AI will produce more software engineers in the world, not fewer, if Patrick Collison's bet holds
Patrick Collison accepts that fewer engineers will be needed to do existing work. His prediction is that the set of work the industry attempts will expand fast enough to more than compensate. That is a specific, checkable claim with real stakes for everyone now entering the profession.
More software engineers in the world, not fewer. That is the call Patrick Collison, Stripe’s chief executive, is making about AI’s net effect on the profession. The claim is counterintuitive enough to deserve scrutiny, and specific enough that it will eventually be checkable against the numbers.
The intuition behind the displacement story is straightforward: if AI can write code faster and cheaper than a human engineer, you need fewer human engineers to produce the same output. Collison does not dispute that premise. He accepts that fewer engineers will be needed to do the things the industry is already doing. His argument is that the set of things the industry will attempt expands faster than per-unit headcount contracts.
That structure maps onto what economists call Jevons’ Paradox, the observation that making a resource cheaper tends to increase its total consumption rather than reduce it. When the cost of writing software falls, the number of problems it becomes economically rational to solve with software rises. Total demand for the underlying work can expand even as the skills that matter shift. The mechanism is not mysterious. It is the same one that turned spreadsheet software from a threat to bookkeepers into a tool that created more financial analysis roles than existed before. The analogy is imperfect, but the general dynamic has historical company.
I think there will be more software engineers in the world. we will need few of them fewer of them to do the things we're already doing, Patrick Collison
What the evidence cannot settle is the pace. The reallocation Collison describes, from engineers doing existing work to engineers attempting things that were previously out of reach, requires new categories of software to emerge fast enough to absorb workers displaced from current workflows. That is not guaranteed. Displacement within existing firms can move quickly. The formation of new markets tends to move slowly. Barriers to entry for building software are falling, and the scale of what can be built is rising, but the expansion depends on entrepreneurs identifying the new categories that cheaper software makes economically possible. Identifying them, capitalizing them, and hiring for them takes time that workers in transition may not have in abundance.
There is also a compositional question the prediction does not fully answer. Even if the total count of software engineers rises, the kind of engineer in demand may shift substantially. Writing boilerplate code and maintaining legacy systems are the tasks AI compresses most visibly. Designing new products, specifying what software should do, and verifying that it actually does it are harder to automate at the same rate. An expansion in total headcount that concentrates new hiring at the design and verification layer, while hollowing out the entry-level execution layer, would technically confirm Collison’s count while producing a very different profession from the one that exists today.
The stakes of Collison getting this right or wrong are not symmetric. If the expansion arrives as he expects, the workers and companies who positioned for volume and variety rather than pure efficiency will be well placed. If the displacement story turns out to be correct, and new categories do not emerge fast enough to reabsorb the workforce, the pain lands on engineers who trained for a labor market that no longer needed them at the same scale. Retraining pipelines are slow. Labor markets do not redistribute surplus workers across new industries on a smooth curve.
Collison has not set a deadline on the call, which limits how precisely it can be evaluated at any given moment. The honest framing is that the prediction operates over a medium horizon, long enough for new software categories to form but short enough that current entrants to the profession have a real stake in whether it holds. Engineers deciding now how to invest in their own skills, and companies deciding how to staff technical teams, are implicitly taking a position on the same question. His answer is that the profession grows. The mechanism he is counting on is demand expansion. The check is in the data that will accumulate over the next several years, and anyone entering the field today is part of the experiment.