Your team ships twice the code with AI. Who's checking it?

Your team is shipping twice the code since everyone got a Copilot licence.

You have doubled the size of your quality problem, not halved it.

I have not seen a technology raise output and review capacity at the same time. AI coding assistants are the most extreme case of that pattern I have watched land on a team.

Deming argued it in Out of the Crisis (1982), and it has aged better than most CI pipelines: you cannot inspect quality into a product. Quality is a property of the system that produces the work, not something bolted on afterwards by looking harder. Feed that system twice the code and you get twice the defects, only faster and with better variable names.

It is like doubling the cars coming off the production line and asking the same bloke with a clipboard to sign them all off before lunch. (He will still take his lunch. He is entitled to it.)

Nobody scaled the part of the system that catches problems. They scaled the part that creates them.

What holds up:

• Review capacity gets planned as deliberately as sprint capacity, not left to whoever's got twenty spare minutes.
• PRs get smaller as generation gets faster, not bigger, because nobody can meaningfully review 800 AI-authored lines in one sitting.
• Someone still has to own why the code exists, not just confirm it compiles and passes the suite it also wrote.
• "The AI wrote the tests too" is treated as a flag to slow down, not a reason to trust the diff more.

AI did not create your quality crisis. It took the one you already had and put it on double speed.

Adopting AI in your delivery pipeline? See AI-Assisted Testing Implementation.