Tag:llm

All the articles with the tag "llm".

• A minimal agent loop and the four ways it breaks

  1 Nov, 2026(updated 9 Nov, 2026)

  I build a minimal agent loop in TypeScript and run it on a local model. It stops too early, crashes on a bad tool call, overflows its context and trusts what it reads. Each failure points to a part of the harness.