David Khourshid: Agents pick useState because training data is flooded with it, not because it's better

DavidKPiano · x · 2026-10-29

David Khourshid (DavidKPiano) argues AI coding agents choose useState not because it's "better" but because the training data is saturated with it. In the quoted context, he notes that architectural state mistakes are less common with reducers, since explicitly modeling all state transitions forces developers — or agents — to confront them. The takeaway: agent coding style preferences largely echo training data distribution rather than optimal engineering judgment.

Original post →

David Khourshid: Agents pick useState because training data is flooded with it, not because it's better

More from coding & agent