AI for thikning developers

Welcome! This is a quick collection of AI ideas and readings for developers. The field is moving fast—so treat everything here as a snapshot, not gospel.

Inspired by Melanie Mitchell’s “AI for Thinking Humans.”

Pages

MCP vs Terminal Calls – Feb 2026

Context, composability, and design tradeoffs for AI tool calls

Key insights: Why CLIs are more flexible and context-friendly than MCPs, practical advice for tool builders and developers, and why this space is evolving fast

Check back soon—new papers and ideas are always popping up. If you have thoughts or questions, I’d love to hear from you!

How I work with AI (as of Feb 2026)

Quick snapshot of my current workflow:

• VS Code Copilot is my daily driver (though I’m eyeing Claude becuase I like command line, and I miss nvim)
• Role prompts at the outset of a chat: Start each chat with “assume the role of…” — keeps it focused and cuts down hallucinations (tip from a Google training)
• Multiple chats: Different roles, different chats. I have a key combo to swap between them fast
• Small steps: Iterate, commit often. Makes it easy to roll back when things go sideways
• Stack your TODOs: Keep a running list of concerns the AI surfaces. Explore depth-first. (I’m tempted to build a shared TODO stack across chats)
• CLIs over MCP: More flexible, better context. I wrote about this
• Share the terminal: Let the AI see what’s happening, as well as interrupt the AI and jump in to run your commands. Context is everything