source: simon willison: understand to participate
level: technical
geoffrey litt, speaking at the aie conference, introduced the concept of understand to participate. he addressed the growing challenge of working with coding agents that produce large, complex changes. as these agents take on more work, developers risk accumulating cognitive debt, where their mental model of the code drifts from reality. litt stressed that without deep understanding, a developer's ability to contribute creatively and fluently is meaningfully limited.
the core idea is that understanding is not just about reviewing code but about building a rich set of concepts in your mind. this mental fluency lets you think creatively about next steps and actively shape the project's direction. if you lack that fluency, you become a passive observer rather than a participant. litt's talk, which will be available on youtube, emphasizes that the goal is to stay engaged in the creative process, not just to verify the agent's output.
this framing has practical implications for ai-assisted development workflows. it suggests that teams should invest time in learning what the agent is doing, perhaps through techniques like having agents record video demos of their work. by prioritizing understanding, developers can avoid the trap of blindly trusting generated code and maintain the ability to steer projects effectively. the talk is part of a larger conversation about how to integrate powerful ai tools without losing human expertise.
why it matters: it highlights a key risk in ai-assisted coding: losing the deep understanding needed to guide and improve ai-generated solutions, which is critical for maintaining code quality and innovation.