OpenAI launches Agents API in public beta
OpenAI has released the Agents API as a public beta, marking a shift from simple question-answering to autonomous task execution. Unlike the Responses API, which requires developers to manage the agent loop, or the Agents SDK, which helps build agents, the Agents API allows OpenAI to manage the infrastructure, including model calls, tool usage, and context management. This enables agents to…
What you can do with it
AI at Work →Run code or file edits with Agents API
What you get You can manage multi-step AI routines.
Runs code, edits files, and manages long tasks for you.
- Who for
- founders, operations and support teams
- Cost
- Price not stated
- Effort
- minutes
Use it for
- Manage multi-step AI routines
- Compress long conversation histories
- Run code or file edits
Watch out Data is stored only
Key points from the news
- OpenAI released the Agents API as a public beta on September 10, 2026.
- The API manages the agent harness, including tool usage and context, in the cloud.
- Agents can use workspaces to handle files, execute code, and delegate tasks to sub-agents.
The system supports sub-agents, allowing a main agent to delegate specific tasks to specialized agents, similar to a human project team. This approach is designed for complex workflows, such as IT onboarding, where an agent can check systems, create accounts, and report status without constant human instruction. The release emphasizes that while the API automates execution, humans must still define authority, approval boundaries, and responsibility for errors.
The article notes that this does not end the era of asking AI questions but changes the focus from crafting prompts to delegating work. It highlights the need for new management skills to oversee AI agents, ensuring they operate within defined limits. The API is currently in public beta, with specifications subject to change.
The era of asking ChatGPT questions might be coming to an end. What is OpenAI's "Agents API"?
note.com · 17 September 2026
Loading the full article…
This text was published by note.com and written by ゆるコンサ. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
Coverage and discussion
2sourcesThe headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Agents & Tools
All →- Perplexity details Brain memory system and Agent API testing methods · 1 src
- Google adds screen-operation AI to Gemini API for developers · 1 src
- Adobe introduces multimodal AI agents to automate tasks beyond chat · 1 src
- Microsoft and UC Berkeley release TokenEconomizer to cut coding agent costs by up to 30.9% · 3 src
- Descript runs live MCP builds with Claude, ChatGPT and Codex · 1 src
Comments
via GitHub Discussions