One of the stronger motives for introducing AI into engineering organizations is the foreseen reduction in development costs. I’m not aware of a single CFO who doesn’t drool over the anticipated headcount reduction that AI is expected to enable. Claude Code does such a great job writing code faster, so how can it possibly not result in less engineers on the payroll?

There are a few ways in which this logic can fail. By one way, the use of AI for engineering may result in more engineers employed in shepherding it. In another scenario, the cost of the compute resources needed to fuel AI may exceed the cost of many of the redundant engineers. But today I’d like to discuss yet a third way in which the OPEX reduction dream might end up not materializing. It boils down to our treatment of the amount of work required for producing a product as if it was a given constant, where in reality it is a constantly-varying result of a market equilibrium function.

Continue reading "How AI might not reduce engineering OPEX"