Simon Willison: 'Understand to Participate' — Cognitive Debt as the Core Risk of AI-Assisted Development

Main idea

Geoffrey Litt argued at the AIE (AI Engineer) conference that developers working with AI agents must maintain deep enough conceptual understanding of their code to remain 'active, creative participants' in the development process. Without this understanding, cognitive debt accumulates with each agent iteration.

Context

Litt delivered this talk at the AIE conference (300+ recorded talks being released over three weeks). Willison comments on it from his perspective as a long-time LLM tools practitioner and author of the llm library. Litt also published a Twitter thread version of the presentation.

Why it matters

This is one of the most practical frameworks for how to work with AI coding agents without losing control. The key question for every developer: when is understanding 'sufficient' and when are you accumulating debt that will manifest as slowed development or production bugs.

Details / arguments

• 'You need a rich set of concepts in your mind to think creatively and fluently about how to move something forward'
• Without understanding, code becomes a black box — the developer is just a prompter, not a creator
• Cognitive debt grows gradually with each unchecked agent iteration
• Both Litt and Willison agree: AI agents accelerate work, not replace understanding