OpenAI Releases 34-Page Whitepaper on Building, Evaluating and Deploying AI Agents
mdancho84 · x · 2026-11-11
OpenAI has published a 34-page whitepaper on how it builds AI agents. The accompanying diagrams show the core architecture — an Input→Agent→Output pipeline where the agent stacks instructions, tools, guardrails and hooks.
A worked example of a refund agent uses gpt-4o-mini for both hallucination/relevance checks and safety classification, layered with the Moderation API and rules-based protections (input length limits, blacklists, regex) to catch prompt injections before handing off via the Agent SDK.
The paper covers three areas: building, evaluating and deploying agents; architectures, tool integration and scaling; and agent ops with evaluation frameworks.
Related event: OpenAI Releases 34-Page White Paper on Building AI Agents(2 posts)→
More from coding & agent
- Hermes agent can now call your phone: /call-me adds calls and voicemail — dean_rie · 2026-11-11
- Hermes looks stuck? Check the tool activity rows before you restart — dean_rie · 2026-11-11
- Zed makes it easy to import your Conductor threads into Delta as Conductor winds down — dean_rie · 2026-11-11
- Claude spawns 1,000+ subagents to build a full OS from an empty folder over 8 days — CodeByPoonam · 2026-11-11
- Google open-sources Harness², an agent that rewrites its own setup per task — Arindam_1729 · 2026-11-11
- The "company brain" folder structure: 26 markdown files to brief your AI agent — VibeMarketer_ · 2026-11-11