AI should help us produce better code

Simon Willison argues that shipping worse code with AI agents is a choice, not an inevitability — and that agentic coding actually enables a zero-tolerance attitude toward technical debt by making refactors (rename concepts, split large files, consolidate duplicate logic) cheap enough to just do. He also highlights using agents for exploratory prototyping to de-risk tech choices, and introduces Every's "Compound Engineering" loop where each project ends with a retrospective that improves future agent instructions. Practical and directly applicable to daily AI-assisted development workflows.