The review bottleneck

Review is the bottleneck now

Agents write code faster than anyone can read it, so the pressure moves to review. Why more code means less understanding, and how to review at the speed agents write.

Writing code got fast. Reviewing it did not. A teammate opens an 800-line pull request written in three hours instead of three days, and someone has to approve it before lunch.

The pressure moved

When generation gets cheap, the expensive part is everything after it: understanding, deciding, owning. Teams feel this as more pull requests, bigger diffs, and approvals that mean less each week.

Faster reading will not fix it

You cannot read 800 lines faster. You can read fewer of them: the decisions, with their evidence beside them, and skip the consequences.

That only works if someone tells you where the decisions are. The person best placed to do that is the one who made them, which is now often the agent.

Let the author do the explaining

deck gives your agent a way to hand you its decisions instead of its diff. One claim per group, the lines lit beside each sentence, a chart when the claim is about load or timing. You review the argument in minutes and push back on the lines that matter. The agent answers with everything it knew when it wrote the code.

Review stays a human decision. It just stops being a reading contest.