- Published on
LLMs got good enough that the hard problems shifted. I stopped writing about prompt engineering and started building autonomous coding agents. Here's what I learned moving from Claude Code to Pi, and why controlling an agent is harder than using one.