2026-08-22
Here is one mental model of a company: a company is an organization that is composed of lots of different people with different roles. Those roles roughly break down into a list of tasks or responsibilities. For example, companies have lawyers, and those lawyers review contracts, manage litigation, and coordinate with outside counsel. Some companies have cashiers, and those cashiers ring up customers and things like that. We can call this the “bottom-up” or “org-chart” model of a company — you take everyone in the company, you make a big list of the things that those people do, and that list is basically the company.
This is an intuitive model of what a company is, because we are individuals who do tasks for companies as part of our jobs, because we usually interact with companies (our own and others) through the people that are employed by them, and because when companies go about hiring people they generally have job postings with roles and salaries and responsibilities, etc. But this is also a fundamentally static view of organizations, and it does not explain why organizational roles or task composition changes over time. If we follow this model too far we will end up believing things that are not true.
Take as an example this article by Jaya Gupta: https://x.com/JayaGup10/status/2089924561441841152.
Some quotes:
“Frontier research is extraordinarily valuable and economically narrow. BLS data shows that life, physical and social-science occupations account for less than 1% of American employment. The United States employs roughly 2,000 mathematicians, 20,000 physicists and 37,000 computer and information research scientists. It employs millions of nurses, managers, administrators, logistics workers and customer-service representatives.”
“Imagine replacing every employee at American Airlines, Home Depot or Medtronic with a math olympiad who approached every task as a problem of discovery. Costs would not fall, but would rise rapidly and dramatically as every routine decision would be re-examined from first principles. The organization would drown in intelligence it cannot productively use or absorb.”
The takeaway is that we should “assume open-weight and specialized systems will keep eating the stack, and reserve frontier models only for the problems that still sit above the rising capability line.”
I guess as an initial matter my reaction is … what? If you had genius-level employees in all your positions you would see this as a bad thing? To me, this is a deficiency in management rather than some sort of iron law.
But more broadly, I think this is a fundamental flaw in the “bottom-up” view of companies. The process of automation in the bottom-up model is to figure out how much intelligence a given employee needs to complete their task-list, and then replace that employee with the cheapest model for that level of intelligence. In this model, companies consume roughly fixed levels of intelligence, and the task is to minimize labor spend for a given level of output.
I am super skeptical of this idea. Yes, we want to be at the efficient frontier of intelligence, and that may be an open model (although if you think you’re going to continue using Chinese open-weight models as a U.S. business indefinitely you may be in for an unpleasant surprise). But the current distribution of intelligence throughout the economy, even if it is efficient (wishing away the arbitrary preferences of people, switching costs between disciplines, etc.), is reflective of the current supply of intelligence! The whole economy is structured around intelligence as a scarce resource, and around fierce competition for small differences in intelligence. In a world where intelligence is much less scarce and the bearers of organization have very different properties, organizational structure looks completely different. It’s not very productive to think about whether AI can replace a particular job in its current form, holding the broader organization fixed. It’s much more productive to think about how you would approach the core task of the organization (the role(s) that actually generate revenue) with a clean slate.
Also, if you approach the problem thinking in this “bottom-up,” fixed-consumption way you end up thinking WAY too small. You could easily imagine these arguments being made in 1900. Why do we need secondary and post-secondary education? 40% of the population are farmworkers! They don’t need to be educated to bale hay! Imagine if those college-educated rabble-rousers with their big ideas were dropped into YOUR organization!!
Well, it turns out that increasing the level of intelligence in an organization can have huge, fundamental changes in organizational structure and in productivity. Work in the 21st century in developed countries looks very different from work in the 19th or early 20th century, and that was on the back of fairly small increases in intelligence (a few hard-fought points per decade). Unless you can identify some limiting principle for intelligence, you are looking down the barrel of much larger changes in a much shorter period. And even if you had average-human intelligence and not superintelligence (i.e., what you have now) you have huge, wildly-disruptive changes to the labor market and to organizational form because of the marginal economics.
If you are worried that your organization could “drown in intelligence it cannot productively use or absorb,” your highest priority should be figuring out how to use or absorb that intelligence, not trying to throttle it back. Otherwise, don’t get bent out of shape when an organization that can handle that intelligence eats your lunch.