Let us imagine that Apple needs X engineer-days per year to produce a new model of an iPhone. Why is it so? Apple could easily get away with 0.5 X engineer-days per year if it was willing to release a new model once every other year rather than once every single year. Apple chooses to ship a new model every single year because it can, because it increases sales enough to make it worthwhile, and most importantly — because that’s what Samsung does. Apple spends X engineer-days per year and not 0.5 X not because X is what it absolutely takes to make any iPhone, but because this is what it takes to make an iPhone that will stay in the market in light of competition, which is also willing to pay its own X.

Once both Samsung and Apple have AI (assuming for simplicity that it’s the same AI with the same capabilities), they can both decide to cash out on the cost reduction, which is the equivalent of both deciding to release a phone every other year, or they can reinvest the saving into improving their edge against the other.

I am making an assumption that AI is a tide that lifts all boats to similar effect; that it plays no favorites. It is possible that one company makes more of AI than the other. While possible and eventually likely, this is not clearly the case at the moment. So far, players in industries that use AI for code generation use more or less the same tools, and even those who create their own tooling rely on a handful of common LLMs under the hood.

I believe that it is reasonable to expect that the higher efficiency of engineering work thanks to AI will end up being used for faster engineering and competitiveness, rather than put towards reducing costs. Competition may lead vendors to prefer directing the savings into producing yet more features we don’t need over, for example, sending people off, handing market dominance to those vendors who don’t.