Blog

Notes from building things: AI agents, LLM evaluation, data.

• 17 October 2026 · 8 min read

  Your AI assistant is picking the wrong tool, and nothing tells you

  AI assistants choose tools by reading one-line descriptions. I built mispick to measure which MCP tools a model confuses, and learned that AI-written fixes often make it worse.

  • MCP
  • LLM Evaluation
  • Open Source
  Read