Code Review as a Skill: For People and for AI

Code Review, Quality, Testing

Why Review Still Matters — and What Changes When Code Is AI-Generated

This article is part of my code review series. It focuses on a question that’s become more relevant with the rise of tools like GitHub Copilot, ChatGPT, and Vibe coding platforms like Cursor Pro:

If the AI can write the code, do we still need review?

Yes — absolutely.
But what you look for changes.

AI Code Is Not Magic — It’s Guesswork at Speed

You may have seen articles claiming that 90% of an app can be written by AI. I’ve seen some of them too — and I don’t buy it.

I did the research. Can it work? In theory — but only with mature, enterprise-level pipelines that include automated code analysis, tests, and a disciplined review process. Without those, the risk of silent regressions or subtle bugs grows quickly.

I tried this myself on a small project using an AI coding assistant (Cursor Pro). The results were mixed:

Some changes were genuinely clever. For example, when I asked it to add a facet filter to a list, it not only implemented the filter but also included a badge list of selected filters — very useful.
Other changes were subtly flawed. One filter, based on the time of day, recalculated on every row — which performed fine for a few items but broke down with larger lists.

This experience confirmed something important:

Whether code is written by AI or by a person, a second pair of eyes makes it better.
Review isn’t about distrust — it’s about improving clarity, reducing mistakes, and aligning expectations.

The Review Focus Changes — But the Need Doesn’t

Whether a person or an AI wrote the code, review is still critical. What shifts is what you’re reviewing for.

When humans write code, you often look for:

Structural issues
Logic errors
Code clarity
Naming and consistency
Adherence to standards

When AI writes code, you also have to consider:

Did it implement the right thing?
Did it change only what it should — or more?
Did it preserve existing behavior?

In my own tests, I’ve seen AI silently remove existing functionality when making unrelated changes on a component. That’s why tests are even more important when using AI: you need something to catch regressions.

The Skills That Matter Now

Using AI to generate code shifts how you spend your time:

Writing code takes less time — the AI handles the boilerplate and repetitive parts
Thinking through requirements takes just as long — vague prompts still produce vague results
Reviewing and testing take more time — because now you’re not just reading code, you’re verifying it behaves as intended

In fact, writing clear requirements and reviewing AI-generated output are becoming two of the most important development skills:

Clearly define what you want — ambiguity leads to unreliable results
Carefully check what you got — no matter how confident the output looks

The SmartBear Peer Code Review Best Practices note that review is cognitively demanding — it requires focus and can’t be sustained continuously.
That’s even more true when reviewing AI-generated code. You’re not just spotting bugs — you’re often reverse-engineering the AI’s intent and verifying that it matches what you asked for.

Testing Matters Even More

This also reinforces the importance of a solid set of tests.
Well-written (and scoped) tests help catch regressions and unexpected changes without adding to the reviewer’s cognitive load.

That said, be cautious when asking AI to fix tests.
This is one area where I’ve consistently seen it struggle:

It sometimes “fixes” a test by changing the assertion, not the code
Or it changes the code just to make the test pass, without validating the actual logic

In short: let tests support your review — but don’t let AI guess what correctness means.

Closing Thoughts

AI may help you write code faster, but it does not remove the need for review — it just shifts the emphasis.

Good code still needs to:

Do what it’s supposed to do
Avoid breaking existing functionality
Remain understandable to the next person (or to future you)

That’s what review is for — and why it’s still a critical skill in the AI era.