I Finally Stopped Reviewing My AI's PRs I Finally Stopped Reviewing My AI's PRs

I Finally Stopped Reviewing My AI's PRs

Opus 5.5 is an incredible model. It’s the first AI I’ve trusted to merge PRs without reviewing the code.

Not for everything. But for my non-critical projects - most of my personal projects and some internal-only tools at work - I’ve stopped reading the diffs. And it’s been great.

That’s a big deal for me. I care a lot about code quality, and I was definitely not on the vibe coding train when it first came out. I used AI as more of a tool and less of a vibe, and I was still very critical of the code it wrote.

Why I Trust It

It understands what I mean. I can give it a half-baked request and it figures out my intention: what to do, and just as importantly, what not to do.

The code is flawless. Clean, readable, and it does what it’s supposed to. I’m not finding bugs after the fact.

It checks its work to the right level. It runs the tests, builds, and verifies the change actually works - without spending an hour testing a one-line typo fix (cough GPT 6.1 Sol cough).

No More Babysitting

Because it’s this good, my velocity has dramatically increased.

Before, using AI felt like managing a really fast intern. I’d review every line, catch the mistakes, explain what went wrong, and repeat. A lot of the time I saved writing code went right back into babysitting it.

Now that’s gone.

A New Workflow

This has completely changed how I work. Instead of breaking work into tiny, reviewable chunks, I can hand AI significantly larger tasks and trust it to get them right without putting a ton of attention on them myself.

It’s also the first AI I’ve used heavily in a cloud environment. I’ve always been big on running AI on my own PC, where I could watch everything it did. That made sense when I was reviewing every PR. Now that I’m not, I’m comfortable moving it into the cloud.

Not Production (Yet)

To be clear, this doesn’t apply to things shipping to big-time production. Like I said in The Dichotomy of Agentic Coding, some systems are complex enough that you still need a human who deeply understands what’s going out the door.

But the line is moving.

A Big Moment

Earlier this year I wrote that either AI would get good enough that code review becomes unnecessary, or we’d build the infrastructure to ship AI code without it. For my non-critical projects, the first one just happened.

It’s exciting to see how far AI has come. I think Opus 5.5 is a pretty big moment in its evolution.


← Back to blog