AI NewsProductivityReported

Christian Ekrem argues reviewers who stop writing code get worse at it

Christian Ekrem argues that a developer who only reviews AI-written code gets worse at reading it over time.

AI News

Editorial2 min read

LinkedInX
Code review on a developer screen

Why it mattersReview capacity is now the slow step on many teams. Letting that skill decay is a hidden cost.

A team using AI to write most of its code will review more code than it writes. Christian Ekrem argued on Thursday that this makes the team's reviewers worse at reading code over time. His blog post reached 346 points on Hacker News. Ekrem is a software developer in Norway who writes on cekrem.github.io.

The argument is short and old. A musician who stops playing gets worse. A writer who only reads gets worse at writing. Ekrem asks why the same would not hold for code. His answer is that it does. "Simply 'reviewing all the code Claude produced' won't quite cut it," he writes in the post. He also warns that a developer who passes hard problems to an AI loses the taste for them.

Why 346 points matters more than the argument

The point itself is old. The traction is what makes it news. The post sits near the top of Hacker News for a day. The 346 points come from a readership that uses AI coding tools every day. Those readers read it as an uneasy agreement that the reviewer's judgement is thinning.

The post is an opinion from one developer, with an analogy to music and writing. Ekrem gives no numbers, cites no study, and names no team that has measured the effect. Treat the claim as that.

What a reviewer loses when all writing moves to the AI

A reviewer's job is to spot what the writer missed. They need to know what a well-written solution looks like in this language, on this codebase, on this deadline. That taste is built by writing. If an AI writes most of the code for a year, the reviewer only ever sees the AI's taste. The reviewer stops noticing a choice that was never made. They stop asking for a shape that was never written.

An eval suite catches a wrong output. A bad design shows up months later, when a change takes twice as long as it should. That is where the review skill matters, and where the AI tool is weakest today. A team that hands all writing to the agent needs a check on this. Count what the senior reviewer catches each quarter. Watch whether the number is falling. If it is, the writing skill is already fading. The fix is to put some writing back in the week. The post is a reminder to do that before the gap becomes a cost no one can name.

Source

Reported byChristian Ekrem

This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.

Share
LinkedInX
Start a project