CEO Brain Limit
I was starting to draft this post last week, based on some thoughts that are percolating and some conversations with my dad, but then oops someone I followed basically already said it smarter than I could’ve: https://www.mironov.com/ai-bottlenecks/
The TL;DR is that even if you take AI boosters at their word on the relative productivity increases to software engineering AI tools can provide1, that does not imply that there won’t be bottlenecks elsewhere in the system, and a key bottleneck being if you build everything you think you should build you will actually have a bad product. This is both because most of the ideas are actually bad, and because once a product gets too big it becomes unwieldy and customers don’t like it. So a hypothetical 15x increase in developer productivity will not and in fact should not lead to a 15x increase in throughput2.
Yeah unsurprisingly I agree with this wholeheartedly! The only other piece I want to add is there’s at least one other operational bottleneck on the surface area of a product: executives.
I’ve worked with a fair number of CEOs/other high-level executives, and while it’s totally true that you have to do some delegation in order to actually get anything meaningful done, surprises are usually bad. A CEO wants to know at least at a high level what are all the things she’s selling, and if she finds out that she’s selling something she didn’t know about, that’s usually not great even if the thing is good. At best case scenario you’ll get a “that’s really cool, but we need to figure out how to keep me in the loop next time.”
So one vision of the possible future is: despite the engineering productivity improvements, the average product surface area of a company stays about the same size because that’s how much executives are able/willing/comfortable with holding in their own head.
That said, there are other kinds of companies that already exceed human brain size. Sundar Pichai or Satya Nadella absolutely do not know the deep product details of everything at their companies (Tim Cook might; Apple is actually a comparatively small product surface area to the behemoths that are Google/Microsoft.) They do, however, eventually ladder down to leaders who own divisions or companies that are more traditionally company-sized in the product surface area, and my assumption is at some level in the hierarchy you can fairly expect an executive to have traditional-CEO-levels of knowledge about everything underneath them.
So the other vision of the possible future could be: we have more companies that take a conglomerate approach. Perhaps it’s now cheaper to run a “normal company sized product”, and so today’s “normal sized” companies become in fact conglomerates of different “company-sized-product” organizations.
That future does require leaders to be very comfortable with delegating and comfortable with them in the end having to take responsibility for and be the face of things they didn’t directly control. And some leaders can do that (again, Pichai! Nadella!) But it is definitely oppositional to the “Founder Mode” idea of a leader who is deep into every detail.
In the end, my guess is the same conclusion as Mironov, but from the organizational direction: where he’s saying that even with AI-assisted developer productivity increases, product surface area should stay roughly the same because there’s only so much surface area customers can take and because there’s only so many good ideas that actually benefit a product, I agree and also think product surface area stays roughly the same because at some point an executive is going to want to actually understand the entirety of what they own. Maybe that knows-everything executive stops being the CEO and drops down to a lower division-level person, or maybe it stays as CEO and companies stay at the same surface area they have now, or (probably most like) both.
But until we have AIs replacing CEOs and executives, you still have the bottleneck of how much can fit in one human’s brain!
Which I don’t think you should, or at least I don’t think you should beyond say 10-20% relative productivity improvement, based both on my experience and the fact that those numbers seem to keep percolating in most reputable studies. My assumption then is that everyone claiming greater improvements is either wrong, lying, or making (stated or unstated) assumptions about how “if it’s 10% now, it’ll obviously be 200% in the next couple years.” But maybe I’m wrong on that, who knows!
That does not necessarily bode well for us developers I guess lol