How much time should you spend trying new AI tools versus getting better at the ones you already have? Developer Jill Cates argues the answer comes from a concept in reinforcement learning called the explore-exploit tradeoff, and she thinks most people get the balance wrong.
The framework is simple. Spend roughly 80% of your time exploiting - going deep with your current tools, building muscle memory, learning shortcuts, optimizing your workflow. Dedicate the other 20% to exploring - deliberately trying new tools and approaches on a set schedule.
The ratio isn't rigid. Newcomers to AI tools might start at 50/50 while figuring out what works. On a tight deadline, 95% exploitation makes sense. But the default should favor depth over breadth.
The Practical Playbook
Cates's specific recommendations cut against how most people actually behave:
- Schedule exploration time. Friday afternoons, for example. Don't let tool-hunting interrupt real work mid-project.
- Set a one-hour limit. If a new tool doesn't click within 60 minutes, move on. Most experiments will fail, and that's fine.
- Avoid reactive switching. Seeing a flashy demo on social media is not a reason to abandon your current setup mid-task.
The core argument: compounding returns come from exploitation, not exploration. Every hour you spend going deep on a committed tool builds on previous hours. Every hour bouncing between new tools starts from zero.
Cates uses Claude Code as her primary example of committed exploitation - a tool she went deep on and keeps getting more productive with over time. She also references Andrej Karpathy's December 2025 post about feeling "behind as a programmer" because of how fast AI tools are evolving. Her point: that feeling of falling behind is exactly the trap. Chasing every new release means mastering nothing.
This maps to what we consistently see with our readers. The people getting the most value from AI tools aren't the ones with accounts on fifteen platforms. They're the ones who picked two or three tools and learned them cold.