If You Use AI, Don't Compromise on Quality

I recently read an article about how AI is being used at work.

The gist of it:

• People submit AI-generated material as-is
• Managers reject it
• It gets sent back for revision, or reassigned to someone else
• This pattern is widespread enough that, in a Stanford × BetterUp study, the average cost is estimated at about $9M per year for a 10,000-person organization

The study is from Stanford University and BetterUp, based on a survey of U.S. workers.

Of course, social-science research isn’t physics — none of this is absolute. The right way to take it is probably, “yeah, that trend feels plausible.” Even so, it lands with real weight.

The Real Problem

The line that stuck with me was this:

“Polished output with no substance destroys trust.”

Using AI isn’t the problem.

The problem is submitting work when you:

• don’t understand it
• haven’t verified it
• aren’t taking responsibility for it

No one will literally say “you used AI here.” But it gets through. People pick up on it. “Ah, this person handed the whole thing to a model.”

And once that thought lands, everything you submit afterward is suspect.

It’s Worse for Engineers

For engineers, the risk is sharper.

Imagine code that:

• runs
• passes tests
• throws no errors

…but:

• misses the requirement
• has no separation of concerns
• duplicates the same logic across the codebase
• has no defensible reason for why it’s designed that way

In the AI era, code like this is going to multiply. The most dangerous case is the one that ends with “well, it works, so we’re fine.”

The Review Moment

Even worse is the review.

Someone asks: “Why is this logic shaped like this?”

And the honest answer is: “The AI wrote it. I don’t really know.”

That moment is brutal. Your credibility as an engineer drops in a single sentence.

How I Use Claude Code

I use Anthropic’s Claude Code heavily.

It’s genuinely fast — implementations that used to take days come together in one sitting.

But trusting it blindly is dangerous.

So I always come back to:

• review
• design check
• responsibility check
• log check
• test-coverage check

If I ran these on every single generation, I’d become the bottleneck myself. So I run them at a coarser grain and lower frequency — but I never skip them. Skipping is where the accidents come from.

The Real Takeaway

This isn’t a “don’t use AI” message.

The opposite — use it.

But you have to internalize one thing:

“You are the one responsible for what AI produced under your name.”

The engineers who get respected in this era won’t be the ones who avoid AI.

They’ll be the ones who use AI without lowering the quality bar.