Type: url
Original: https://simonwillison.net/2025/Sep/30/designing-agentic-loops/
Status: generated
Summary
A blog post naming “designing agentic loops” as a distinct skill for working with coding agents: reducing a problem to a clear goal plus a set of tools an agent can iterate against, so the agent can brute-force its way to a solution. It works through the safety question this raises (the post’s “YOLO mode”, where every command is approved by default, and the three ways to survive it: a sandbox, someone else’s computer, or accepting the risk), argues that shell commands usually beat MCP as the tools to expose, recommends issuing tightly scoped and budget-capped credentials, and identifies the problems worth this treatment as those with clear success criteria and tedious trial and error.