AI Founder Mode

The second claim

"I think AI is going to create the equivalent of an AI founder mode. Now you've got to be even more hands-on. In the era of AI there should be no pure people managers, because you're so close to the details, to the data, that everyone has the opportunity to be hands-on. It's hard to imagine only managing people and not agents."¹ Brian Chesky is not restating the case for a hands-on CEO when he says this. The case for the CEO is about one person's calendar; this is about a layer of the organization several rungs below him, and it says that layer is losing its reason to exist.

The doctrine being extended, leadership as presence rather than absence, installed one room at a time, is the practice Paul Graham named founder mode after watching Chesky run Airbnb, and it is set out in the file on founder mode as presence not absence. This file takes that argument as settled and asks only the narrower question: what changes when the work itself is increasingly done by agents.

What the coordination layer was for

The case for a manager who writes no code and touches no product was always a coordination case. Someone has to sequence work across people who cannot all talk to each other, resolve their contention for scarce colleagues, and carry information between queues. Chesky's claim dissolves the premise from underneath rather than arguing the manager is doing the job badly. An individual contributor directing agents is not waiting in anyone's queue, so most of what the pure manager was coordinating is now coordination the operator performs for themselves. What is left of the role is judgment about the work, and judgment about the work requires being in it.

The organizational shape this implies is not a flatter version of the same chart but a different unit: everyone a hybrid of individual contributor and manager, closer to a player-coach than to an overseer. That shift is treated at length in manage through the work and DRI / IC / player-coach model. Chesky's measure of how far it has already gone at his own company is that roughly 60 percent of Airbnb's code is now written by AI, which he places at twice the benchmark of competitors and peers.¹

The epistemic floor

The extension rests on a distinction Ryan Petersen states from a business with no AI story attached to it: execution can be handed off, the ground truth needed to decide well cannot. "You can't delegate understanding."² Read against Chesky's claim, that line explains why the pure manager is the role at risk and the individual contributor is not. Execution was always the delegable half, and agents make it cheap; a manager who delegated understanding along with it is left holding nothing that was ever theirs to hold. Petersen's own reversal, away from a bottoms-up culture and toward constant skip levels and a product roadmap he defines largely himself, is held in the file on you can't delegate understanding, and the collapse and recovery in customer satisfaction that bracketed it are in the file on owner-operator versus professional management.

Tensions and open questions

"No pure people managers" is asserted rather than demonstrated, and the evidence offered does not reach it: 60 percent of code written by AI measures how much of the work agents do, not whether a coordination layer thinned. Large organizations still need people whose full-time job is sequencing and arbitration across teams, which leaves the claim ambiguous between a strong reading, that the layer disappears, and a weaker one, that anyone occupying it must also be in the work. The claim also comes from an unusually design-driven company that is unusually far along on agent adoption, which is precisely the population that would see the effect first and the least safe population to generalize from while it is happening.