← All Nuggets Jan 23-29, 2026

#1: Agents, Atlas, and exhaustion

Quick Thoughts

  • Claude Code is impressive (but expensive).
    • I finally played around with Claude Code this week. (Late to the party, I know!)
    • The friction: Even on the Pro plan, I hit rate limits within an hour. The disruption to my flow made it clear this isn’t quite ready for heavy individual consumer use yet, or perhaps I just haven’t calibrated my sense of token cost.
    • The insight: My workflow naturally shifted from reviewer to orchestrator. Instead of reading every line of diff, I found myself defining guardrails and setting a vision for the agent to execute. I was moderately successful, but it forced me to define “skills” in a much more structured way. There is so much more to explore here.
  • Tooling Shift:
    • Grok: I’m slowly switching to Grok for deep technical exploration. I’m finding its responses preferable for the “hard stuff.”
    • ChatGPT Study: Still my go-to for scaffolding new topics.
    • Gemini: Essential when my personal context (emails, docs) is part of the query.
  • One wish: I really want ChatGPT Atlas to succeed. The browser-agent concept is too good to fail.