AI Code Reviews: When Your Robot Colleague Is Smarter Than You

I need to tell you about the Workfront proofs thing, because it perfectly captures my relationship with AI.

At my job at StudioRx - an Omnicom agency, team of 30+ devs - we use Adobe Workfront for project management. Routing proofs through Workfront’s approval workflow is this notoriously painful manual process. Multiple people had tried to automate it. The consensus was that it couldn’t be done. The API didn’t support it, the documentation was incomplete, it simply wasn’t possible.

I spent an entire day with Claude trying anyway. Not because I thought everyone was wrong. Because I couldn’t stop. I kept rephrasing the problem, kept adjusting the approach, kept pushing. And at some point - after more attempts than I’d care to count at finding the right way to describe what I needed - it worked. We automated the proof routing through the approval workflow. It saves real manual hours that used to get eaten by hand.

That’s my relationship with AI in a nutshell: obsessive, relentless, and usually successful because I refuse to stop before everyone else would have.

The Rust Story (Or: How I Wrote a Language I’ve Never Learned)

In December 2024, I built an Electron app that included Rust binaries. I have never written Rust. I have never read a Rust tutorial. I don’t know what a borrow checker is (okay, I’ve been told, but I couldn’t explain it to you). ChatGPT wrote approximately 95% of the Rust code. It compiled. It ran. It shipped.

This isn’t a cute anecdote about AI being helpful. This is a fundamental shift in what’s possible. A self-taught developer with a GED and no CS degree built working software in a systems programming language he doesn’t know, in weeks instead of the months it would take to learn Rust properly.

If that doesn’t change how you think about code reviews, I don’t know what will.

AI Reviews Catch What You Won’t

Here’s what I’ve learned from using AI as my primary code review tool: it’s not “slightly better” at catching certain things. It’s embarrassingly better. The gap between what I catch on a manual review and what Claude catches is humbling.

Security vulnerabilities. I’ve been writing WordPress code for over a decade. I know about SQL injection, XSS, nonce verification, capability checks. And AI still catches sanitization gaps I miss. Not because I’m bad at security - because human attention is finite and AI’s isn’t.

Edge cases. The “what if this value is null” and “what if this array is empty” stuff. I handle the happy path and the obvious error cases. AI catches the weird ones - the race conditions, the type coercion gotchas, the off-by-one errors that only manifest with specific data.

WordPress-specific patterns. Improper hook priorities, missing wp_kses on output, using $wpdb->query() when $wpdb->prepare() should be mandatory. AI knows the WordPress coding standards better than any human because it’s read every page of the handbook and never forgets any of it.

How I Actually Use It

My workflow isn’t fancy. I don’t have some elaborate CI/CD pipeline with automated AI reviews (though that’s coming). I paste code into Claude. That’s it. That’s the workflow.

But how I prompt matters. “Review this code” gets you generic feedback. What actually works:

• “Review this WordPress plugin code for security vulnerabilities. Assume an attacker has access to subscriber-level accounts.”
• “This function handles user-submitted data. What sanitization am I missing?”
• “Find the performance bottleneck in this WP_Query loop. The page takes 4 seconds to load with 500 posts.”

Specific questions get specific answers. The AI is only as good as what you ask it, and most people don’t ask it enough.

The Part Nobody Talks About: AI Makes You Learn Faster

The best thing about AI code reviews isn’t catching bugs. It’s the education. Every time Claude explains why something is a vulnerability, I learn the pattern. Every time it suggests a more efficient query, I understand the optimization. It’s like having a senior developer who never gets tired of explaining things and never judges you for not knowing something.

I taught myself to code by reading WordPress source code line by line. No CS degree, no bootcamp, just a GED and a library card and an obsessive personality. AI has accelerated that self-teaching process by an order of magnitude. The things I’m learning now, in the way I’m learning them - through building, through review, through having my mistakes explained in context - it’s the fastest I’ve ever grown as a developer.

Where AI Reviews Fall Flat

AI doesn’t understand your project. It doesn’t know that the weird code pattern in your legacy module exists because the client’s payment processor has a bizarre API that requires it. It doesn’t know that your “inefficient” database query is intentionally denormalized because the reporting team needs the data in that shape.

Business context is the thing AI cannot review. Architecture decisions, trade-offs, “we did it this way because of X constraint” - that’s still entirely on you. AI reviews the code. You review the decisions.

It also hallucinates. Not often with code review (it’s better at analyzing existing code than generating new code), but it happens. I’ve had it flag “vulnerabilities” that weren’t real and suggest “optimizations” that would break functionality. You cannot turn off your brain. The AI is a tool, not an authority.

The Duality

Here’s the thing I wrestle with: AI is making me a dramatically better developer while simultaneously making the skills I’ve spent 16 years building less unique. The Rust story is a perfect example. I couldn’t write Rust before AI. Now I can ship Rust code. That’s incredible for me and terrifying for Rust developers who spent years learning the language properly.

I believe AI will eventually take my job. I’ve said this publicly and I mean it. And I also can’t stop using it, can’t stop pushing it further, can’t stop finding new ways to make it do things people say are impossible. The excitement and the existential dread coexist, and I’ve stopped pretending one cancels out the other.

Just Fu**ing Try It

If you’re not using AI for code reviews yet, start today. Not next week. Today. Paste your most recent pull request into Claude or ChatGPT and ask it to review for security and performance issues. I guarantee it will find something you missed.

And when it does, don’t feel bad about it. Feel grateful that you caught it before production. Then do it again on the next PR. And the next one. Make it a habit.

The developers who will thrive aren’t the ones who can write the most flawless code from memory. They’re the ones who use every available tool to ship better software faster. AI code review is the highest-leverage tool most developers aren’t using yet.

Your robot colleague is smarter than you at certain things. That’s not an insult. That’s an opportunity.