#software-design

Pol Alvarez Vecino (Answer.AI) applies Peter Naur's 'Programming as Theory Building' to LLM coding: the real program is the Theory held in engineers' minds, not the code, so LLMs can't reduce complexity that isn't measurable in the code itself. Illustrates with Answer.AI's own Stripe billing redesign for Solveit, where they chose a simpler 'Theory' (charge a saved card manually or automatically) over Stripe's more capable but more complex managed-subscription model, even when an LLM later suggested reverting to subscriptions to fix an edge case in India.

Hadley Wickham writes up a talk on his ambivalence about AI-assisted coding — the "software factory" dystopia of Steve Yegge's Gas Town versus using AI to write higher-quality code — and argues chat interfaces leave ~90% of an LLM's value on the table for programming, compared to giving it a harness to write, run, and iterate on code with a feedback loop of errors and docs.