02 25 2026
AI is everywhere, and everyone is expressing their opinions on the matter. Will AI take all our jobs, will it cure all diseases, will it create mirror bacteria and kill us all, what will happen to the labor economy, and on and on. Even Ben Affleck is weighing in on diminishing returns due to LLM scaling laws and the economics of capital expenditure on data centers (footnote: he is surprisingly well read on the topic).
Rather than dwell on the future I want to focus on how diffusion of current capabilities is happening in industry today. And in short, within the tech world there appears to be something of an ongoing mass hysteria about what current AI tooling enables.
Lest my comments be taken as some defense that AI will never replace software engineers, I'll preface with this: I know that these capabilities are already amazing, I agree that it has completely changed the nature of software work, and I fully expect capabilities to improve as the ongoing capacity buildouts continue.
That said, my critique is on the sentiment in the field which is centered around a belief that somehow speed of output of code was the bottleneck for creating innovation, increasing revenue, and building impactful products. If that were the case then why does innovation not scale with headcount already? Raw output has rarely been the bottleneck to innovation. It may even be the case that the actual act of typing code induces some cognitive process that helps with the creative process, as we already know occurs with writing.
Instead the misguided belief that velocity is innovation has been pushing firms to mandate AI usage and to mistake volume of output as a measure of productivity. The result is misplaced effort and incorrect incentive design for technologists. People are seeing that signaling "AI native" skills is getting rewarded. Thus we are reinforcing a push for pure velocity and quantity, producing as much code as possible by parallelizing your work: "gas towns", "Ralph Wiggum" loops, subagents, agent teams, etc. So far this doesn't appear to be producing much innovation beyond more ways to parallelize further code generation.
Many interesting things will happen this year as capabilities are bound to improve, but my hope is that the industry will temper expectations and begin to see more clearly how to leverage AI as a tool for high value work. Quantity is not quality, and we will have to learn to stop treating the two as equivalent. Hopefully the sooner we get there, the sooner we can put our focus on innovating rather than measuring success on the volume of work created.