Fable's judgement

According to Simon Willison, the Claude Code team shared a counterintuitive tip: let AI models use their own judgment rather than over-constraining their behavior. Instead of dictating 'only write tests for large features,' just ask the model to decide when testing makes sense. Willison extended this idea to model selection—delegating smaller coding tasks to subagents running lower-power models like Sonnet or Haiku while keeping judgment-heavy work with Fable. With Fable token prices rising in days, this strategy lets developers get more work done while burning through their allowance less quickly.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton