Simon Willison: let Fable pick its own cheaper subagent

Let the strong model use its own judgement about when to spawn cheaper subagents. That is the small workflow tip Simon Willison wrote up on his blog, pulling from a fireside chat he hosted with Cat Wu and Thariq Shihipar from the Claude Code team.

The example they gave was testing. You can spell out to Fable exactly when to write and run automated tests, and when to skip them for small copy or design changes. Or you can just tell Fable to decide for itself. Willison extends the idea with a related tip from Jesse Vincent, framed around the few days remaining before Fable prices go up: tell Fable to route coding tasks to lower power models in subagents, again using its own judgement about which model fits.

His actual prompt was 'For all coding tasks use your judgement to decide an appropriate lower power model and run that in a subagent.' Claude Code saved it as a persistent memory file with the reasoning that implementation rarely needs the top tier model while judgement, review, and synthesis stay with the main loop. Sonnet for substantive implementation, Haiku for mechanical edits, and Fable stays reserved for design, auditing, and anything judgement heavy.

Willison reports getting a lot of work done and watching his Fable allowance shrink less quickly than before. The honest caveat is that this is one person's experience over a few days, with a price change he does not quantify. What the post does not give you is any measurement of how well Fable actually judges which model fits which task, or whether the subagent output needs Fable to rework it. If the pattern holds up, it is a cheap trick for anyone feeling the sting of premium tier limits.