Code Reviews Will (and Have to) Die!

Reviewing code has become dramatically more expensive (relative to the cost of writing code). In the past, code reviews were relatively cheap (or maybe even paid for themselves in the long run). 1 or 2 days of coding roughly equaled 1 or 2 hours of code reviewing. But things have changed.

Agents Build Features in Minutes, Not Days

Agents produce the same features I build in 1 to 2 days in 15 to 30 minutes. Yet, (human) code reviewing time stays the same while its benefits diminish. While issues found and fixed during code review might have reduced future (expensive) coding work, now expensive code reviewing might save cents in coding costs down the line.

The Economics Do Not Work Out

What was eye-opening to me was when I flipped the roles: while at work, I still have a hard time accepting that skipping code reviews might be the way to go, when I let AI build software for myself, and I’m mainly the user of the software instead of the craftsmen writing it, I care about outcomes, not how well the code is written. I realized that I barely review the code produced when I let OpenCode build software that I personally use and where I care a lot more about what it does as opposed to how it does it.

Wer Zahlt, Schafft An

This is a German idiom equivalent to “He who pays the piper calls the tune.” If I put myself in an employer’s shoes, I’d have a hard time accepting that my developers spend 2 hours reviewing (already working!) code that an AI wrote in 30 minutes. And the better LLMs get at not only writing but also reviewing code themselves, the harder I find it to justify.