AI should help us produce better code - Agentic Engineering Patterns

Simon Willison argues that AI coding agents should be used to raise code quality rather than lower it. He frames the issue around technical debt: many common debt fixes are conceptually simple but time-consuming, making them ideal tasks for async coding agents like Gemini Jules, OpenAI Codex, or Claude Code. Agents can also support exploratory prototyping to validate technology choices cheaply. He closes by advocating for a 'compound engineering' loop where each project ends with a retrospective that improves future agent instructions, enabling teams to ship new features and improve quality simultaneously.