How to evaluate an AI agent before you let it run
A five-part evaluation method with a concrete test for permissions, approval and recovery.
Read more →The Construct / Field notes for operators
Sharp writing for teams turning AI into real work.
Practical guidance for understanding what a system can do, where its authority ends and how to check the result.
What MCP is and why it matters for AI agents
Understand the connection protocol and the controls an application still needs.
Read more →What trust boundaries mean
Map the identities, data and actions that cross the boundaries of an AI workflow.
Read more →