Reported OpenAI Code Points to Managed Agents as Agent Builder Nears Shutdown
The unannounced service could move the agent conversation from designing workflows to controlling the environments where agents run.
Listen to this story
The audio brief
Story brief
3 key pointsOpenAI plans to retire its visual Agent Builder and Evals after November 30, 2026, directing developers toward the Agents SDK or Workspace Agents depending on use case. Reported unreleased code points to a separate operating layer for hosted or self-managed environments, skills, plugins, deployment controls, and multi-agent management—but OpenAI has not announced it, and the design could change or disappear. The...
- 01
Retirement takes effect after November 30, 2026; OpenAI recommends the Agents SDK for code and Workspace Agents for natural-language use cases.
- 02
Reported references include hosted and self-managed environments, configurable skills, plugins, deployment controls, and multi-agent management.
- 03
Workspace Agents already support scheduled tasks, connected applications, and administrator-defined permissions for Business, Enterprise, and Edu customers.
OpenAI has set a shutdown date for its visual Agent Builder, while reported code and unreleased product screens point to a possible service for running agents in configured environments. The contrast suggests a potential shift away from laying out workflow steps and toward the infrastructure that keeps agents working.
The visual builder now has an end date
OpenAI launched Agent Builder in October 2025 as part of AgentKit. The drag-and-drop tool was designed to compose multi-agent workflows, connect tools and add guardrails. OpenAI now says Agent Builder and Evals will shut down after November 30, 2026. It has directed code-based workflows to the Agents SDK and natural-language use cases to Workspace Agents in ChatGPT.
Signals from an unannounced operating layer
The reported code references hosted and self-managed agent environments, configurable skills, plugins and deployment controls, as well as options for managing multiple agents. The interface is not publicly accessible. OpenAI has not announced the product, and the features could change or be abandoned before release.
A different question from workflow design
OpenAI already offers Workspace Agents to Business, Enterprise and Edu customers. Those shared agents can run scheduled tasks, use connected applications and work under administrator-defined permissions. A service built around configurable environments would put more attention on the setup and deployment of agents, rather than the canvas used to arrange a workflow.
Anthropic already offers Claude Managed Agents in public beta as a hosted service with long-running sessions, scoped tool permissions, managed credentials and audit logs. The reported OpenAI references cover environments, skills and plugins, but do not confirm an equivalent set of controls or release terms. The practical question is whether a future service can make agents easier to run without taking away the deployment choices teams need.
DevDay is the next public checkpoint
OpenAI’s September 29 DevDay is the next scheduled event that could clarify whether it plans to disclose a new developer-facing agent service. Until then, the confirmed change is the retirement plan for the visual builder; the possible runtime layer remains an early signal, not a released product.
Loading discussion...
Reader comments
Newest comments first. Replies stay oldest first.