> It's not like there was some fixed pool of work to be done and you just had to hire enough to exhaust the pool.

I'm my opinion you are failing to consider other bottlenecks, a la the theory of constraints.

An analogy: Imagine you have a widget factory that requires 3 machines, executed in sequence, to produce one widget.

Now imagine one of those machines gets 2x-5x more efficient. What will you do? Buy more of the faster machines? Of course not! Maybe you'll scale up by buying more of the slower machines (which are now your bottleneck) so they can match the output of the faster one, but that's only if you can acquire the raw material inputs fast enough to make use of them, and also that you can sell the output fast enough to not end up with a massive unsold inventory.

Bringing this back to software engineering: there are other processes in the software development lifecycle besides writing code -- namely gathering requirements, testing with users (getting feedback), and deployment / operations. And human coordination across these processes is hard, and hard to scale with agents.

These other aspects are much harder to scale (for now, at least) with agents. This is the core reason why agentic development will lead to fewer developers -- because you just don't need as many developers to deliver the same amount of development velocity.

The same logic explains (at least in part) why US companies don't simply continue hiring more and more outsourced developers. At a certain point, more raw development velocity isn't helpful because you're limited by other constraints.

On the other hand, agentic development DOES mean a boon to solo developers, who can MUCH more easily scale just themselves. It's much easier to coordinate between the product team, the development team, the ops team, and the customer support team when all the teams are in the same person's head.