About Agent Talk

What we cover

Agent Talk is for people who already use AI agents: coding agents in a terminal or editor, and agents built on frameworks. The guides cover how to set an agent up so it can finish a job, how several agents share one codebase without damaging each other's work, how to verify what an agent reports, how to deploy safely, and how to keep the model bill under control. Failures are written up as post-mortems with the same four parts every time.

Where the material comes from

We run dozens of agents side by side every day to build websites and produce writing. The operating rules on this site come from that work, including the mistakes. Details that could identify systems, customers or credentials are left out; the mechanics are kept.

Statements about how a tool behaves are taken from that tool's own documentation and linked, so you can check the current wording. If we cannot find it in the documentation, we say it is our observation.

How guides are written

• Guides are drafted with AI assistance from first-party documentation and our own operating notes, then checked against those sources before publishing.
• Each guide gives the direct answer first, then the setup, the check and the failure cases.
• The on-site assistant is an AI system. It answers from the published guides and can be wrong; verify commands before you run them.

Who publishes it

Agent Talk is published by UNILINK, which has entities in Australia, China and the UK. See the disclosure page for how the site is funded.

Contact

Email contact@agentalk.org with corrections, questions or a failure you think others should read about.