tutorial
Fable's judgement
For developers building AI workflows, trusting agents to self-optimize reduces prompt complexity and operational costs, enabling more scalable and autonomous systems.
What happened
Simon Willison shared a practical tip from a Fireside Chat with Cat Wu and Thariq Shihipar of the Claude Code team: when working with AI agents like Fable or Opus, it's often more effective to let them exercise their own judgment rather than prescribing explicit instructions. For example, instead of telling Fable to only run tests for large features, simply instruct it to decide when testing is appropriate. Willison also relayed advice from Jesse Vincent on conserving tokens by having Fable delegate smaller coding tasks to cheaper subagents. He applied this by prompting Claude Code to use its judgment to select a lower-power model for coding work, which Claude saved as a project memory file. The approach emphasizes trust in the AI's reasoning, reducing overhead from overly detailed prompts while also controlling costs as token prices rise.
Key takeaways
- Letting AI agents use their own judgment can be more efficient than micromanaging task rules.
- Fable and Opus perform better when given high-level goals instead of step-by-step instructions.
- Delegating small coding tasks to cheaper models saves tokens and reduces costs.
- Simon Willison configured Claude Code to automatically choose a lower-power model for coding subagents.
- The tip is especially relevant as token prices increase, making cost efficiency a priority.
Why it matters
For developers building AI workflows, trusting agents to self-optimize reduces prompt complexity and operational costs, enabling more scalable and autonomous systems.
This is an original editorial digest by AI Workflow Pro. Full reporting at the source:
Read the original on Simon WillisonMore AI news
All news →
Join the AI Workflow Pro Community