OpenAI's new tools for building agents

This OpenAI article is useful as a baseline for Responses API, built-in tools, tracing, evaluations, and how agent platforms are getting structured.

Source

New tools for building agents

Why I saved it

This is a good baseline read for the direction of agent APIs.

The useful part is not only “call a model.” It is the combination of model calls, built-in tools, tracing, and evaluations. Those are the pieces that make agent behavior easier to inspect and improve.

My notes

  • The Responses API is designed for combining models and tools.
  • Tracing matters because agents can fail across many small steps.
  • Evaluations matter because prompts and model choices should be measured.
  • Stored context and tool outputs need clear privacy and data rules.

What I want to remember

An agent product should include evaluation and tracing early. Otherwise every improvement becomes a feeling instead of evidence.