Have Your AI Make Your AI Better

Let a different AI attack your work, then make the final call yourself.

Jeff Sauer
October 24, 2026

One AI writes something. A different AI tears it apart. They go back and forth until it's done. That's an adversarial review, and it's probably the biggest quality breakthrough I've had with AI lately.

When I mention this technique to people, they usually say: "What does that even mean?"

I picked it up from code reviews, where one engineer writes and another hunts for problems. It’s a staple in my current “vibe coding” sprints, so I use it all the time.

Here's how to do an adversarial review:

You create something with one AI
Then you bring in a different AI to say what's wrong with it
Instead of fixing it yourself, you have the first AI make the fix and resubmit for another review.

They work against each other in the same window, round after round, until it's ready to go.

Here’s my code review setup:

I'll build something in Claude Code using Fable 5.1
Then I run it against Opus 5.5, ChatGPT's Codex, and sometimes GLM, a Chinese model.
It does this over and over again, telling me "we're on round two, round three, round four."

One round might catch 12 errors, and the next round catches the new ones the fixes created.

The same thing works for writing. "Tell me what makes this sound like AI, then remove it." You can do this for every email, landing page, etc. Write it, then have AI go through it as a user and tell you what the user doesn't like. That's an adversarial review.

The cost is close to nothing. If you're already on a $20/month ChatGPT plan or higher, Claude Code can call Codex in for a review using that plan for no extra charge. It adds maybe 5 to 10 minutes to your workflow, and it saves you years of embarrassment.

One thing I'll add: with writing and marketing, I'm always the final adversarial reviewer. The AI assembles my thoughts, but at the end I make sure it sounds like me.

Adversarial reviews can work without a human in the loop, but they work better when a human closes that loop.

Try it on the next thing you create with AI. Let a different AI attack it, and then be the final adversary yourself.

Until next Saturday,
Jeff Sauer

P.S. If you want help bringing AI into your organization, or getting better at it yourself, I do coaching for exactly that. Just reply, and we can see if it's a good fit.

New On YouTube