You might have heard about Pi (pi.dev), most probable you haven’t. What about Claude Code? Cursor? Codex? Antigravity? For sure the others ones. Now, why Pi?

Models have been progressing aggressively, getting better and better, but better models doesn’t mean better processes or code, automatically. We still have to deal with different things, shaping what we want with the tools we have, context windows, hallucinations, and much more. Recently Claude Code launched the feature of Agent Teams, in a nutshell, allowing you to spawn other Claude Code sessions with isolated context and protect us against the crazies hallucinations - a topic I want to deepdive in a little bit -.

In the hot topic of Claude Code, how many MCPs (Model Context Protocol) integrations you have? How are they treating you? Have you tried to run with dangerously skip permissions? Are you letting your model really work without babysit?

talking points:

  • building extensions
  • spawning agents
  • extensions better than mcps
  • controlling context
  • no permission requirements
  • skills
  • best agents to get the work done