-toc__link\">Preguntas frecuentes\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-61\" class=\"editorjs-toc__link\">Glosario\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-63\" class=\"editorjs-toc__link\">Fuentes principales y lecturas complementarias\u003C\u002Fa>\u003C\u002Fli>\u003C\u002Fol>\u003C\u002Fnav>",{"time":212,"blocks":213,"version":798},1790366117379,[214,222,228,236,243,250,255,260,265,270,306,311,316,321,326,331,345,350,361,366,371,376,381,386,391,396,401,407,412,417,422,427,432,478,483,512,517,546,551,556,561,566,582,587,592,597,602,607,612,621,626,631,636,641,646,651,656,661,666,671,701,706,733,738,748,757,766,775,784,793],{"id":215,"data":216,"type":220,"tunes":221},"75qcDhG72q",{"title":217,"maxLevel":218,"minLevel":219},"Índice",3,2,"tableOfContents",{},{"id":223,"data":224,"type":226,"tunes":227},"intro",{"text":225},"El stack de agentes de OpenAI cambió sustancialmente en septiembre de 2026. La nueva Agents API introdujo un harness de Codex gestionado para agentes duraderos en la nube, mientras que el Agents SDK anterior pasó a una fase de mantenimiento con funcionalidad completa. La Responses API sigue siendo la superficie de nivel inferior para aplicaciones que requieren llamadas directas al modelo o gestionar el bucle del agente por sí mismas. No son tres wrappers intercambiables sobre lo mismo: sitúan el límite del runtime en lugares distintos.","paragraph",{},{"id":229,"data":230,"type":234,"tunes":235},"direct",{"body":231,"title":232,"variant":233},"\u003Cstrong>Para una nueva aplicación de agentes nativa de OpenAI en septiembre de 2026, comience con la Agents API a menos que tenga una razón concreta para controlar una mayor parte del runtime.\u003C\u002Fstrong> Utilice la Responses API cuando su aplicación deba gestionar el bucle, la estrategia de estado y la orquestación. Conserve el Agents SDK para aplicaciones existentes o brechas a corto plazo donde la Agents API aún no admita una capacidad requerida. Si desea el harness de Codex en sí pero necesita ejecutarlo en infraestructura operada por usted, evalúe el Codex SDK en lugar de forzar al Agents SDK a asumir esa función.","Respuesta directa","info","callout",{},{"id":237,"data":238,"type":234,"tunes":242},"date-warning",{"body":239,"title":240,"variant":241},"Este artículo refleja el estado de la plataforma de OpenAI al \u003Cstrong>25 de septiembre de 2026\u003C\u002Fstrong>. La Agents API se lanzó en beta pública el 10 de septiembre de 2026. Las superficies en beta, la cobertura de funciones, los precios y las directrices de migración pueden cambiar; verifique la documentación actual antes de comprometer una arquitectura de producción.","Decisión de arquitectura sensible al tiempo","warning",{},{"id":244,"data":245,"type":234,"tunes":249},"model-note",{"body":246,"title":247,"variant":248},"La prueba de control del runtime (Runtime Ownership Test) que figura a continuación es un modelo práctico de arquitectura, no una taxonomía de productos de OpenAI. Plantea qué partes del runtime del agente debe controlar su equipo frente a qué partes desea que opere OpenAI.","El modelo de decisión utilizado aquí","note",{},{"id":251,"data":252,"type":42,"tunes":254},"h-boundary",{"text":253,"level":219},"La arquitectura ha cambiado: elija un límite de runtime, no una biblioteca",{},{"id":256,"data":257,"type":226,"tunes":259},"p-boundary-1",{"text":258},"La decisión importante ya no es simplemente: «¿Qué SDK debo instalar?». Se trata de quién controla el harness, el bucle del agente, el estado duradero de la sesión, la compactación del contexto, la recuperación, el entorno de ejecución y el ciclo de vida de la aplicación.",{},{"id":261,"data":262,"type":226,"tunes":264},"p-boundary-2",{"text":263},"La descripción general actual de Agents de OpenAI hace explícito este límite. La Agents API ejecuta un harness de Codex alojado y gestiona la orquestación además del estado de sesión duradero. La Responses API proporciona respuestas del modelo y capacidades alojadas mientras su aplicación gestiona el bucle del agente circundante. El Agents SDK ejecuta el bucle en su aplicación y ahora está en estado de funcionalidad completa (feature complete) en lugar de ser la vía a seguir para nuevas características importantes de agentes.",{},{"id":266,"data":267,"type":42,"tunes":269},"h-short",{"text":268,"level":219},"Comparativa resumida",{},{"id":271,"data":272,"type":304,"tunes":305},"short-table",{"content":273,"stretched":43,"withHeadings":14},[274,280,286,292,298],[275,276,277,278,279],"Opción","Mejor opción en 2026","¿Quién gestiona el bucle del agente?","Responsabilidad sobre la sesión \u002F contexto","Estado estratégico",[281,282,283,284,285],"Agents API","Nuevos agentes duraderos nativos de OpenAI","Harness de Codex gestionado por OpenAI","OpenAI gestiona las sesiones, la orquestación, la compactación y la recuperación","Punto de partida recomendado para nuevas aplicaciones de agentes; beta pública",[287,288,289,290,291],"Responses API","Integraciones directas de modelos y runtimes de agentes personalizados","Su aplicación","Usted elige el encadenamiento de respuestas, Conversations, el almacenamiento y la lógica del bucle","Primitiva principal de la API; recomendada sobre Chat Completions para nuevos proyectos",[293,294,295,296,297],"Agents SDK","Aplicaciones existentes con el SDK o brechas temporales de funcionalidad","Su aplicación a través del ejecutor del SDK","Su aplicación opera el despliegue, el almacenamiento y el comportamiento en runtime","Funcionalidad completa; se mantienen el soporte y la compatibilidad, no se prevén nuevas funciones principales",[299,300,301,302,303],"Codex SDK","Harness de Codex en infraestructura operada por usted","Harness de Codex en su entorno","Usted opera el alojamiento y el ciclo de vida del harness","Opción independiente cuando se desea el harness sin el runtime alojado de la Agents API","table",{},{"id":307,"data":308,"type":42,"tunes":310},"h-agents-api",{"text":309,"level":219},"1. Agents API: harness gestionado, agente duradero en la nube",{},{"id":312,"data":313,"type":226,"tunes":315},"p-api-1",{"text":314},"La Agents API expone el harness de Codex a través de un servicio gestionado por OpenAI. OpenAI gestiona las sesiones, la orquestación, la compactación del contexto y la recuperación. Su aplicación sigue proporcionando herramientas y eligiendo el entorno de ejecución.",{},{"id":317,"data":318,"type":226,"tunes":320},"p-api-2",{"text":319},"Esa última distinción es importante. «Agente gestionado» no significa necesariamente que «todo el cómputo se ejecute dentro de OpenAI». La arquitectura de la Agents API admite no tener entorno, un entorno alojado por OpenAI o un entorno autohospedado conectado al harness gestionado. Con un entorno autohospedado, su aplicación asume el aprovisionamiento, la reconexión, el apagado y los archivos persistentes, mientras que el harness permanece gestionado.",{},{"id":322,"data":323,"type":226,"tunes":325},"p-api-3",{"text":324},"Por lo tanto, la Agents API es un servicio de runtime, no un simple formato de solicitud. Las sesiones pueden persistir, transmitir el progreso en tiempo real, recibir tareas adicionales, usar herramientas, trabajar con archivos y recuperarse a lo largo de trabajos de ejecución prolongada.",{},{"id":327,"data":328,"type":42,"tunes":330},"h-api-gain",{"text":329,"level":218},"Qué se obtiene con la Agents API",{},{"id":332,"data":333,"type":343,"tunes":344},"api-gains",{"meta":334,"items":335,"style":342},{},[336,337,338,339,340,341],"Un harness de Codex gestionado en lugar de construir y operar el bucle principal del agente por su cuenta.","Sesiones duraderas para tareas que abarcan múltiples turnos y procesos de larga duración.","Orquestación gestionada, compactación de contexto y recuperación.","Entornos de ejecución alojados en OpenAI o autohospedados, según los requisitos de la carga de trabajo.","Streaming y webhooks para el seguimiento del progreso y eventos del ciclo de vida.","Una dirección de plataforma que OpenAI recomienda explícitamente para nuevas aplicaciones de agentes.","unordered","list",{},{"id":346,"data":347,"type":42,"tunes":349},"h-api-own",{"text":348,"level":218},"Qué sigue bajo su control",{},{"id":351,"data":352,"type":343,"tunes":360},"api-own",{"meta":353,"items":354,"style":342},{},[355,356,357,358,359],"Su producto y servidor de aplicaciones.","Implementaciones de herramientas de funciones y lógica de negocio.","Decisiones de autorización y políticas relativas a sus propios sistemas.","El ciclo de vida del entorno de ejecución al optar por cómputo autohospedado.","Evaluación, criterios de aceptación, barreras de seguridad (guardrails) específicas del dominio y la decisión sobre qué tiene permitido hacer el agente.",{},{"id":362,"data":363,"type":42,"tunes":365},"h-responses",{"text":364,"level":219},"2. API Responses: controle el bucle, use las primitivas de la plataforma",{},{"id":367,"data":368,"type":226,"tunes":370},"p-resp-1",{"text":369},"La API Responses es la opción de menor nivel cuando desea las capacidades de modelos y herramientas de OpenAI sin delegar el entorno de ejecución general del agente. OpenAI describe Responses como la primitiva de API recomendada para nuevos proyectos y como una evolución de Chat Completions con herramientas integradas, opciones de estado multiturno, entrada multimodal y uso agéntico de herramientas.",{},{"id":372,"data":373,"type":226,"tunes":375},"p-resp-2",{"text":374},"Una solicitud a Responses puede invocar herramientas por sí misma, pero su aplicación sigue siendo responsable del flujo de trabajo general cuando crea un agente a su alrededor. Eso significa que su código decide cómo persistir el estado de la aplicación, cuándo continuar, cómo recuperarse, cómo coordinar especialistas, cómo compactar historiales largos y cómo representar el trabajo reanudable.",{},{"id":377,"data":378,"type":226,"tunes":380},"p-resp-3",{"text":379},"Esto no es intrínsecamente inferior. Es el límite adecuado cuando el comportamiento del agente debe estar profundamente integrado en la lógica de la aplicación existente, cuando necesita un modelo de estado personalizado o cuando un entorno de ejecución administrado ocultaría un control que realmente necesita.",{},{"id":382,"data":383,"type":42,"tunes":385},"h-sdk",{"text":384,"level":219},"3. Agents SDK: con soporte, pero ya no es la vía predeterminada hacia el futuro",{},{"id":387,"data":388,"type":226,"tunes":390},"p-sdk-1",{"text":389},"El Agents SDK sigue siendo un marco de código abierto para ejecutar flujos de trabajo de agentes en su aplicación. Proporciona definiciones de agentes, herramientas, transferencias (handoffs), barreras de seguridad (guardrails), trazado, sesiones y el bucle de ejecución en TypeScript y Python.",{},{"id":392,"data":393,"type":226,"tunes":395},"p-sdk-2",{"text":394},"Pero su estatus estratégico cambió. OpenAI ahora califica el Agents SDK como completo en características (feature complete): el mantenimiento, las correcciones de seguridad, las correcciones de errores críticos y el trabajo de compatibilidad continúan, pero no se planean nuevas características importantes. OpenAI recomienda la API Agents para nuevas aplicaciones.",{},{"id":397,"data":398,"type":226,"tunes":400},"p-sdk-3",{"text":399},"Eso no significa que una aplicación existente construida con el SDK deba reescribirse de inmediato. Significa que la arquitectura debe dejar de asumir que el SDK es donde llegarán las próximas capacidades importantes del entorno de ejecución de agentes.",{},{"id":402,"data":403,"type":234,"tunes":406},"sdk-warning",{"body":404,"title":405,"variant":241},"OpenAI no ha calificado el Agents SDK como sin soporte. Las aplicaciones existentes pueden seguir utilizándolo, y el mantenimiento, la seguridad, las correcciones de errores críticos y el trabajo de compatibilidad continúan. El cambio importante es que no se planean nuevas características principales y las nuevas aplicaciones se orientan hacia la API Agents.","No interprete «feature complete» como «en desuso»",{},{"id":408,"data":409,"type":42,"tunes":411},"h-codex-sdk",{"text":410,"level":219},"Dónde encaja el Codex SDK",{},{"id":413,"data":414,"type":226,"tunes":416},"p-codex-1",{"text":415},"La elección actual no es una simple bifurcación de tres vías. La descripción general del entorno de ejecución de OpenAI incluye el Codex SDK como la opción para ejecutar el entorno de Codex en la infraestructura que usted opera. Eso es arquitectónicamente diferente tanto de la API Agents alojada como del Agents SDK.",{},{"id":418,"data":419,"type":226,"tunes":421},"p-codex-2",{"text":420},"Si su requisito real es «Quiero el entorno de Codex, pero necesito operarlo yo mismo», el Codex SDK es la superficie a evaluar. Si su requisito es «Quiero controlar el bucle alrededor de las llamadas al modelo», evalúe Responses. Si su requisito es «Ya tengo una aplicación funcional con Agents SDK», el SDK existente puede seguir siendo válido mientras planifica en función de su estado de mantenimiento.",{},{"id":423,"data":424,"type":42,"tunes":426},"h-test",{"text":425,"level":219},"La prueba de propiedad del entorno de ejecución",{},{"id":428,"data":429,"type":226,"tunes":431},"p-test-intro",{"text":430},"Una decisión de arquitectura útil comienza por determinar qué debe poseer su equipo. Califique cada requisito como: debe controlar, prefiere controlar o prefiere administrado.",{},{"id":433,"data":434,"type":476,"tunes":477},"ownership-test",{"rows":435,"title":465,"layout":304,"columns":466},[436,441,445,449,453,457,461],{"id":437,"label":438,"values":439},"loop","Bucle del agente",[440,440,440],"",{"id":442,"label":443,"values":444},"session","Sesiones duraderas",[440,440,440],{"id":446,"label":447,"values":448},"harness","Entorno de ejecución (harness)",[440,440,440],{"id":450,"label":451,"values":452},"environment","Entorno de ejecución",[440,440,440],{"id":454,"label":455,"values":456},"orchestration","Semántica de orquestación",[440,440,440],{"id":458,"label":459,"values":460},"portability","Flexibilidad de proveedor \u002F transporte",[440,440,440],{"id":462,"label":463,"values":464},"ops","Carga operativa",[440,440,440],"Prueba de propiedad del entorno de ejecución",[467,470,473],{"id":468,"label":469},"decision","Decisión",{"id":471,"label":472},"managed","Si prefiere administrado",{"id":474,"label":475},"control","Si requiere control","comparison",{},{"id":479,"data":480,"type":42,"tunes":482},"h-tree",{"text":481,"level":219},"Un árbol de decisión para nuevos sistemas",{},{"id":484,"data":485,"type":510,"tunes":511},"decision-flow",{"steps":486,"title":508,"orientation":509},[487,490,493,496,499,502,505],{"label":488,"description":489},"1. ¿Es esta una nueva aplicación de agente?","Si la respuesta es no, no migre únicamente porque existe una superficie más nueva. Evalúe primero las limitaciones reales de la aplicación actual.",{"label":491,"description":492},"2. ¿Desea un entorno administrado de agentes de larga ejecución?","Si la respuesta es sí, comience con la API Agents como la vía recomendada por OpenAI para nuevas aplicaciones de agentes.",{"label":494,"description":495},"3. ¿Necesita el entorno de Codex pero debe operarlo usted mismo?","Evalúe el Codex SDK en lugar de reconstruir el comportamiento del entorno sobre el Agents SDK.",{"label":497,"description":498},"4. ¿Necesita controlar el bucle del agente y el modelo de estado?","Use la API Responses como la primitiva de plataforma de menor nivel y construya el bucle a su alrededor.",{"label":500,"description":501},"5. ¿Ya está utilizando el Agents SDK?","Continúe si cumple con sus requisitos; no se planean nuevas características importantes, por lo que debe considerar la futura adopción de la plataforma como una decisión explícita en su hoja de ruta.",{"label":503,"description":504},"6. ¿Falta alguna capacidad requerida en la API Agents?","OpenAI permite explícitamente el Agents SDK como una opción a corto plazo para nuevas aplicaciones que necesitan capacidades aún no admitidas.",{"label":506,"description":507},"7. Valide con una prueba de concepto (spike) representativa de producción","Pruebe herramientas, entorno, aprobaciones, latencia, observabilidad, recuperación ante fallos y ciclo de vida antes de fijar la arquitectura definitiva.","Elija el entorno de ejecución según el límite de control","auto","processFlow",{},{"id":513,"data":514,"type":42,"tunes":516},"h-not-drive",{"text":515,"level":219},"Qué no debería motivar la decisión",{},{"id":518,"data":519,"type":304,"tunes":545},"weak-rules",{"content":520,"stretched":43,"withHeadings":14},[521,525,529,533,537,541],[522,523,524],"Regla de decisión débil","Por qué falla","Mejor pregunta",[526,527,528],"“La API más nueva debe ser la mejor.”","Lo más nuevo puede ser estratégicamente preferible y, aun así, carecer de una capacidad que necesitas.","¿Qué responsabilidades del entorno de ejecución deben ser gestionadas versus pertenecer a la aplicación?",[530,531,532],"“Ya conocemos el SDK.”","La familiaridad del equipo puede preservar una arquitectura cuya hoja de ruta ha cambiado.","¿Cuál es el coste de quedarse frente a migrar durante el próximo ciclo de producto?",[534,535,536],"“Gestionado significa sin infraestructura.”","La Agents API todavía puede usar entornos autohospedados y tu aplicación sigue siendo propietaria de la lógica de producto.","¿Qué capa de infraestructura se está delegando realmente?",[538,539,540],"“Responses es solo para llamadas simples.”","Responses proporciona herramientas integradas y primitivas con estado; puede ser la base de bucles de agentes personalizados.","¿Necesitamos que la plataforma sea dueña del arnés, o solo de las primitivas de modelos\u002Fherramientas?",[542,543,544],"“Que esté completa a nivel de funciones significa que debemos migrar ahora.”","El SDK sigue manteniéndose para las aplicaciones existentes.","¿Qué requisito futuro concreto se ve bloqueado por quedarse?",{},{"id":547,"data":548,"type":42,"tunes":550},"h-migration",{"text":549,"level":219},"La migración es un cambio arquitectónico, no un renombramiento de importaciones",{},{"id":552,"data":553,"type":226,"tunes":555},"p-migration-1",{"text":554},"Pasar del Agents SDK a la Agents API cambia la titularidad. En el SDK, el bucle se ejecuta en tu aplicación. En la Agents API, OpenAI ejecuta el arnés y la sesión, mientras que tu aplicación se integra a través de tareas, eventos, herramientas y límites de entorno.",{},{"id":557,"data":558,"type":226,"tunes":560},"p-migration-2",{"text":559},"Por lo tanto, un plan de migración real debe mapear el estado de la sesión, la orquestación personalizada, los traspasos (handoffs), la ejecución de herramientas, las aprobaciones, el almacenamiento, el rastreo, los reintentos, la recuperación ante fallos, el ciclo de vida del entorno y cualquier abstracción específica del proveedor. El volumen de código puede disminuir mientras los supuestos operativos cambian.",{},{"id":562,"data":563,"type":42,"tunes":565},"h-migration-inventory",{"text":564,"level":218},"El inventario de migración",{},{"id":567,"data":568,"type":343,"tunes":581},"migration-list",{"meta":569,"items":570,"style":342},{},[571,572,573,574,575,576,577,578,579,580],"Definiciones de agentes y titularidad de las instrucciones.","Definiciones de herramientas y dónde se ejecuta cada herramienta.","Traspasos (handoffs), patrones de gestor\u002Fespecialista y comportamiento de subagentes.","Identificadores de sesión, estado de la conversación, capacidad de reanudación y retención del historial.","Aprobaciones humanas y semántica de interrupción.","Lógica personalizada de recorte o compactación de contexto.","Rastreo (tracing), evaluaciones, observabilidad y depuración en producción.","Archivos autohospedados, contenedores, acceso a redes privadas u otras dependencias de ejecución.","Abstracción de proveedores o dependencias de modelos ajenos a OpenAI.","Supuestos de reintento, tiempo de espera, idempotencia, recuperación y ciclo de vida.",{},{"id":583,"data":584,"type":42,"tunes":586},"h-beta",{"text":585,"level":219},"La beta pública cambia el modelo de riesgo",{},{"id":588,"data":589,"type":226,"tunes":591},"p-beta-1",{"text":590},"La Agents API es la dirección recomendada para nuevas aplicaciones de agentes, pero también está en beta pública. Esos hechos no son contradictorios. La dirección estratégica responde a “¿hacia dónde va la plataforma?”. El estado beta responde a “¿cuánto cambio operativo y de interfaz debo presupuestar?”.",{},{"id":593,"data":594,"type":226,"tunes":596},"p-beta-2",{"text":595},"Para los sistemas en producción, aísla la integración detrás de un límite de la aplicación. Mantén el estado del dominio, los permisos, los datos de auditoría y las reglas de negocio fuera de los objetos de sesión específicos del proveedor siempre que sea posible. Esto facilita absorber la evolución de la API sin convertir el entorno de ejecución del agente en la fuente de verdad de todo tu producto.",{},{"id":598,"data":599,"type":42,"tunes":601},"h-default",{"text":600,"level":219},"Una arquitectura predeterminada práctica",{},{"id":603,"data":604,"type":226,"tunes":606},"p-default-1",{"text":605},"Para muchas aplicaciones nativas de OpenAI nuevas, un valor predeterminado razonable para 2026 es: Agents API para el arnés gestionado y la sesión duradera, servicios de dominio y autorización propios de la aplicación, herramientas de función explícitas para acciones comerciales, y ejecución alojada en OpenAI o autohospedada según los requisitos de datos y cómputo.",{},{"id":608,"data":609,"type":226,"tunes":611},"p-default-2",{"text":610},"Eso mantiene la potencia del entorno de ejecución del agente sin convertirlo en el propietario de la verdad del negocio. La aplicación sigue decidiendo qué puede hacer un usuario, qué datos son autoritativos, qué acciones requieren aprobación y cómo se validan los resultados.",{},{"id":613,"data":614,"type":619,"tunes":620},"ref-reliability",{"url":615,"title":616,"excerpt":617,"ctaLabel":618},"https:\u002F\u002Fstajic.de\u002Fes\u002Fblog\u002Fai-agent-reliability-why-the-final-answer-is-not-enough","Fiabilidad de los agentes de IA: por qué la respuesta final no es suficiente","Un arnés gestionado reduce la carga operativa, pero la fiabilidad aún requiere una ejecución trazable, criterios de aceptación explícitos y evaluación más allá de la respuesta final.","Leer el artículo sobre fiabilidad","referralArticle",{},{"id":622,"data":623,"type":42,"tunes":625},"h-change",{"text":624,"level":219},"¿Qué cambiaría esta respuesta?",{},{"id":627,"data":628,"type":226,"tunes":630},"p-change-1",{"text":629},"La recomendación cambia si la Agents API añade o elimina capacidades, sale de la beta con contratos diferentes, cambia los límites de entorno o de precios, o introduce herramientas de migración que reduzcan las diferencias de titularidad. También cambia si tu aplicación depende de la portabilidad de proveedores, semánticas de orquestación personalizadas, ejecución exclusivamente local o una capacidad que el arnés alojado no pueda admitir.",{},{"id":632,"data":633,"type":226,"tunes":635},"p-change-2",{"text":634},"Para una aplicación existente de Agents SDK, la respuesta también varía en función del coste de migración. Si el sistema es estable, está bien evaluado y no se ve bloqueado por el estado de funciones completas del SDK, la migración inmediata puede generar más riesgos que valor. Si la hoja de ruta del producto depende de capacidades que solo llegan a la Agents API, retrasar la migración puede generar un tipo diferente de deuda.",{},{"id":637,"data":638,"type":42,"tunes":640},"h-limitations",{"text":639,"level":219},"Limitaciones",{},{"id":642,"data":643,"type":226,"tunes":645},"p-limit-1",{"text":644},"Esta comparación se centra en la titularidad del entorno de ejecución y la dirección de la plataforma declarada por OpenAI. No evalúa la latencia, la calidad ni el coste total para una carga de trabajo específica. Esas propiedades dependen de la elección del modelo, el uso de herramientas, el entorno, la duración de la tarea, el almacenamiento en caché, el uso de sandboxes y la arquitectura de la aplicación.",{},{"id":647,"data":648,"type":226,"tunes":650},"p-limit-2",{"text":649},"La API de Agents también es lo suficientemente nueva como para que la experiencia en producción aún se esté acumulando. Por lo tanto, un diseño debe validarse con cargas de trabajo representativas en lugar de elegirse únicamente a partir del posicionamiento del producto.",{},{"id":652,"data":653,"type":42,"tunes":655},"h-conclusion",{"text":654,"level":219},"Conclusión",{},{"id":657,"data":658,"type":226,"tunes":660},"p-conclusion-1",{"text":659},"La decisión sobre agentes de OpenAI en 2026 trata fundamentalmente sobre la propiedad del entorno de ejecución (runtime). La API de Agents significa que OpenAI asume una mayor parte del arnés y de la maquinaria de sesiones duraderas. Responses implica que su aplicación es propietaria del bucle en torno a las primitivas de la plataforma. El Agents SDK sigue siendo válido para los sistemas existentes, pero ya no es el destino predeterminado para las nuevas características importantes del runtime de agentes.",{},{"id":662,"data":663,"type":226,"tunes":665},"p-conclusion-2",{"text":664},"Para una aplicación nueva, siga la dirección de la plataforma a menos que un requisito real lo empuje hacia una capa inferior de la pila. Comience con la API de Agents, pase a Responses cuando necesite controlar el bucle, evalúe Codex SDK cuando necesite el arnés en su propia infraestructura y conserve el Agents SDK donde las inversiones existentes o las brechas temporales de capacidades lo justifiquen.",{},{"id":667,"data":668,"type":42,"tunes":670},"h-faq",{"text":669,"level":219},"Preguntas frecuentes",{},{"id":672,"data":673,"type":672,"tunes":700},"faq",{"items":674,"title":699},[675,679,683,687,691,695],{"id":676,"answer":677,"question":678},"faq1","Actualmente, OpenAI recomienda la API de Agents para nuevas aplicaciones de agentes. El Agents SDK cuenta con todas las funciones completas y sigue teniendo soporte para aplicaciones existentes, continuando con las labores de mantenimiento, seguridad, corrección de errores críticos y compatibilidad.","¿Debería usar la API de Agents de OpenAI o el Agents SDK para un nuevo proyecto?",{"id":680,"answer":681,"question":682},"faq2","OpenAI lo describe como completo a nivel de funciones (feature complete), no como sin soporte. No se prevén nuevas características importantes, pero continúan el mantenimiento, los parches de seguridad, la corrección de errores críticos y el trabajo de compatibilidad.","¿Está en desuso el Agents SDK?",{"id":684,"answer":685,"question":686},"faq3","Utilice Responses cuando su aplicación deba controlar el bucle de agentes, la estrategia de estado, la orquestación y la lógica de continuación mientras sigue utilizando los modelos y las herramientas de la plataforma de OpenAI.","¿Cuándo debería usar la API de Responses en lugar de la API de Agents?",{"id":688,"answer":689,"question":690},"faq4","No. La API de Agents puede no utilizar ningún entorno de ejecución, un entorno alojado en OpenAI o un entorno autoalojado conectado al arnés gestionado.","¿Requiere la API de Agents capacidad de cómputo alojada en OpenAI?",{"id":692,"answer":693,"question":694},"faq5","OpenAI posiciona el Codex SDK para ejecutar el arnés de Codex en una infraestructura operada por usted. Es la opción relevante cuando desea el arnés pero no el runtime alojado de la API de Agents.","¿Dónde encaja el Codex SDK?",{"id":696,"answer":697,"question":698},"faq6","No de forma automática. Evalúe si el sistema actual se ve limitado por el estado de funciones completas del SDK, si las capacidades requeridas existen en la API de Agents y si el beneficio de la migración supera el cambio operativo y arquitectónico.","¿Debería migrar inmediatamente una aplicación existente basada en Agents SDK?","Opciones de runtime para agentes de OpenAI en 2026",{},{"id":702,"data":703,"type":42,"tunes":705},"h-glossary",{"text":704,"level":219},"Glosario",{},{"id":707,"data":708,"type":707,"tunes":732},"glossary",{"title":709,"entries":710},"Términos clave del runtime",[711,714,718,722,725,728],{"term":712,"anchor":446,"definition":713},"Arnés (Harness)","El bucle de ejecución y la maquinaria de soporte que coordina las llamadas al modelo, herramientas, contexto, sesiones y la ejecución del agente.",{"term":715,"anchor":716,"definition":717},"API de Agents","agents-api","La API gestionada de OpenAI para agentes en la nube duraderos que utiliza un arnés de Codex alojado.",{"term":719,"anchor":720,"definition":721},"API de Responses","responses-api","La primitiva de API de bajo nivel de OpenAI para respuestas del modelo, herramientas alojadas e interacciones con estado alrededor de las cuales las aplicaciones pueden construir su propio bucle de agentes.",{"term":293,"anchor":723,"definition":724},"agents-sdk","El framework de código abierto de OpenAI para ejecutar flujos de trabajo de agentes en el código de la aplicación; con características completas a partir de septiembre de 2026.",{"term":299,"anchor":726,"definition":727},"codex-sdk","La opción de runtime que OpenAI enumera para ejecutar el arnés de Codex en infraestructura operada por usted.",{"term":729,"anchor":730,"definition":731},"Propiedad del runtime","runtime-ownership","El límite arquitectónico que describe qué partes del bucle del agente, el estado de la sesión, el entorno de ejecución y el ciclo de vida son operados por la plataforma frente a la aplicación.",{},{"id":734,"data":735,"type":42,"tunes":737},"h-sources",{"text":736,"level":219},"Fuentes principales y lecturas complementarias",{},{"id":739,"data":740,"type":746,"tunes":747},"src-launch",{"link":741,"meta":742},"https:\u002F\u002Fopenai.com\u002Findex\u002Fintroducing-the-agents-api\u002F",{"image":743,"title":744,"description":745},{"url":440},"OpenAI — Presentación de la API de Agents","Anuncio del lanzamiento de la API de Agents el 10 de septiembre de 2026, que describe el arnés gestionado de Codex y la versión beta pública.","linkTool",{},{"id":749,"data":750,"type":746,"tunes":756},"src-agents-overview",{"link":751,"meta":752},"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fagents",{"image":753,"title":754,"description":755},{"url":440},"OpenAI — Descripción general del runtime de agentes","Comparación actual de la API de Agents, Codex SDK y la API de Responses, incluyendo el estado de soporte del Agents SDK.",{},{"id":758,"data":759,"type":746,"tunes":765},"src-agents-api",{"link":760,"meta":761},"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fagents-api\u002Foverview",{"image":762,"title":763,"description":764},{"url":440},"OpenAI — Descripción general de la API de Agents","Documentación sobre agentes duraderos en la nube, sesiones, orquestación, compactación de contexto, recuperación y opciones de entorno.",{},{"id":767,"data":768,"type":746,"tunes":774},"src-architecture",{"link":769,"meta":770},"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fagents-api\u002Farchitecture",{"image":771,"title":772,"description":773},{"url":440},"OpenAI — Arquitectura de la API de Agents","Límite arquitectónico entre el arnés alojado, el servidor de aplicaciones y los entornos de ejecución sin entorno\u002Falojados en OpenAI\u002Fautoalojados.",{},{"id":776,"data":777,"type":746,"tunes":783},"src-sdk",{"link":778,"meta":779},"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fagents\u002Fsdk",{"image":780,"title":781,"description":782},{"url":440},"OpenAI — Agents SDK","Aviso actual de soporte para el Agents SDK y explicación del bucle de agente controlado por la aplicación.",{},{"id":785,"data":786,"type":746,"tunes":792},"src-responses",{"link":787,"meta":788},"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fmigrate-to-responses",{"image":789,"title":790,"description":791},{"url":440},"OpenAI — Migración a la API de Responses","Posicionamiento actual de la API de Responses, herramientas integradas, contexto con estado y primitivas de agentes.",{},{"id":794,"data":795,"type":220,"tunes":797},"toc",{"title":796,"maxLevel":218,"minLevel":219},"Contenidos",{},"2.31","El stack de agentes de OpenAI cambió en septiembre de 2026. Esta guía de arquitectura separa la Agents API, el Agents SDK, la Responses API y el Codex SDK por propiedad del runtime—para que los equipos puedan elegir el límite de control adecuado en lugar de comparar nombres de productos.","\u002Fuploads\u002F2026\u002F09\u002Fopenai-agents-api-vs-agents-sdk-vs-responses-api-what-should-you-build-on-in-2026-1790351846714-zi7lus.webp","openai-agents-api-vs-agents-sdk-vs-responses-api-what-should-you-build-on-in-2026-1790351846714-zi7lus","PUBLISHED","2026-09-25T11:56:00.000Z","2026-09-25T15:56:16.274Z","2026-09-25T19:58:41.459Z",{"en":807,"de":808,"sr":809,"es":810,"fr":811,"it":812,"ru":813,"zh":814},"\u002Fblog\u002Fopenai-agents-api-vs-agents-sdk-vs-responses-api-what-should-you-build-on-in-2026","\u002Fde\u002Fblog\u002Fopenai-agents-api-vs-agents-sdk-vs-responses-api-what-should-you-build-on-in-2026","\u002Fsr\u002Fblog\u002Fopenai-agents-api-vs-agents-sdk-vs-responses-api-what-should-you-build-on-in-2026","\u002Fes\u002Fblog\u002Fopenai-agents-api-vs-agents-sdk-vs-responses-api-what-should-you-build-on-in-2026","\u002Ffr\u002Fblog\u002Fopenai-agents-api-vs-agents-sdk-vs-responses-api-what-should-you-build-on-in-2026","\u002Fit\u002Fblog\u002Fopenai-agents-api-vs-agents-sdk-vs-responses-api-what-should-you-build-on-in-2026","\u002Fru\u002Fblog\u002Fopenai-agents-api-vs-agents-sdk-vs-responses-api-what-should-you-build-on-in-2026","\u002Fzh\u002Fblog\u002Fopenai-agents-api-vs-agents-sdk-vs-responses-api-what-should-you-build-on-in-2026",[816,819,823],{"id":101,"name":817,"slug":818},"Overview","overview-digital-platform",{"id":820,"name":821,"slug":822},48,"Capacidades","capabilities",{"id":824,"name":825,"slug":826},46,"Resumen","overview",{"id":828,"login":829,"email":830,"displayName":831},"20","rooth8233","aleksandar@stajic.de","Aleksandar Stajić",[833,1304],{"lang":834,"title":835,"content":836,"contentJson":837,"excerpt":1303},"en","OpenAI Agents API vs Agents SDK vs Responses API: What Should You Build On in 2026?","{\"time\":1790352010644,\"blocks\":[{\"id\":\"75qcDhG72q\",\"type\":\"tableOfContents\",\"data\":{\"title\":\"Contents\",\"minLevel\":2,\"maxLevel\":3},\"tunes\":{}},{\"id\":\"intro\",\"type\":\"paragraph\",\"data\":{\"text\":\"OpenAI's agent stack changed materially in September 2026. The new Agents API introduced a managed Codex harness for durable cloud agents, while the older Agents SDK moved into a feature-complete maintenance phase. The Responses API remains the lower-level surface for applications that want direct model calls or to own the agent loop themselves. These are not three interchangeable wrappers around the same thing: they place the runtime boundary in different places.\"},\"tunes\":{}},{\"id\":\"direct\",\"type\":\"callout\",\"data\":{\"variant\":\"info\",\"title\":\"Direct answer\",\"body\":\"\u003Cstrong>For a new OpenAI-native agent application in September 2026, start with the Agents API unless you have a concrete reason to own more of the runtime.\u003C\u002Fstrong> Use the Responses API when your application should own the loop, state strategy, and orchestration. Keep the Agents SDK for existing applications or short-term gaps where the Agents API does not yet support a required capability. If you want the Codex harness itself but need to run it in infrastructure you operate, evaluate the Codex SDK rather than forcing the Agents SDK into that role.\"},\"tunes\":{}},{\"id\":\"date-warning\",\"type\":\"callout\",\"data\":{\"variant\":\"warning\",\"title\":\"Time-sensitive architecture decision\",\"body\":\"This article reflects OpenAI's platform state on \u003Cstrong>25 September 2026\u003C\u002Fstrong>. The Agents API launched in public beta on 10 September 2026. Beta surfaces, feature coverage, pricing, and migration guidance can change; verify current documentation before committing a production architecture.\"},\"tunes\":{}},{\"id\":\"model-note\",\"type\":\"callout\",\"data\":{\"variant\":\"note\",\"title\":\"The decision model used here\",\"body\":\"The Runtime Ownership Test below is a practical architecture model, not an OpenAI product taxonomy. It asks which parts of the agent runtime your team must control versus which parts you want OpenAI to operate.\"},\"tunes\":{}},{\"id\":\"h-boundary\",\"type\":\"header\",\"data\":{\"text\":\"The architecture changed: choose a runtime boundary, not a library\",\"level\":2},\"tunes\":{}},{\"id\":\"p-boundary-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"The important decision is no longer simply “Which SDK should I install?” It is who owns the harness, the agent loop, durable session state, context compaction, recovery, execution environment, and application lifecycle.\"},\"tunes\":{}},{\"id\":\"p-boundary-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"OpenAI's current Agents overview makes this boundary explicit. The Agents API runs a hosted Codex harness and manages orchestration plus durable session state. The Responses API provides model responses and hosted capabilities while your application owns the surrounding agent loop. The Agents SDK runs the loop in your application and is now feature complete rather than the forward path for major new agent features.\"},\"tunes\":{}},{\"id\":\"h-short\",\"type\":\"header\",\"data\":{\"text\":\"The short comparison\",\"level\":2},\"tunes\":{}},{\"id\":\"short-table\",\"type\":\"table\",\"data\":{\"withHeadings\":true,\"stretched\":false,\"content\":[[\"Option\",\"Best fit in 2026\",\"Who owns the agent loop?\",\"Session \u002F context responsibility\",\"Strategic status\"],[\"Agents API\",\"New OpenAI-native durable agents\",\"OpenAI-managed Codex harness\",\"OpenAI manages sessions, orchestration, compaction and recovery\",\"Recommended starting point for new agent apps; public beta\"],[\"Responses API\",\"Direct model integrations and custom agent runtimes\",\"Your application\",\"You choose response chaining, Conversations, storage and loop logic\",\"Core API primitive; recommended over Chat Completions for new projects\"],[\"Agents SDK\",\"Existing SDK applications or temporary capability gaps\",\"Your application via the SDK runner\",\"Your application operates deployment, storage and runtime behaviour\",\"Feature complete; maintenance and compatibility continue, major new features not planned\"],[\"Codex SDK\",\"Codex harness in infrastructure you operate\",\"Codex harness in your environment\",\"You operate harness hosting and lifecycle\",\"Separate option when you want the harness without the hosted Agents API runtime\"]]},\"tunes\":{}},{\"id\":\"h-agents-api\",\"type\":\"header\",\"data\":{\"text\":\"1. Agents API: managed harness, durable cloud agent\",\"level\":2},\"tunes\":{}},{\"id\":\"p-api-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"The Agents API exposes the Codex harness through an OpenAI-managed service. OpenAI manages sessions, orchestration, context compaction, and recovery. Your application still provides tools and chooses the execution environment.\"},\"tunes\":{}},{\"id\":\"p-api-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"That last distinction matters. “Managed agent” does not necessarily mean “all compute runs inside OpenAI.” The Agents API architecture supports no environment, an OpenAI-hosted environment, or a self-hosted environment connected to the hosted harness. With a self-hosted environment, your application owns provisioning, reconnection, shutdown, and persistent files while the harness remains managed.\"},\"tunes\":{}},{\"id\":\"p-api-3\",\"type\":\"paragraph\",\"data\":{\"text\":\"The Agents API is therefore a runtime service, not merely a request format. Sessions can persist, stream progress, receive additional tasks, use tools, work with files, and recover across long-running work.\"},\"tunes\":{}},{\"id\":\"h-api-gain\",\"type\":\"header\",\"data\":{\"text\":\"What you gain with the Agents API\",\"level\":3},\"tunes\":{}},{\"id\":\"api-gains\",\"type\":\"list\",\"data\":{\"style\":\"unordered\",\"meta\":{},\"items\":[\"A managed Codex harness instead of building and operating the main agent loop yourself.\",\"Durable sessions for work that spans turns and long-running tasks.\",\"Managed orchestration, context compaction, and recovery.\",\"OpenAI-hosted or self-hosted execution environments, depending on workload requirements.\",\"Streaming and webhooks for progress and lifecycle events.\",\"A platform direction that OpenAI explicitly recommends for new agent applications.\"]},\"tunes\":{}},{\"id\":\"h-api-own\",\"type\":\"header\",\"data\":{\"text\":\"What you still own\",\"level\":3},\"tunes\":{}},{\"id\":\"api-own\",\"type\":\"list\",\"data\":{\"style\":\"unordered\",\"meta\":{},\"items\":[\"Your product and application server.\",\"Function-tool implementations and business logic.\",\"Authorization and policy decisions around your own systems.\",\"The execution environment lifecycle when you choose self-hosted compute.\",\"Evaluation, acceptance criteria, domain-specific guardrails, and the decision of what the agent is allowed to do.\"]},\"tunes\":{}},{\"id\":\"h-responses\",\"type\":\"header\",\"data\":{\"text\":\"2. Responses API: own the loop, use the platform primitives\",\"level\":2},\"tunes\":{}},{\"id\":\"p-resp-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"The Responses API is the lower-level choice when you want OpenAI model and tool capabilities without delegating the overall agent runtime. OpenAI describes Responses as the recommended API primitive for new projects and as an evolution of Chat Completions with built-in tools, multi-turn state options, multimodal input, and agentic tool use.\"},\"tunes\":{}},{\"id\":\"p-resp-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"A Responses request can itself invoke tools, but your application remains responsible for the larger workflow when you build an agent around it. That means your code decides how to persist application state, when to continue, how to recover, how to coordinate specialists, how to compact long histories, and how to represent resumable work.\"},\"tunes\":{}},{\"id\":\"p-resp-3\",\"type\":\"paragraph\",\"data\":{\"text\":\"This is not inherently inferior. It is the right boundary when agent behaviour must be deeply embedded in existing application logic, when you need a custom state model, or when a managed harness would hide control you actually need.\"},\"tunes\":{}},{\"id\":\"h-sdk\",\"type\":\"header\",\"data\":{\"text\":\"3. Agents SDK: supported, but no longer the default forward path\",\"level\":2},\"tunes\":{}},{\"id\":\"p-sdk-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"The Agents SDK remains an open-source framework for running agent workflows in your application. It provides agent definitions, tools, handoffs, guardrails, tracing, sessions, and the runner loop in TypeScript and Python.\"},\"tunes\":{}},{\"id\":\"p-sdk-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"But its strategic status changed. OpenAI now labels the Agents SDK feature complete: maintenance, security fixes, critical bug fixes, and compatibility work continue, but major new features are not planned. OpenAI recommends the Agents API for new applications.\"},\"tunes\":{}},{\"id\":\"p-sdk-3\",\"type\":\"paragraph\",\"data\":{\"text\":\"That does not mean an existing SDK application should be rewritten immediately. It means the architecture should stop assuming that the SDK is where the next major agent-runtime capabilities will land.\"},\"tunes\":{}},{\"id\":\"sdk-warning\",\"type\":\"callout\",\"data\":{\"variant\":\"warning\",\"title\":\"Do not translate “feature complete” into “deprecated”\",\"body\":\"OpenAI has not described the Agents SDK as unsupported. Existing applications can continue to use it, and maintenance, security, critical bug fixes, and compatibility work continue. The important change is that major new features are not planned and new applications are directed to the Agents API.\"},\"tunes\":{}},{\"id\":\"h-codex-sdk\",\"type\":\"header\",\"data\":{\"text\":\"Where the Codex SDK fits\",\"level\":2},\"tunes\":{}},{\"id\":\"p-codex-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"The current choice is not a simple three-way fork. OpenAI's runtime overview includes the Codex SDK as the option for running the Codex harness in infrastructure you operate. That is architecturally different from both the hosted Agents API and the Agents SDK.\"},\"tunes\":{}},{\"id\":\"p-codex-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"If your real requirement is “I want the Codex harness, but I need to operate it myself,” the Codex SDK is the surface to evaluate. If your requirement is “I want to own the loop around model calls,” evaluate Responses. If your requirement is “I already have a working Agents SDK application,” the existing SDK can remain valid while you plan against its maintenance status.\"},\"tunes\":{}},{\"id\":\"h-test\",\"type\":\"header\",\"data\":{\"text\":\"The Runtime Ownership Test\",\"level\":2},\"tunes\":{}},{\"id\":\"p-test-intro\",\"type\":\"paragraph\",\"data\":{\"text\":\"A useful architecture decision starts by deciding what your team must own. Score each requirement as either must control, prefer to control, or prefer managed.\"},\"tunes\":{}},{\"id\":\"ownership-test\",\"type\":\"comparison\",\"data\":{\"title\":\"Runtime Ownership Test\",\"layout\":\"table\",\"columns\":[{\"id\":\"decision\",\"label\":\"Decision\"},{\"id\":\"managed\",\"label\":\"If you prefer managed\"},{\"id\":\"control\",\"label\":\"If you require control\"}],\"rows\":[{\"id\":\"loop\",\"label\":\"Agent loop\",\"values\":[\"\",\"\",\"\"]},{\"id\":\"session\",\"label\":\"Durable sessions\",\"values\":[\"\",\"\",\"\"]},{\"id\":\"harness\",\"label\":\"Harness runtime\",\"values\":[\"\",\"\",\"\"]},{\"id\":\"environment\",\"label\":\"Execution environment\",\"values\":[\"\",\"\",\"\"]},{\"id\":\"orchestration\",\"label\":\"Orchestration semantics\",\"values\":[\"\",\"\",\"\"]},{\"id\":\"portability\",\"label\":\"Provider \u002F transport flexibility\",\"values\":[\"\",\"\",\"\"]},{\"id\":\"ops\",\"label\":\"Operational burden\",\"values\":[\"\",\"\",\"\"]}]},\"tunes\":{}},{\"id\":\"h-tree\",\"type\":\"header\",\"data\":{\"text\":\"A decision tree for new systems\",\"level\":2},\"tunes\":{}},{\"id\":\"decision-flow\",\"type\":\"processFlow\",\"data\":{\"title\":\"Choose the runtime by the control boundary\",\"orientation\":\"auto\",\"steps\":[{\"label\":\"1. Is this a new agent application?\",\"description\":\"If no, do not migrate solely because a newer surface exists. Evaluate the current application's actual constraints first.\"},{\"label\":\"2. Do you want a managed long-running agent harness?\",\"description\":\"If yes, start with the Agents API as OpenAI's recommended path for new agent applications.\"},{\"label\":\"3. Do you need the Codex harness but must operate it yourself?\",\"description\":\"Evaluate the Codex SDK rather than rebuilding harness behaviour on top of the Agents SDK.\"},{\"label\":\"4. Do you need to own the agent loop and state model?\",\"description\":\"Use the Responses API as the lower-level platform primitive and build the loop around it.\"},{\"label\":\"5. Are you already on the Agents SDK?\",\"description\":\"Continue if it meets requirements; major new features are not planned, so treat future platform adoption as an explicit roadmap decision.\"},{\"label\":\"6. Is a required capability missing from the Agents API?\",\"description\":\"OpenAI explicitly allows the Agents SDK as a short-term option for new applications that need unsupported capabilities.\"},{\"label\":\"7. Validate with a production-shaped spike\",\"description\":\"Test tools, environment, approvals, latency, observability, failure recovery, and lifecycle before locking the architecture.\"}]},\"tunes\":{}},{\"id\":\"h-not-drive\",\"type\":\"header\",\"data\":{\"text\":\"What should not drive the decision\",\"level\":2},\"tunes\":{}},{\"id\":\"weak-rules\",\"type\":\"table\",\"data\":{\"withHeadings\":true,\"stretched\":false,\"content\":[[\"Weak decision rule\",\"Why it fails\",\"Better question\"],[\"“The newest API must be best.”\",\"Newer can be strategically preferred while still lacking a capability you need.\",\"Which runtime responsibilities should be managed versus application-owned?\"],[\"“We already know the SDK.”\",\"Team familiarity can preserve an architecture whose roadmap has changed.\",\"What is the cost of staying versus moving over the next product cycle?\"],[\"“Managed means no infrastructure.”\",\"Agents API can still use self-hosted environments and your application still owns product logic.\",\"Which infrastructure layer is actually being delegated?\"],[\"“Responses is only for simple calls.”\",\"Responses provides built-in tools and stateful primitives; it can be the foundation of custom agent loops.\",\"Do we need the platform to own the harness, or only model\u002Ftool primitives?\"],[\"“Feature complete means we must migrate now.”\",\"The SDK remains maintained for existing applications.\",\"What concrete future requirement is blocked by staying?\"]]},\"tunes\":{}},{\"id\":\"h-migration\",\"type\":\"header\",\"data\":{\"text\":\"Migration is an architectural change, not an import rename\",\"level\":2},\"tunes\":{}},{\"id\":\"p-migration-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"Moving from the Agents SDK to the Agents API changes ownership. In the SDK, the loop runs in your application. In the Agents API, OpenAI runs the harness and session while your application integrates through tasks, events, tools, and environment boundaries.\"},\"tunes\":{}},{\"id\":\"p-migration-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"A real migration plan therefore needs to map session state, custom orchestration, handoffs, tool execution, approvals, storage, tracing, retries, failure recovery, environment lifecycle, and any provider-specific abstractions. Code volume may fall while operational assumptions change.\"},\"tunes\":{}},{\"id\":\"h-migration-inventory\",\"type\":\"header\",\"data\":{\"text\":\"The migration inventory\",\"level\":3},\"tunes\":{}},{\"id\":\"migration-list\",\"type\":\"list\",\"data\":{\"style\":\"unordered\",\"meta\":{},\"items\":[\"Agent definitions and instruction ownership.\",\"Tool definitions and where each tool executes.\",\"Handoffs, manager\u002Fspecialist patterns, and subagent behaviour.\",\"Session identifiers, conversation state, resumability, and history retention.\",\"Human approvals and interruption semantics.\",\"Custom context trimming or compaction logic.\",\"Tracing, evals, observability, and production debugging.\",\"Self-hosted files, containers, private-network access, or other execution dependencies.\",\"Provider abstraction or non-OpenAI model dependencies.\",\"Retry, timeout, idempotency, recovery, and lifecycle assumptions.\"]},\"tunes\":{}},{\"id\":\"h-beta\",\"type\":\"header\",\"data\":{\"text\":\"Public beta changes the risk model\",\"level\":2},\"tunes\":{}},{\"id\":\"p-beta-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"The Agents API is the recommended direction for new agent applications, but it is also in public beta. Those facts are not contradictory. Strategic direction answers “where is the platform going?” Beta status answers “how much interface and operational change should I budget for?”\"},\"tunes\":{}},{\"id\":\"p-beta-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"For production systems, isolate the integration behind an application boundary. Keep domain state, permissions, audit data, and business rules outside vendor-specific session objects where possible. This makes it easier to absorb API evolution without turning the agent runtime into the source of truth for your entire product.\"},\"tunes\":{}},{\"id\":\"h-default\",\"type\":\"header\",\"data\":{\"text\":\"A practical default architecture\",\"level\":2},\"tunes\":{}},{\"id\":\"p-default-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"For many new OpenAI-native applications, a reasonable 2026 default is: Agents API for the managed harness and durable session, application-owned domain services and authorization, explicit function tools for business actions, and either OpenAI-hosted or self-hosted execution depending on data and compute requirements.\"},\"tunes\":{}},{\"id\":\"p-default-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"That keeps the agent runtime powerful without making it the owner of business truth. The application still decides what a user may do, which data is authoritative, which actions require approval, and how results are validated.\"},\"tunes\":{}},{\"id\":\"ref-reliability\",\"type\":\"referralArticle\",\"data\":{\"url\":\"https:\u002F\u002Fstajic.de\u002Fblog\u002Fai-agent-reliability-why-the-final-answer-is-not-enough\",\"title\":\"AI Agent Reliability: Why the Final Answer Is Not Enough\",\"excerpt\":\"A managed harness reduces operational burden, but reliability still requires traceable execution, explicit acceptance criteria and evaluation beyond the final answer.\",\"ctaLabel\":\"Read the reliability article\"},\"tunes\":{}},{\"id\":\"h-change\",\"type\":\"header\",\"data\":{\"text\":\"What would change this answer?\",\"level\":2},\"tunes\":{}},{\"id\":\"p-change-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"The recommendation changes if the Agents API adds or removes capabilities, exits beta with different contracts, changes environment or pricing boundaries, or introduces migration tooling that reduces ownership differences. It also changes if your application depends on provider portability, custom orchestration semantics, local-only execution, or a capability the hosted harness cannot support.\"},\"tunes\":{}},{\"id\":\"p-change-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"For an existing Agents SDK application, the answer also changes with migration cost. If the system is stable, well evaluated, and not blocked by the SDK's feature-complete status, immediate migration may create more risk than value. If the product roadmap depends on capabilities landing only in the Agents API, delaying migration can create a different kind of debt.\"},\"tunes\":{}},{\"id\":\"h-limitations\",\"type\":\"header\",\"data\":{\"text\":\"Limitations\",\"level\":2},\"tunes\":{}},{\"id\":\"p-limit-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"This comparison focuses on runtime ownership and OpenAI's stated platform direction. It does not benchmark latency, quality, or total cost for a specific workload. Those properties depend on model choice, tool usage, environment, task length, caching, sandbox usage, and application architecture.\"},\"tunes\":{}},{\"id\":\"p-limit-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"The Agents API is also new enough that production experience is still accumulating. A design should therefore be validated with representative workloads rather than chosen solely from product positioning.\"},\"tunes\":{}},{\"id\":\"h-conclusion\",\"type\":\"header\",\"data\":{\"text\":\"Conclusion\",\"level\":2},\"tunes\":{}},{\"id\":\"p-conclusion-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"The 2026 OpenAI agent decision is fundamentally about runtime ownership. Agents API means OpenAI owns more of the harness and durable-session machinery. Responses means your application owns the loop around platform primitives. The Agents SDK remains valid for existing systems but is no longer the default destination for major new agent-runtime features.\"},\"tunes\":{}},{\"id\":\"p-conclusion-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"For a new application, follow the platform direction unless a real requirement pushes you lower in the stack. Start with the Agents API, move to Responses when you need to own the loop, evaluate Codex SDK when you need the harness in your infrastructure, and keep the Agents SDK where existing investments or temporary capability gaps justify it.\"},\"tunes\":{}},{\"id\":\"h-faq\",\"type\":\"header\",\"data\":{\"text\":\"FAQ\",\"level\":2},\"tunes\":{}},{\"id\":\"faq\",\"type\":\"faq\",\"data\":{\"title\":\"OpenAI agent runtime choices in 2026\",\"items\":[{\"id\":\"faq1\",\"question\":\"Should I use the OpenAI Agents API or Agents SDK for a new project?\",\"answer\":\"OpenAI currently recommends the Agents API for new agent applications. The Agents SDK is feature complete and remains supported for existing applications, with maintenance, security, critical bug fixes and compatibility work continuing.\"},{\"id\":\"faq2\",\"question\":\"Is the Agents SDK deprecated?\",\"answer\":\"OpenAI describes it as feature complete, not as unsupported. Major new features are not planned, but maintenance, security fixes, critical bug fixes and compatibility work continue.\"},{\"id\":\"faq3\",\"question\":\"When should I use the Responses API instead of the Agents API?\",\"answer\":\"Use Responses when your application should own the agent loop, state strategy, orchestration and continuation logic while still using OpenAI models and platform tools.\"},{\"id\":\"faq4\",\"question\":\"Does the Agents API require OpenAI-hosted compute?\",\"answer\":\"No. The Agents API can use no execution environment, an OpenAI-hosted environment, or a self-hosted environment connected to the managed harness.\"},{\"id\":\"faq5\",\"question\":\"Where does the Codex SDK fit?\",\"answer\":\"OpenAI positions the Codex SDK for running the Codex harness in infrastructure you operate. It is the relevant option when you want the harness but not the hosted Agents API runtime.\"},{\"id\":\"faq6\",\"question\":\"Should an existing Agents SDK app migrate immediately?\",\"answer\":\"Not automatically. Evaluate whether the current system is blocked by the SDK's feature-complete status, whether required capabilities exist in the Agents API, and whether the migration benefit exceeds the operational and architectural change.\"}]},\"tunes\":{}},{\"id\":\"h-glossary\",\"type\":\"header\",\"data\":{\"text\":\"Glossary\",\"level\":2},\"tunes\":{}},{\"id\":\"glossary\",\"type\":\"glossary\",\"data\":{\"title\":\"Key runtime terms\",\"entries\":[{\"term\":\"Harness\",\"definition\":\"The runtime loop and supporting machinery that coordinates model calls, tools, context, sessions and agent execution.\",\"anchor\":\"harness\"},{\"term\":\"Agents API\",\"definition\":\"OpenAI's managed API for durable cloud agents using a hosted Codex harness.\",\"anchor\":\"agents-api\"},{\"term\":\"Responses API\",\"definition\":\"OpenAI's lower-level API primitive for model responses, hosted tools and stateful interactions around which applications can build their own agent loop.\",\"anchor\":\"responses-api\"},{\"term\":\"Agents SDK\",\"definition\":\"OpenAI's open-source framework for running agent workflows in application code; feature complete as of September 2026.\",\"anchor\":\"agents-sdk\"},{\"term\":\"Codex SDK\",\"definition\":\"The runtime option OpenAI lists for running the Codex harness in infrastructure you operate.\",\"anchor\":\"codex-sdk\"},{\"term\":\"Runtime ownership\",\"definition\":\"The architectural boundary describing which parts of the agent loop, session state, execution environment and lifecycle are operated by the platform versus the application.\",\"anchor\":\"runtime-ownership\"}]},\"tunes\":{}},{\"id\":\"h-sources\",\"type\":\"header\",\"data\":{\"text\":\"Primary sources and further reading\",\"level\":2},\"tunes\":{}},{\"id\":\"src-launch\",\"type\":\"linkTool\",\"data\":{\"link\":\"https:\u002F\u002Fopenai.com\u002Findex\u002Fintroducing-the-agents-api\u002F\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"OpenAI — Introducing the Agents API\",\"description\":\"Launch announcement for the Agents API on 10 September 2026, describing the managed Codex harness and public beta.\"}},\"tunes\":{}},{\"id\":\"src-agents-overview\",\"type\":\"linkTool\",\"data\":{\"link\":\"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fagents\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"OpenAI — Agents runtime overview\",\"description\":\"Current comparison of Agents API, Codex SDK and Responses API, including the support status of the Agents SDK.\"}},\"tunes\":{}},{\"id\":\"src-agents-api\",\"type\":\"linkTool\",\"data\":{\"link\":\"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fagents-api\u002Foverview\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"OpenAI — Agents API overview\",\"description\":\"Documentation for durable cloud agents, sessions, orchestration, context compaction, recovery and environment choices.\"}},\"tunes\":{}},{\"id\":\"src-architecture\",\"type\":\"linkTool\",\"data\":{\"link\":\"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fagents-api\u002Farchitecture\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"OpenAI — Agents API architecture\",\"description\":\"Architecture boundary between the hosted harness, application server and no\u002Fopenai-hosted\u002Fself-hosted execution environments.\"}},\"tunes\":{}},{\"id\":\"src-sdk\",\"type\":\"linkTool\",\"data\":{\"link\":\"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fagents\u002Fsdk\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"OpenAI — Agents SDK\",\"description\":\"Current Agents SDK support notice and explanation of the application-owned agent loop.\"}},\"tunes\":{}},{\"id\":\"src-responses\",\"type\":\"linkTool\",\"data\":{\"link\":\"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fmigrate-to-responses\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"OpenAI — Migrate to the Responses API\",\"description\":\"Current Responses API positioning, built-in tools, stateful context and agentic primitives.\"}},\"tunes\":{}},{\"id\":\"toc\",\"type\":\"tableOfContents\",\"data\":{\"title\":\"Contents\",\"minLevel\":2,\"maxLevel\":3},\"tunes\":{}}],\"version\":\"2.31.6\"}",{"time":838,"blocks":839,"version":1302},1790352010644,[840,844,848,853,858,863,867,871,875,879,909,913,917,921,925,929,940,944,954,958,962,966,970,974,978,982,986,991,995,999,1003,1007,1011,1044,1048,1074,1078,1106,1110,1114,1118,1122,1137,1141,1145,1149,1153,1157,1161,1168,1172,1176,1180,1184,1188,1192,1196,1200,1204,1208,1231,1235,1254,1258,1265,1272,1279,1286,1292,1299],{"id":215,"data":841,"type":220,"tunes":843},{"title":842,"maxLevel":218,"minLevel":219},"Contents",{},{"id":223,"data":845,"type":226,"tunes":847},{"text":846},"OpenAI's agent stack changed materially in September 2026. The new Agents API introduced a managed Codex harness for durable cloud agents, while the older Agents SDK moved into a feature-complete maintenance phase. The Responses API remains the lower-level surface for applications that want direct model calls or to own the agent loop themselves. These are not three interchangeable wrappers around the same thing: they place the runtime boundary in different places.",{},{"id":229,"data":849,"type":234,"tunes":852},{"body":850,"title":851,"variant":233},"\u003Cstrong>For a new OpenAI-native agent application in September 2026, start with the Agents API unless you have a concrete reason to own more of the runtime.\u003C\u002Fstrong> Use the Responses API when your application should own the loop, state strategy, and orchestration. Keep the Agents SDK for existing applications or short-term gaps where the Agents API does not yet support a required capability. If you want the Codex harness itself but need to run it in infrastructure you operate, evaluate the Codex SDK rather than forcing the Agents SDK into that role.","Direct answer",{},{"id":237,"data":854,"type":234,"tunes":857},{"body":855,"title":856,"variant":241},"This article reflects OpenAI's platform state on \u003Cstrong>25 September 2026\u003C\u002Fstrong>. The Agents API launched in public beta on 10 September 2026. Beta surfaces, feature coverage, pricing, and migration guidance can change; verify current documentation before committing a production architecture.","Time-sensitive architecture decision",{},{"id":244,"data":859,"type":234,"tunes":862},{"body":860,"title":861,"variant":248},"The Runtime Ownership Test below is a practical architecture model, not an OpenAI product taxonomy. It asks which parts of the agent runtime your team must control versus which parts you want OpenAI to operate.","The decision model used here",{},{"id":251,"data":864,"type":42,"tunes":866},{"text":865,"level":219},"The architecture changed: choose a runtime boundary, not a library",{},{"id":256,"data":868,"type":226,"tunes":870},{"text":869},"The important decision is no longer simply “Which SDK should I install?” It is who owns the harness, the agent loop, durable session state, context compaction, recovery, execution environment, and application lifecycle.",{},{"id":261,"data":872,"type":226,"tunes":874},{"text":873},"OpenAI's current Agents overview makes this boundary explicit. The Agents API runs a hosted Codex harness and manages orchestration plus durable session state. The Responses API provides model responses and hosted capabilities while your application owns the surrounding agent loop. The Agents SDK runs the loop in your application and is now feature complete rather than the forward path for major new agent features.",{},{"id":266,"data":876,"type":42,"tunes":878},{"text":877,"level":219},"The short comparison",{},{"id":271,"data":880,"type":304,"tunes":908},{"content":881,"stretched":43,"withHeadings":14},[882,888,893,898,903],[883,884,885,886,887],"Option","Best fit in 2026","Who owns the agent loop?","Session \u002F context responsibility","Strategic status",[281,889,890,891,892],"New OpenAI-native durable agents","OpenAI-managed Codex harness","OpenAI manages sessions, orchestration, compaction and recovery","Recommended starting point for new agent apps; public beta",[287,894,895,896,897],"Direct model integrations and custom agent runtimes","Your application","You choose response chaining, Conversations, storage and loop logic","Core API primitive; recommended over Chat Completions for new projects",[293,899,900,901,902],"Existing SDK applications or temporary capability gaps","Your application via the SDK runner","Your application operates deployment, storage and runtime behaviour","Feature complete; maintenance and compatibility continue, major new features not planned",[299,904,905,906,907],"Codex harness in infrastructure you operate","Codex harness in your environment","You operate harness hosting and lifecycle","Separate option when you want the harness without the hosted Agents API runtime",{},{"id":307,"data":910,"type":42,"tunes":912},{"text":911,"level":219},"1. Agents API: managed harness, durable cloud agent",{},{"id":312,"data":914,"type":226,"tunes":916},{"text":915},"The Agents API exposes the Codex harness through an OpenAI-managed service. OpenAI manages sessions, orchestration, context compaction, and recovery. Your application still provides tools and chooses the execution environment.",{},{"id":317,"data":918,"type":226,"tunes":920},{"text":919},"That last distinction matters. “Managed agent” does not necessarily mean “all compute runs inside OpenAI.” The Agents API architecture supports no environment, an OpenAI-hosted environment, or a self-hosted environment connected to the hosted harness. With a self-hosted environment, your application owns provisioning, reconnection, shutdown, and persistent files while the harness remains managed.",{},{"id":322,"data":922,"type":226,"tunes":924},{"text":923},"The Agents API is therefore a runtime service, not merely a request format. Sessions can persist, stream progress, receive additional tasks, use tools, work with files, and recover across long-running work.",{},{"id":327,"data":926,"type":42,"tunes":928},{"text":927,"level":218},"What you gain with the Agents API",{},{"id":332,"data":930,"type":343,"tunes":939},{"meta":931,"items":932,"style":342},{},[933,934,935,936,937,938],"A managed Codex harness instead of building and operating the main agent loop yourself.","Durable sessions for work that spans turns and long-running tasks.","Managed orchestration, context compaction, and recovery.","OpenAI-hosted or self-hosted execution environments, depending on workload requirements.","Streaming and webhooks for progress and lifecycle events.","A platform direction that OpenAI explicitly recommends for new agent applications.",{},{"id":346,"data":941,"type":42,"tunes":943},{"text":942,"level":218},"What you still own",{},{"id":351,"data":945,"type":343,"tunes":953},{"meta":946,"items":947,"style":342},{},[948,949,950,951,952],"Your product and application server.","Function-tool implementations and business logic.","Authorization and policy decisions around your own systems.","The execution environment lifecycle when you choose self-hosted compute.","Evaluation, acceptance criteria, domain-specific guardrails, and the decision of what the agent is allowed to do.",{},{"id":362,"data":955,"type":42,"tunes":957},{"text":956,"level":219},"2. Responses API: own the loop, use the platform primitives",{},{"id":367,"data":959,"type":226,"tunes":961},{"text":960},"The Responses API is the lower-level choice when you want OpenAI model and tool capabilities without delegating the overall agent runtime. OpenAI describes Responses as the recommended API primitive for new projects and as an evolution of Chat Completions with built-in tools, multi-turn state options, multimodal input, and agentic tool use.",{},{"id":372,"data":963,"type":226,"tunes":965},{"text":964},"A Responses request can itself invoke tools, but your application remains responsible for the larger workflow when you build an agent around it. That means your code decides how to persist application state, when to continue, how to recover, how to coordinate specialists, how to compact long histories, and how to represent resumable work.",{},{"id":377,"data":967,"type":226,"tunes":969},{"text":968},"This is not inherently inferior. It is the right boundary when agent behaviour must be deeply embedded in existing application logic, when you need a custom state model, or when a managed harness would hide control you actually need.",{},{"id":382,"data":971,"type":42,"tunes":973},{"text":972,"level":219},"3. Agents SDK: supported, but no longer the default forward path",{},{"id":387,"data":975,"type":226,"tunes":977},{"text":976},"The Agents SDK remains an open-source framework for running agent workflows in your application. It provides agent definitions, tools, handoffs, guardrails, tracing, sessions, and the runner loop in TypeScript and Python.",{},{"id":392,"data":979,"type":226,"tunes":981},{"text":980},"But its strategic status changed. OpenAI now labels the Agents SDK feature complete: maintenance, security fixes, critical bug fixes, and compatibility work continue, but major new features are not planned. OpenAI recommends the Agents API for new applications.",{},{"id":397,"data":983,"type":226,"tunes":985},{"text":984},"That does not mean an existing SDK application should be rewritten immediately. It means the architecture should stop assuming that the SDK is where the next major agent-runtime capabilities will land.",{},{"id":402,"data":987,"type":234,"tunes":990},{"body":988,"title":989,"variant":241},"OpenAI has not described the Agents SDK as unsupported. Existing applications can continue to use it, and maintenance, security, critical bug fixes, and compatibility work continue. The important change is that major new features are not planned and new applications are directed to the Agents API.","Do not translate “feature complete” into “deprecated”",{},{"id":408,"data":992,"type":42,"tunes":994},{"text":993,"level":219},"Where the Codex SDK fits",{},{"id":413,"data":996,"type":226,"tunes":998},{"text":997},"The current choice is not a simple three-way fork. OpenAI's runtime overview includes the Codex SDK as the option for running the Codex harness in infrastructure you operate. That is architecturally different from both the hosted Agents API and the Agents SDK.",{},{"id":418,"data":1000,"type":226,"tunes":1002},{"text":1001},"If your real requirement is “I want the Codex harness, but I need to operate it myself,” the Codex SDK is the surface to evaluate. If your requirement is “I want to own the loop around model calls,” evaluate Responses. If your requirement is “I already have a working Agents SDK application,” the existing SDK can remain valid while you plan against its maintenance status.",{},{"id":423,"data":1004,"type":42,"tunes":1006},{"text":1005,"level":219},"The Runtime Ownership Test",{},{"id":428,"data":1008,"type":226,"tunes":1010},{"text":1009},"A useful architecture decision starts by deciding what your team must own. Score each requirement as either must control, prefer to control, or prefer managed.",{},{"id":433,"data":1012,"type":476,"tunes":1043},{"rows":1013,"title":1035,"layout":304,"columns":1036},[1014,1017,1020,1023,1026,1029,1032],{"id":437,"label":1015,"values":1016},"Agent loop",[440,440,440],{"id":442,"label":1018,"values":1019},"Durable sessions",[440,440,440],{"id":446,"label":1021,"values":1022},"Harness runtime",[440,440,440],{"id":450,"label":1024,"values":1025},"Execution environment",[440,440,440],{"id":454,"label":1027,"values":1028},"Orchestration semantics",[440,440,440],{"id":458,"label":1030,"values":1031},"Provider \u002F transport flexibility",[440,440,440],{"id":462,"label":1033,"values":1034},"Operational burden",[440,440,440],"Runtime Ownership Test",[1037,1039,1041],{"id":468,"label":1038},"Decision",{"id":471,"label":1040},"If you prefer managed",{"id":474,"label":1042},"If you require control",{},{"id":479,"data":1045,"type":42,"tunes":1047},{"text":1046,"level":219},"A decision tree for new systems",{},{"id":484,"data":1049,"type":510,"tunes":1073},{"steps":1050,"title":1072,"orientation":509},[1051,1054,1057,1060,1063,1066,1069],{"label":1052,"description":1053},"1. Is this a new agent application?","If no, do not migrate solely because a newer surface exists. Evaluate the current application's actual constraints first.",{"label":1055,"description":1056},"2. Do you want a managed long-running agent harness?","If yes, start with the Agents API as OpenAI's recommended path for new agent applications.",{"label":1058,"description":1059},"3. Do you need the Codex harness but must operate it yourself?","Evaluate the Codex SDK rather than rebuilding harness behaviour on top of the Agents SDK.",{"label":1061,"description":1062},"4. Do you need to own the agent loop and state model?","Use the Responses API as the lower-level platform primitive and build the loop around it.",{"label":1064,"description":1065},"5. Are you already on the Agents SDK?","Continue if it meets requirements; major new features are not planned, so treat future platform adoption as an explicit roadmap decision.",{"label":1067,"description":1068},"6. Is a required capability missing from the Agents API?","OpenAI explicitly allows the Agents SDK as a short-term option for new applications that need unsupported capabilities.",{"label":1070,"description":1071},"7. Validate with a production-shaped spike","Test tools, environment, approvals, latency, observability, failure recovery, and lifecycle before locking the architecture.","Choose the runtime by the control boundary",{},{"id":513,"data":1075,"type":42,"tunes":1077},{"text":1076,"level":219},"What should not drive the decision",{},{"id":518,"data":1079,"type":304,"tunes":1105},{"content":1080,"stretched":43,"withHeadings":14},[1081,1085,1089,1093,1097,1101],[1082,1083,1084],"Weak decision rule","Why it fails","Better question",[1086,1087,1088],"“The newest API must be best.”","Newer can be strategically preferred while still lacking a capability you need.","Which runtime responsibilities should be managed versus application-owned?",[1090,1091,1092],"“We already know the SDK.”","Team familiarity can preserve an architecture whose roadmap has changed.","What is the cost of staying versus moving over the next product cycle?",[1094,1095,1096],"“Managed means no infrastructure.”","Agents API can still use self-hosted environments and your application still owns product logic.","Which infrastructure layer is actually being delegated?",[1098,1099,1100],"“Responses is only for simple calls.”","Responses provides built-in tools and stateful primitives; it can be the foundation of custom agent loops.","Do we need the platform to own the harness, or only model\u002Ftool primitives?",[1102,1103,1104],"“Feature complete means we must migrate now.”","The SDK remains maintained for existing applications.","What concrete future requirement is blocked by staying?",{},{"id":547,"data":1107,"type":42,"tunes":1109},{"text":1108,"level":219},"Migration is an architectural change, not an import rename",{},{"id":552,"data":1111,"type":226,"tunes":1113},{"text":1112},"Moving from the Agents SDK to the Agents API changes ownership. In the SDK, the loop runs in your application. In the Agents API, OpenAI runs the harness and session while your application integrates through tasks, events, tools, and environment boundaries.",{},{"id":557,"data":1115,"type":226,"tunes":1117},{"text":1116},"A real migration plan therefore needs to map session state, custom orchestration, handoffs, tool execution, approvals, storage, tracing, retries, failure recovery, environment lifecycle, and any provider-specific abstractions. Code volume may fall while operational assumptions change.",{},{"id":562,"data":1119,"type":42,"tunes":1121},{"text":1120,"level":218},"The migration inventory",{},{"id":567,"data":1123,"type":343,"tunes":1136},{"meta":1124,"items":1125,"style":342},{},[1126,1127,1128,1129,1130,1131,1132,1133,1134,1135],"Agent definitions and instruction ownership.","Tool definitions and where each tool executes.","Handoffs, manager\u002Fspecialist patterns, and subagent behaviour.","Session identifiers, conversation state, resumability, and history retention.","Human approvals and interruption semantics.","Custom context trimming or compaction logic.","Tracing, evals, observability, and production debugging.","Self-hosted files, containers, private-network access, or other execution dependencies.","Provider abstraction or non-OpenAI model dependencies.","Retry, timeout, idempotency, recovery, and lifecycle assumptions.",{},{"id":583,"data":1138,"type":42,"tunes":1140},{"text":1139,"level":219},"Public beta changes the risk model",{},{"id":588,"data":1142,"type":226,"tunes":1144},{"text":1143},"The Agents API is the recommended direction for new agent applications, but it is also in public beta. Those facts are not contradictory. Strategic direction answers “where is the platform going?” Beta status answers “how much interface and operational change should I budget for?”",{},{"id":593,"data":1146,"type":226,"tunes":1148},{"text":1147},"For production systems, isolate the integration behind an application boundary. Keep domain state, permissions, audit data, and business rules outside vendor-specific session objects where possible. This makes it easier to absorb API evolution without turning the agent runtime into the source of truth for your entire product.",{},{"id":598,"data":1150,"type":42,"tunes":1152},{"text":1151,"level":219},"A practical default architecture",{},{"id":603,"data":1154,"type":226,"tunes":1156},{"text":1155},"For many new OpenAI-native applications, a reasonable 2026 default is: Agents API for the managed harness and durable session, application-owned domain services and authorization, explicit function tools for business actions, and either OpenAI-hosted or self-hosted execution depending on data and compute requirements.",{},{"id":608,"data":1158,"type":226,"tunes":1160},{"text":1159},"That keeps the agent runtime powerful without making it the owner of business truth. The application still decides what a user may do, which data is authoritative, which actions require approval, and how results are validated.",{},{"id":613,"data":1162,"type":619,"tunes":1167},{"url":1163,"title":1164,"excerpt":1165,"ctaLabel":1166},"https:\u002F\u002Fstajic.de\u002Fblog\u002Fai-agent-reliability-why-the-final-answer-is-not-enough","AI Agent Reliability: Why the Final Answer Is Not Enough","A managed harness reduces operational burden, but reliability still requires traceable execution, explicit acceptance criteria and evaluation beyond the final answer.","Read the reliability article",{},{"id":622,"data":1169,"type":42,"tunes":1171},{"text":1170,"level":219},"What would change this answer?",{},{"id":627,"data":1173,"type":226,"tunes":1175},{"text":1174},"The recommendation changes if the Agents API adds or removes capabilities, exits beta with different contracts, changes environment or pricing boundaries, or introduces migration tooling that reduces ownership differences. It also changes if your application depends on provider portability, custom orchestration semantics, local-only execution, or a capability the hosted harness cannot support.",{},{"id":632,"data":1177,"type":226,"tunes":1179},{"text":1178},"For an existing Agents SDK application, the answer also changes with migration cost. If the system is stable, well evaluated, and not blocked by the SDK's feature-complete status, immediate migration may create more risk than value. If the product roadmap depends on capabilities landing only in the Agents API, delaying migration can create a different kind of debt.",{},{"id":637,"data":1181,"type":42,"tunes":1183},{"text":1182,"level":219},"Limitations",{},{"id":642,"data":1185,"type":226,"tunes":1187},{"text":1186},"This comparison focuses on runtime ownership and OpenAI's stated platform direction. It does not benchmark latency, quality, or total cost for a specific workload. Those properties depend on model choice, tool usage, environment, task length, caching, sandbox usage, and application architecture.",{},{"id":647,"data":1189,"type":226,"tunes":1191},{"text":1190},"The Agents API is also new enough that production experience is still accumulating. A design should therefore be validated with representative workloads rather than chosen solely from product positioning.",{},{"id":652,"data":1193,"type":42,"tunes":1195},{"text":1194,"level":219},"Conclusion",{},{"id":657,"data":1197,"type":226,"tunes":1199},{"text":1198},"The 2026 OpenAI agent decision is fundamentally about runtime ownership. Agents API means OpenAI owns more of the harness and durable-session machinery. Responses means your application owns the loop around platform primitives. The Agents SDK remains valid for existing systems but is no longer the default destination for major new agent-runtime features.",{},{"id":662,"data":1201,"type":226,"tunes":1203},{"text":1202},"For a new application, follow the platform direction unless a real requirement pushes you lower in the stack. Start with the Agents API, move to Responses when you need to own the loop, evaluate Codex SDK when you need the harness in your infrastructure, and keep the Agents SDK where existing investments or temporary capability gaps justify it.",{},{"id":667,"data":1205,"type":42,"tunes":1207},{"text":1206,"level":219},"FAQ",{},{"id":672,"data":1209,"type":672,"tunes":1230},{"items":1210,"title":1229},[1211,1214,1217,1220,1223,1226],{"id":676,"answer":1212,"question":1213},"OpenAI currently recommends the Agents API for new agent applications. The Agents SDK is feature complete and remains supported for existing applications, with maintenance, security, critical bug fixes and compatibility work continuing.","Should I use the OpenAI Agents API or Agents SDK for a new project?",{"id":680,"answer":1215,"question":1216},"OpenAI describes it as feature complete, not as unsupported. Major new features are not planned, but maintenance, security fixes, critical bug fixes and compatibility work continue.","Is the Agents SDK deprecated?",{"id":684,"answer":1218,"question":1219},"Use Responses when your application should own the agent loop, state strategy, orchestration and continuation logic while still using OpenAI models and platform tools.","When should I use the Responses API instead of the Agents API?",{"id":688,"answer":1221,"question":1222},"No. The Agents API can use no execution environment, an OpenAI-hosted environment, or a self-hosted environment connected to the managed harness.","Does the Agents API require OpenAI-hosted compute?",{"id":692,"answer":1224,"question":1225},"OpenAI positions the Codex SDK for running the Codex harness in infrastructure you operate. It is the relevant option when you want the harness but not the hosted Agents API runtime.","Where does the Codex SDK fit?",{"id":696,"answer":1227,"question":1228},"Not automatically. Evaluate whether the current system is blocked by the SDK's feature-complete status, whether required capabilities exist in the Agents API, and whether the migration benefit exceeds the operational and architectural change.","Should an existing Agents SDK app migrate immediately?","OpenAI agent runtime choices in 2026",{},{"id":702,"data":1232,"type":42,"tunes":1234},{"text":1233,"level":219},"Glossary",{},{"id":707,"data":1236,"type":707,"tunes":1253},{"title":1237,"entries":1238},"Key runtime terms",[1239,1242,1244,1246,1248,1250],{"term":1240,"anchor":446,"definition":1241},"Harness","The runtime loop and supporting machinery that coordinates model calls, tools, context, sessions and agent execution.",{"term":281,"anchor":716,"definition":1243},"OpenAI's managed API for durable cloud agents using a hosted Codex harness.",{"term":287,"anchor":720,"definition":1245},"OpenAI's lower-level API primitive for model responses, hosted tools and stateful interactions around which applications can build their own agent loop.",{"term":293,"anchor":723,"definition":1247},"OpenAI's open-source framework for running agent workflows in application code; feature complete as of September 2026.",{"term":299,"anchor":726,"definition":1249},"The runtime option OpenAI lists for running the Codex harness in infrastructure you operate.",{"term":1251,"anchor":730,"definition":1252},"Runtime ownership","The architectural boundary describing which parts of the agent loop, session state, execution environment and lifecycle are operated by the platform versus the application.",{},{"id":734,"data":1255,"type":42,"tunes":1257},{"text":1256,"level":219},"Primary sources and further reading",{},{"id":739,"data":1259,"type":746,"tunes":1264},{"link":741,"meta":1260},{"image":1261,"title":1262,"description":1263},{"url":440},"OpenAI — Introducing the Agents API","Launch announcement for the Agents API on 10 September 2026, describing the managed Codex harness and public beta.",{},{"id":749,"data":1266,"type":746,"tunes":1271},{"link":751,"meta":1267},{"image":1268,"title":1269,"description":1270},{"url":440},"OpenAI — Agents runtime overview","Current comparison of Agents API, Codex SDK and Responses API, including the support status of the Agents SDK.",{},{"id":758,"data":1273,"type":746,"tunes":1278},{"link":760,"meta":1274},{"image":1275,"title":1276,"description":1277},{"url":440},"OpenAI — Agents API overview","Documentation for durable cloud agents, sessions, orchestration, context compaction, recovery and environment choices.",{},{"id":767,"data":1280,"type":746,"tunes":1285},{"link":769,"meta":1281},{"image":1282,"title":1283,"description":1284},{"url":440},"OpenAI — Agents API architecture","Architecture boundary between the hosted harness, application server and no\u002Fopenai-hosted\u002Fself-hosted execution environments.",{},{"id":776,"data":1287,"type":746,"tunes":1291},{"link":778,"meta":1288},{"image":1289,"title":781,"description":1290},{"url":440},"Current Agents SDK support notice and explanation of the application-owned agent loop.",{},{"id":785,"data":1293,"type":746,"tunes":1298},{"link":787,"meta":1294},{"image":1295,"title":1296,"description":1297},{"url":440},"OpenAI — Migrate to the Responses API","Current Responses API positioning, built-in tools, stateful context and agentic primitives.",{},{"id":794,"data":1300,"type":220,"tunes":1301},{"title":842,"maxLevel":218,"minLevel":219},{},"2.31.6","OpenAI’s agent stack changed in September 2026. This architecture guide separates the Agents API, Agents SDK, Responses API, and Codex SDK by runtime ownership—so teams can choose the right control boundary instead of comparing product names.",{"lang":7,"title":208,"content":210,"contentJson":1305,"excerpt":799},{"time":212,"blocks":1306,"version":798},[1307,1310,1313,1316,1319,1322,1325,1328,1331,1334,1343,1346,1349,1352,1355,1358,1363,1366,1371,1374,1377,1380,1383,1386,1389,1392,1395,1398,1401,1404,1407,1410,1413,1435,1438,1449,1452,1462,1465,1468,1471,1474,1479,1482,1485,1488,1491,1494,1497,1500,1503,1506,1509,1512,1515,1518,1521,1524,1527,1530,1540,1543,1553,1556,1561,1566,1571,1576,1581,1586],{"id":215,"data":1308,"type":220,"tunes":1309},{"title":217,"maxLevel":218,"minLevel":219},{},{"id":223,"data":1311,"type":226,"tunes":1312},{"text":225},{},{"id":229,"data":1314,"type":234,"tunes":1315},{"body":231,"title":232,"variant":233},{},{"id":237,"data":1317,"type":234,"tunes":1318},{"body":239,"title":240,"variant":241},{},{"id":244,"data":1320,"type":234,"tunes":1321},{"body":246,"title":247,"variant":248},{},{"id":251,"data":1323,"type":42,"tunes":1324},{"text":253,"level":219},{},{"id":256,"data":1326,"type":226,"tunes":1327},{"text":258},{},{"id":261,"data":1329,"type":226,"tunes":1330},{"text":263},{},{"id":266,"data":1332,"type":42,"tunes":1333},{"text":268,"level":219},{},{"id":271,"data":1335,"type":304,"tunes":1342},{"content":1336,"stretched":43,"withHeadings":14},[1337,1338,1339,1340,1341],[275,276,277,278,279],[281,282,283,284,285],[287,288,289,290,291],[293,294,295,296,297],[299,300,301,302,303],{},{"id":307,"data":1344,"type":42,"tunes":1345},{"text":309,"level":219},{},{"id":312,"data":1347,"type":226,"tunes":1348},{"text":314},{},{"id":317,"data":1350,"type":226,"tunes":1351},{"text":319},{},{"id":322,"data":1353,"type":226,"tunes":1354},{"text":324},{},{"id":327,"data":1356,"type":42,"tunes":1357},{"text":329,"level":218},{},{"id":332,"data":1359,"type":343,"tunes":1362},{"meta":1360,"items":1361,"style":342},{},[336,337,338,339,340,341],{},{"id":346,"data":1364,"type":42,"tunes":1365},{"text":348,"level":218},{},{"id":351,"data":1367,"type":343,"tunes":1370},{"meta":1368,"items":1369,"style":342},{},[355,356,357,358,359],{},{"id":362,"data":1372,"type":42,"tunes":1373},{"text":364,"level":219},{},{"id":367,"data":1375,"type":226,"tunes":1376},{"text":369},{},{"id":372,"data":1378,"type":226,"tunes":1379},{"text":374},{},{"id":377,"data":1381,"type":226,"tunes":1382},{"text":379},{},{"id":382,"data":1384,"type":42,"tunes":1385},{"text":384,"level":219},{},{"id":387,"data":1387,"type":226,"tunes":1388},{"text":389},{},{"id":392,"data":1390,"type":226,"tunes":1391},{"text":394},{},{"id":397,"data":1393,"type":226,"tunes":1394},{"text":399},{},{"id":402,"data":1396,"type":234,"tunes":1397},{"body":404,"title":405,"variant":241},{},{"id":408,"data":1399,"type":42,"tunes":1400},{"text":410,"level":219},{},{"id":413,"data":1402,"type":226,"tunes":1403},{"text":415},{},{"id":418,"data":1405,"type":226,"tunes":1406},{"text":420},{},{"id":423,"data":1408,"type":42,"tunes":1409},{"text":425,"level":219},{},{"id":428,"data":1411,"type":226,"tunes":1412},{"text":430},{},{"id":433,"data":1414,"type":476,"tunes":1434},{"rows":1415,"title":465,"layout":304,"columns":1430},[1416,1418,1420,1422,1424,1426,1428],{"id":437,"label":438,"values":1417},[440,440,440],{"id":442,"label":443,"values":1419},[440,440,440],{"id":446,"label":447,"values":1421},[440,440,440],{"id":450,"label":451,"values":1423},[440,440,440],{"id":454,"label":455,"values":1425},[440,440,440],{"id":458,"label":459,"values":1427},[440,440,440],{"id":462,"label":463,"values":1429},[440,440,440],[1431,1432,1433],{"id":468,"label":469},{"id":471,"label":472},{"id":474,"label":475},{},{"id":479,"data":1436,"type":42,"tunes":1437},{"text":481,"level":219},{},{"id":484,"data":1439,"type":510,"tunes":1448},{"steps":1440,"title":508,"orientation":509},[1441,1442,1443,1444,1445,1446,1447],{"label":488,"description":489},{"label":491,"description":492},{"label":494,"description":495},{"label":497,"description":498},{"label":500,"description":501},{"label":503,"description":504},{"label":506,"description":507},{},{"id":513,"data":1450,"type":42,"tunes":1451},{"text":515,"level":219},{},{"id":518,"data":1453,"type":304,"tunes":1461},{"content":1454,"stretched":43,"withHeadings":14},[1455,1456,1457,1458,1459,1460],[522,523,524],[526,527,528],[530,531,532],[534,535,536],[538,539,540],[542,543,544],{},{"id":547,"data":1463,"type":42,"tunes":1464},{"text":549,"level":219},{},{"id":552,"data":1466,"type":226,"tunes":1467},{"text":554},{},{"id":557,"data":1469,"type":226,"tunes":1470},{"text":559},{},{"id":562,"data":1472,"type":42,"tunes":1473},{"text":564,"level":218},{},{"id":567,"data":1475,"type":343,"tunes":1478},{"meta":1476,"items":1477,"style":342},{},[571,572,573,574,575,576,577,578,579,580],{},{"id":583,"data":1480,"type":42,"tunes":1481},{"text":585,"level":219},{},{"id":588,"data":1483,"type":226,"tunes":1484},{"text":590},{},{"id":593,"data":1486,"type":226,"tunes":1487},{"text":595},{},{"id":598,"data":1489,"type":42,"tunes":1490},{"text":600,"level":219},{},{"id":603,"data":1492,"type":226,"tunes":1493},{"text":605},{},{"id":608,"data":1495,"type":226,"tunes":1496},{"text":610},{},{"id":613,"data":1498,"type":619,"tunes":1499},{"url":615,"title":616,"excerpt":617,"ctaLabel":618},{},{"id":622,"data":1501,"type":42,"tunes":1502},{"text":624,"level":219},{},{"id":627,"data":1504,"type":226,"tunes":1505},{"text":629},{},{"id":632,"data":1507,"type":226,"tunes":1508},{"text":634},{},{"id":637,"data":1510,"type":42,"tunes":1511},{"text":639,"level":219},{},{"id":642,"data":1513,"type":226,"tunes":1514},{"text":644},{},{"id":647,"data":1516,"type":226,"tunes":1517},{"text":649},{},{"id":652,"data":1519,"type":42,"tunes":1520},{"text":654,"level":219},{},{"id":657,"data":1522,"type":226,"tunes":1523},{"text":659},{},{"id":662,"data":1525,"type":226,"tunes":1526},{"text":664},{},{"id":667,"data":1528,"type":42,"tunes":1529},{"text":669,"level":219},{},{"id":672,"data":1531,"type":672,"tunes":1539},{"items":1532,"title":699},[1533,1534,1535,1536,1537,1538],{"id":676,"answer":677,"question":678},{"id":680,"answer":681,"question":682},{"id":684,"answer":685,"question":686},{"id":688,"answer":689,"question":690},{"id":692,"answer":693,"question":694},{"id":696,"answer":697,"question":698},{},{"id":702,"data":1541,"type":42,"tunes":1542},{"text":704,"level":219},{},{"id":707,"data":1544,"type":707,"tunes":1552},{"title":709,"entries":1545},[1546,1547,1548,1549,1550,1551],{"term":712,"anchor":446,"definition":713},{"term":715,"anchor":716,"definition":717},{"term":719,"anchor":720,"definition":721},{"term":293,"anchor":723,"definition":724},{"term":299,"anchor":726,"definition":727},{"term":729,"anchor":730,"definition":731},{},{"id":734,"data":1554,"type":42,"tunes":1555},{"text":736,"level":219},{},{"id":739,"data":1557,"type":746,"tunes":1560},{"link":741,"meta":1558},{"image":1559,"title":744,"description":745},{"url":440},{},{"id":749,"data":1562,"type":746,"tunes":1565},{"link":751,"meta":1563},{"image":1564,"title":754,"description":755},{"url":440},{},{"id":758,"data":1567,"type":746,"tunes":1570},{"link":760,"meta":1568},{"image":1569,"title":763,"description":764},{"url":440},{},{"id":767,"data":1572,"type":746,"tunes":1575},{"link":769,"meta":1573},{"image":1574,"title":772,"description":773},{"url":440},{},{"id":776,"data":1577,"type":746,"tunes":1580},{"link":778,"meta":1578},{"image":1579,"title":781,"description":782},{"url":440},{},{"id":785,"data":1582,"type":746,"tunes":1585},{"link":787,"meta":1583},{"image":1584,"title":790,"description":791},{"url":440},{},{"id":794,"data":1587,"type":220,"tunes":1588},{"title":796,"maxLevel":218,"minLevel":219},{},"Post erfolgreich abgerufen",{"items":1591,"source":1620,"manualIds":1621,"manualMatchedIds":1622},[1592,1599,1606,1613],{"id":1593,"slug":1594,"title":1595,"excerpt":1596,"featuredImage":1597,"publishedAt":1598},"382","a-practical-monorepo-architecture-next-js-platform-admin-fastify-api-prisma-and-nginx","Una Arquitectura Monorepo Práctica con Next.js, Fastify, Prisma y NGINX","Explora una arquitectura monorepo práctica utilizando Next.js, Fastify, Prisma y NGINX, destacando la integración y el flujo de trabajo en el mundo real.","\u002Fuploads\u002F2026\u002F01\u002Fa-practical-monorepo-architecture-next-js-platform-admin-fastify-api-prisma-and-nginx-1769885526116-q1100n.webp","2026-01-31T08:18:00.000Z",{"id":1600,"slug":1601,"title":1602,"excerpt":1603,"featuredImage":1604,"publishedAt":1605},"381","enterprise-grade-multi-tenant-architecture-for-an-international-platform","Arquitectura Multi-Inquilino de Grado Empresarial para una Plataforma Internacional","Loving Rocks es una plataforma de bodas de nivel empresarial diseñada con una verdadera arquitectura multiinquilino, bases de datos aisladas por inquilino e internacionalización integrada para escalabilidad global, seguridad y estabilidad operativa a largo plazo.","\u002Fuploads\u002F2026\u002F01\u002Fenterprise-grade-multi-tenant-architecture-for-an-international-platform-1769789121298-b6v7ak.webp","2026-01-30T12:04:00.000Z",{"id":1607,"slug":1608,"title":1609,"excerpt":1610,"featuredImage":1611,"publishedAt":1612},"478","what-is-rag-the-simplest-explanation-of-how-it-works","¿Qué es RAG? La explicación más sencilla de cómo funciona","RAG suena complicado, pero la idea es simple: antes de que una IA responda, primero busca información útil de una fuente de conocimiento y le da esa información al modelo de lenguaje. Esta guía explica RAG, los LLM, el estado, la memoria y las herramientas usando un modelo mental simple.","\u002Fuploads\u002F2026\u002F09\u002Fwhat-is-rag-the-simplest-explanation-of-how-it-works-1790377492124-khjagt.webp","2026-09-25T19:03:00.000Z",{"id":1614,"slug":1615,"title":1616,"excerpt":1617,"featuredImage":1618,"publishedAt":1619},"476","mcp-vs-a2a-vs-ucp-vs-ap2-vs-a2ui-the-agent-protocol-stack-explained","MCP vs A2A vs UCP vs AP2 vs A2UI: La pila de protocolos de agentes explicada","MCP, A2A, UCP, AP2 y A2UI a menudo se presentan como estándares de agentes competidores. En su mayoría, resuelven diferentes problemas de interoperabilidad. Esta guía mapea cada protocolo con el límite que realmente estandariza—y muestra cómo pueden funcionar juntos en un sistema de producción.","\u002Fuploads\u002F2026\u002F09\u002Fmcp-vs-a2a-vs-ucp-vs-ap2-vs-a2ui-the-agent-protocol-stack-explained-1790352625869-2ezle0.webp","2026-09-25T12:09:00.000Z","fallback",[],[]]