{"slug": "openai-s-code-points-to-managed-agents-as-agent-builder-winds-down", "title": "OpenAI's code points to managed agents as Agent Builder winds down", "summary": "OpenAI's codebase contains references to a managed agent service with configurable environments, skills, plugins, and deployment controls, according to TestingCatalog, ahead of OpenAI's DevDay on September 29th at Fort Mason in San Francisco. The service would provide a hosted runtime for long-running agents, potentially closing a product gap with Anthropic's Claude Managed Agents, as OpenAI winds down its Agent Builder after November 30th, 2026.", "body_md": "# OpenAI's code points to managed agents as Agent Builder winds down\n\n**TestingCatalog found references to hosted and self-managed environments, skills and plugins ahead of OpenAI's September 29th DevDay.**\n\n        By [Ryan Merket](/author/ryan-merket)\n        · Published \n\nPrimary source: [TestingCatalog](https://www.testingcatalog.com/openai-prepares-managed-agents-for-devday-2026/)\n\n## Why it matters\n\nA managed runtime would let OpenAI sell the infrastructure around long-running agents, where pricing, permissions and reliability matter as much as model quality. Connecting those agents to ads could also turn ChatGPT from a discovery channel into the place where commercial workflows are completed.\n\nThis time, it wasn't us: Alexey Shabanov at TestingCatalog found references in OpenAI's codebase to a managed agent service with configurable environments, skills, plugins and deployment controls, according to [his report published Sunday](https://www.testingcatalog.com/openai-prepares-managed-agents-for-devday-2026/). The findings point to an expansion of OpenAI's agent platform for developers and businesses at DevDay later this month.\n\nThe interface is not accessible, and the report is based on underlying code and unreleased product screens. TestingCatalog found options for creating and managing multiple agents and environments, using a developer plugin to configure agents, and deploying them in self-hosted environments. OpenAI could change or abandon those features before release.\n\nThe timing centers the reported service on [OpenAI DevDay](https://devday.openai.com/), scheduled for September 29th at Fort Mason in San Francisco. OpenAI says CEO Sam Altman will lead the 10 a.m. keynote, followed by technical sessions, product demonstrations and workshops. The opening keynote will be livestreamed.\n\n### OpenAI is replacing a year-old agent stack\n\nA managed agent service would give OpenAI a hosted runtime for long-running work rather than another visual tool for wiring individual workflow steps together. That distinction matters because OpenAI is already retiring the builder it introduced at its previous DevDay.\n\nOpenAI [launched Agent Builder in October 2025](https://openai.com/index/introducing-agentkit/) as part of AgentKit, describing it as a drag-and-drop canvas for composing multi-agent workflows, connecting tools and adding guardrails. On June 3rd, OpenAI updated the announcement to say Agent Builder and its Evals product would shut down after November 30th, 2026. OpenAI directed code-based workflows to its Agents SDK and natural-language use cases to Workspace Agents in ChatGPT.\n\nOpenAI's [Workspace Agents](https://openai.com/business/workspace-agents/) already let Business, Enterprise and Edu customers create shared agents that run scheduled tasks, use connected applications and act under administrator-defined permissions. The reported managed service appears to package a similar operating model for developers and deployment teams, including environments that can be configured outside the ChatGPT workspace interface.\n\nThat would close a product gap with Anthropic. [Claude Managed Agents](https://www.anthropic.com/news/finance-agents) runs custom agents as hosted services on Anthropic's platform, with long-running sessions, scoped tool permissions, managed credentials and audit logs. Anthropic offers the service in public beta for programmatic use.\n\nTestingCatalog reported that OpenAI's code references many of the same building blocks, including skills, plugins and distinct execution environments. Those similarities make pricing and operational controls more consequential than the basic feature list. Enterprises can already assemble agents from models, sandboxes and orchestration software. A managed service earns its margin by removing that infrastructure work while preserving enough control for security teams.\n\nOpenAI has not announced pricing or release terms for the reported product. Runtime limits, credential handling, auditability and the cost of keeping an agent active across a long task will determine whether developers use OpenAI's runtime or continue operating agents on cheaper infrastructure.\n\n### An agent could become the landing page\n\nTestingCatalog also found references to an advertising flow in which a ChatGPT ad links to an agent that guides the user through a commercial process. The report describes agents handling and personalizing conversion workflows rather than sending users to a conventional landing page.\n\nThat feature is more speculative than the managed-agent interface, and TestingCatalog said it could arrive after DevDay. OpenAI's public advertising roadmap points in the same direction without confirming an agent-led ad format.\n\nOn August 31st, OpenAI [said ChatGPT Ads had reached a $1 billion annualized revenue run rate](https://openai.com/index/expanding-access-to-ai-with-chatgpt-ads/) less than 200 days after launch. The figure extrapolates the platform's recent sales pace rather than reporting a completed year of advertising revenue. OpenAI also said tens of thousands of advertisers were using the platform and that it planned to explore more native ways for businesses to interact with consumers inside ChatGPT.\n\nLinking an ad to an agent would provide one such interaction. A retailer, software vendor or service provider could use the agent to collect requirements, compare options, answer questions and move a customer toward a purchase without immediately handing the session to an external website.\n\nFor OpenAI, that would connect two businesses it has built separately: enterprise agents that complete workflows and advertising that monetizes commercial intent. The managed runtime would supply the execution layer, while ChatGPT would supply distribution. DevDay will show whether OpenAI is ready to put those pieces into one product or is still testing them in code.", "url": "https://wpnews.pro/news/openai-s-code-points-to-managed-agents-as-agent-builder-winds-down", "canonical_source": "https://runtimewire.com/article/openai-managed-agents-devday-2026", "published_at": "2026-09-07 21:52:19+00:00", "updated_at": "2026-09-07 22:00:50.643424+00:00", "lang": "en", "topics": ["ai-agents", "ai-products", "ai-infrastructure"], "entities": ["OpenAI", "TestingCatalog", "Alexey Shabanov", "Sam Altman", "Agent Builder", "AgentKit", "Workspace Agents", "Anthropic"], "alternates": {"html": "https://wpnews.pro/news/openai-s-code-points-to-managed-agents-as-agent-builder-winds-down", "markdown": "https://wpnews.pro/news/openai-s-code-points-to-managed-agents-as-agent-builder-winds-down.md", "text": "https://wpnews.pro/news/openai-s-code-points-to-managed-agents-as-agent-builder-winds-down.txt", "jsonld": "https://wpnews.pro/news/openai-s-code-points-to-managed-agents-as-agent-builder-winds-down.jsonld"}}