AI is faster. It can dramatically reduce development time, automate repetitive tasks and take a huge amount of workload away from engineering teams.

But there’s a problem that doesn’t get talked about nearly enough:

AI can write code. It doesn’t necessarily know whether the code it has written is good.

I use the term “cringe-generated” for AI-generated code that looks impressive on the surface but falls apart when you examine what it’s actually doing.

I’ve seen code generated by AI that consumed significantly more CPU and resources than necessary simply because nobody had properly reviewed or optimised it.

The result?

End users started complaining that the platform was painfully slow.

Support tickets started flooding in.

And the error everyone kept seeing?

HTTP 500.

One client found themselves in an incredibly expensive situation. They ultimately had to spend an additional $310,000 bringing in specialist software engineering resource just to untangle and rebuild parts of an AI-generated codebase.

Meanwhile, productivity ground to a halt.

Sales dried up.

Refunds had to be issued.

That added another $140,000 to the bill.

$450,000 in additional cost.

Not because AI was inherently bad.

Not because the technology didn’t work.

But because nobody had the expertise or processes in place to properly validate, benchmark, secure and optimise the code AI was producing.

And that’s the part of the AI conversation I think businesses need to hear.

AI is a tool, not a replacement for engineering expertise.

Use AI to accelerate development.

Use it to automate.

Use it to prototype.

Use it to solve problems faster.

But then test the code. Profile it. Benchmark it. Review the architecture. Look at resource consumption. Stress-test it. Check what happens when things go wrong.

Because there’s a massive difference between:

“AI built it.”

and

“AI built it, and an experienced engineer proved that it works efficiently, securely and at scale.”

AI absolutely has a place in modern infrastructure and software development.

In fact, I’d argue that businesses not embracing it are going to fall behind.

But blindly deploying AI-generated infrastructure without knowing how to verify what’s underneath it?

That can become an incredibly expensive shortcut.

If you’re considering using AI to build, modernise or scale your infrastructure and want to make sure the efficiency gains don’t become technical debt, let’s talk.

todd.gilbey@bluecitycapital.com