Simon Willison
Understand to participate
agentic
What happened
Highlighting Geoffrey Litt's talk at the AI Engineer World's Fair, this piece introduces the concept of "Understand to participate" in the context of coding agents. As agents generate large, complex codebases, human developers risk accumulating "cognitive debt" if they don't understand the underlying code, ultimately limiting their ability to creatively guide and collaborate with the agent.
Why it matters
Human-agent collaboration breaks down when the human loses the mental model of the code, creating a bottleneck for agentic software engineering.
The take
This is a crucial UX/interaction paradigm for coding agents. "One-click generation" is a trap; the real challenge is keeping the human in the loop with enough context to steer. Builders of coding agents need to design interfaces that explain *why* and *how* changes are made, not just *what* was changed.
Do this
Watch Geoffrey Litt's talk on YouTube (when released) and design your agentic workflows to actively maintain human comprehension rather than just outputting raw code.
Don't read this site daily. Get it in your inbox.
The daily brief and Sunday deep dive — distilled, scored, and opinionated. For builders only.