Understand to participate
Simon Willison reflects on a talk by Geoffrey Litt at AIE, who argued that when collaborating with coding agents, developers must understand the code deeply enough to remain active participants in the creative process, avoiding cognitive debt from drifting understanding.
Background
Simon Willison (creator of Datasette, prolific tech blogger) reports on a talk by Geoffrey Litt at the AI Engineer World's Fair (AIE) 2026. Litt researches tool-building and human-AI collaboration at the intersection of programming languages and HCI — he previously worked at MIT's Dynamicland and created the "llm-notes" concept for AI-assisted work. The core phrase "understand to participate" addresses a growing problem in AI-assisted coding: as LLM-based coding agents (like Claude with Artifacts, Cursor, or GitHub Copilot) write increasingly large chunks of code autonomously, developers risk "cognitive debt" — their mental model of the codebase diverges from what the agent actually produced, making it hard to steer the project intelligently later. Litt argues you need to keep your understanding deep enough to remain a true collaborator, not just an approving user.