As I See It: Who tests the code the AI wrote?

There's a giddy pleasure in watching code appear faster than you can type it.  

A developer describes what they want, an LLM spins it up (or, to use one of Claude's own coinages, "wibbles" it), and a feature that might have taken days is working before lunch.

Whole teams have reorganised themselves around that speed, and the gain is real: the code arrives in a fraction of the time. But somewhere in the celebration a question goes unasked: once the machine has written it, who finds out whether it works?

 More code, same standard 

This is a present concern. By recent estimates, AI now writes or drafts close to half of all new code, and the figure climbs with every survey. That's a significant shift in how teams make software. The way we judge software stays the same: if it fails to do what it should, or breaks existing features, a line of code has no value, however quickly it arrived.

There's evidence that quality needs closer attention as AI-assisted coding scales.

The largest study of its kind to date, from the security firm Veracode, tested more than a hundred models across four programming languages and found that 45% of the code they produced introduced a security flaw. So the picture goes beyond a bigger volume of code to check.

AI-generated code deserves at least the scrutiny we give hand-written code.

 Why polished code gets less scrutiny

The bigger issue here is human rather than technical. We're inclined to trust a confident-looking answer, and code that arrives polished and instant feels finished in a way a colleague's half-commented draft never does.

A Stanford study found that developers using an AI assistant wrote less secure code than those working without one, while reporting greater confidence that their code was sound. The fluency that makes the tool a pleasure to use also lowers the scrutiny it needs.

The faster the code appears, the more finished it looks, and the less closely anyone tends to examine it.

Speed and accountability 

This is a case for testing alongside the tools. They let developers do what was previously out of reach, and they're particularly good at routine work like the boilerplate and scaffolding nobody enjoyed writing.

That's real time returned to people. What stays constant is the need to check the output. Hours of work may vanish, but healthy scepticism about the software should stay. Testers bring the same rigour and the same painstaking, systematic checks to AI-written code as to any other.

With the premium now on how quickly we can produce code, someone in all this acceleration still has to be accountable for checking it. 

By Simon Royall, Test Analyst at Mando Group