François Chollet, creator of the Keras deep learning library and ARC-AGI benchmark, describes directing AI coding agents by positioning verifiable constraints as barriers that guide blind-squirrel-like random exploration toward target outcomes.

Danielle Fong quoted the post and compared the method to guiding lemmings in the classic video game through environmental design rather than direct commands.

Original post

OPFC#37François Chollet|@FCHOLLET

A mental model for working with coding agents is that they're blind squirrels running into a maze and bumping into walls. You must place the walls (verifiable constraints) strategically so that they end up in the general region you want them in.

8:46 AM · May 18, 2026 View on X

ORIGINAL POSTFC#37François Chollet|@FCHOLLET

A mental model for working with coding agents is that they're blind squirrels running into a maze and bumping into walls. You must place the walls (verifiable constraints) strategically so that they end up in the general region you want them in.

3:46 PM · May 18, 2026 · 32.6K Views

REPLYND#1894Nick Dobos|@NICKADOBOS

@fchollet @DanielleFong I prefer surfing on the back of a dragon but squirrels work too

FCFrançois Chollet@fchollet

A mental model for working with coding agents is that they're blind squirrels running into a maze and bumping into walls. You must place the walls (verifiable constraints) strategically so that they end up in the general region you want them in.

3:46 PM · May 18, 2026 · 32.6K Views

10:50 PM · May 18, 2026 · 131 Views