Review Is Becoming the Main Work. The Promotion Case Still Counts What Shipped.

Watch the video version:

At companies that adopted AI agents, more engineers now review code on top of writing it. A Harvard working paper by Fiona Chen and James Stratton followed 718 companies through March 2026; yesterday's piece used it to look at the review queue, and this finding is about the people standing in it. The paper estimates that AI agents raised the share of workers who review code in a given month from a baseline of about 29% to about 33%, with no significant change in the share who write it.

Review is becoming a primary activity in the AI era. In the paper's data, it is about a fifth of an engineer's week: the paper puts review at 18.6% of engineering time from its own data, and the engineers it surveyed put it at 21.8%. Agents are pulling more people into it, and the paper's theoretical model predicts the staffing mix shifts toward review, never toward coding, as AI raises coding productivity or the bug rate. A fifth of the week is already real work, and the promotion case has no artifact for it.

So think about the last calibration you sat in. Which engineer spent the quarter inside other people's code?

The standard advice goes to that engineer. Track your reviews, write them into your self-assessment, find a sponsor. If the ladder mentions review somewhere under technical leadership, the reasoning goes, a strong reviewer only has to make the case. That advice puts the work on the wrong person. It is not up to the individual contributor. The manager needs to be able to tie review work to business goals and revenue, especially now.

A calibration room runs on evidence, and authored work brings its own: a feature with a name on it, a launch, a commit history. The value of review is mostly an absence. It is the defect that never shipped, the design that never forked, the outage nobody had. A manager who walks in with artifacts argues with the artifacts they have, and the person who shipped more wins by default, not on merit.

The team reads the result. Review becomes the tax paid before the real work, and the strongest reviewers either move their hours back into authoring or keep holding the codebase together and stall. Either way, the reviews they stop doing usually land on the manager. That is the trap. The person whose job is to tie review to revenue is now doing the reviews, and the manager slides into Centralized Competence: their bandwidth becomes the single routing point for quality, and the next calibration arrives with less evidence than the last.

This is what I call the Rewards-and-Penalties Inversion: the systems that evaluate leadership performance actively reward the behaviors that damage teams. A manager who personally clears the review queue looks productive. A manager who builds the evidence for a reviewer's promotion produces nothing anyone counts until the calibration room. Last week's piece on rating AI usage found the same inversion one level down, in a rating that scores careful review as nothing. This time it sits on the leader's own scorecard, and the work it leaves uncounted is the work more of the team now does.

The managers who fix this are no better at advocacy than you are. They stopped treating review as a favor the reviewer does and started treating it as work they have to account for.

If your best reviewer lost the last calibration to the person who shipped more, the evidence that should have won it was yours to bring. Not theirs. Every review you quietly took back from them is an hour you spent not making that case. Treat review as owned, budgeted work: for every engineer who carries a real share of your team's reviews, keep a running record of what that review protected in business terms.

Name the engineer on your team who spent the most hours in review last quarter. Now write one sentence tying that review work to a business outcome: a release it cleared, a customer it protected, revenue that rode on a change they signed off. If you cannot write that sentence today, you will not be able to say it in the calibration room either, and they will lose to whoever shipped more. If you can write it, you already hold the evidence that room has been missing.

Quiet Confidence goes deeper into the Rewards-and-Penalties Inversion, and why the work that holds a team together never shows up on the scoreboard that rates it: https://TheQuietConfidenceBook.com

You are being graded on a second scorecard no dashboard shows, and it measures what you held back.