Review intent, not code

For decades, code review was the primary quality gate in software development. One engineer writes code, another reviews it. This worked because humans wrote all the code, and human attention was the best way to verify human work.

Two things changed. Coding agents now write a significant and growing share of production code. And CI has evolved from simple test runners to sophisticated harness systems that combine deterministic checks with agent-based review. When agents write the code and agents review the code, the human in the loop on code review is no longer the highest-leverage use of their time.

Intent review is the alternative: reviewing the plan and direction before agents run, instead of reviewing the diff after. An engineer drafts a plan — often starting with a prototype built by an agent — and shares it with their team. The plan isn't a static document. It's a living artifact that includes code, prototypes, and the full conversation that shaped it. Teammates review the intent, leave comments, fork directions, and approve before full implementation begins.

The quality risk has moved from correctness to coherence. The danger isn't that any one agent writes bad code. It's that five agents write good code pointed in five different architectural directions. The cost of this technical debt in terms of lost future velocity and increased risk is the biggest concern for teams now. Deep focus on intent review is the best way to mitigate this risk.