Most companies aren't getting the speed they hoped for from AI. The difference is engineering foundations - patterns for AI to extend, or gaps it fills by improvising.
It's no surprise to me that most companies aren't getting the speed they hoped for from AI.
If you have good engineering foundations - what kind of code goes where, how it's tested, how quickly you find out and can respond when something isn't behaving as expected - then AI has established patterns to work off and keep extending.
If they're missing or under-enforced, AI improvises a decision in every gap, at the speed it generates code - and all you get back is more code to review than the team can hold the line on. The most likely scenario is that more under-reviewed code gets into the code base, which further adds on to future problems.
The teams that will accelerate had something before AI existed: well established foundations set by experienced people that were enforceable. When a requirement lands that no existing pattern covers - that's where decisions get debated and made.
source & further reading
thetruthasiseeitnow.com — original article
Using "AI slop" as feedback tells as much about the commenter as the creator.
You can only get AI scale if you let AI do the reviewing too.
AI changes the economics of software rewrites