saved
Understanding is the new bottleneck
Hraness wrote this summary from a saved copy of the source. Quotations are taken word for word from the source.
gist
Geoffrey Litt argues that as agents write code faster than humans can absorb it, understanding remains necessary so people can participate in the next design loop, not only to verify correctness. He demonstrates explainer docs with quizzes, micro-worlds that make system behavior playable, and shared Notion spaces where teams build the same mental model.
ideas
- Understand to participate, not only to verify. Agents get better at checking their own work; human fluency still drives the next creative iteration across many loops.
- Cognitive debt is real. Skipping understanding works briefly, then limits what you can change without getting lost.
- Explainer docs beat raw diffs. Background, intuition before details, interactive figures, and literate diffs teach the change before line-by-line review.
- Quizzes regulate speed. Five questions after an explainer force a pause so the AI loop cannot outrun comprehension.
- Agents can build understanding tools. Debuggers, command centers, and shared agent plans help humans inhabit the system instead of rubber-stamping output.
quotes
“Hot take: I think it's still important to understand the code that our agents write!”
“That's where another answer comes in: we can understand to participate.”
“A quiz is a speed regulator.”
“The point here is that agents can write bits of code that help us humans understand other code.”