Two OpenAI employees who began managing teams at 24 sketch three ways AI might enter the firm: beneath people, beside people, and between people. They emphasize the essay reflects their own views, not OpenAI’s, and that they are interested in a transitioning economy rather than total automation. None of the scenarios will play out exactly, but aspects of all three are already happening.

The ‘beneath us’ scenario gives every individual contributor an AI workforce, effectively promoting everyone to manager. The agents start as eager interns and eventually form their own miniature org chart, with manager AIs and grunt AIs, accountants and lawyers, reporting up to a human board of directors. Adoption starts at the bottom because ground-level workers know the work; a department head unfamiliar with the methods cannot yet leverage the tools. The scenario is framed as automation without pink slips: AI takes over much of what a person used to do, while the person’s job expands to a larger objective. Entry-level drudgery disappears, along with some traditional apprenticeship routes, but junior employees gain resources and scope to experiment.

The ‘beside us’ scenario interleaves AI as colleagues reporting to existing managers. An AI marketing team can run an end-to-end campaign and proactively flag details—such as a sofa looking too blue—that a human VP would never have noticed. The authors describe an ‘escalator of scope’: automating a task, then a job, then a team, then a firm. Bigger scope is harder to build but easier to adopt, because labor learns the firm and onboards itself, whereas firms must learn new technologies and reshape around them. They also argue that using the ‘job’ as the AGI standard undervalues humans, since most output comes from coordination inside teams and firms, not lone workers.

The ‘between us’ scenario is an automation sandwich: AI sits above people and acts as an ever-present management consultant for executives. It reads Slack channels, documents, and code, investigates problems proactively, and can surface miscommunications before they escalate. The essay notes that managerial work is more cognitive and less entangled with atoms than ground-level work, so some managerial functions may automate before the work they oversee. This scenario centralizes power at the top, but the same intelligent monitoring could increase trust: a monitor that understands real progress, rather than clicks and status dots, might let firms offer more freedom, remote work, and contracting.

The final four questions are red tape, worker agency, firm size, and pay disparity. Cheap building should move firms from planning-heavy ceremonies toward prototyping and iteration—’aim less, fire more’—though the same technology can also enable more oversight. Workers are likely to gain agency on net, because AI gives everyone better advice and executives can rely on AI consultants to flag consequential exceptions. On firm size, they invoke Ronald Coase’s question of why production is organized inside firms rather than through contracts; AI coordination could support larger firms or make outside contracting cheap enough to shrink firms. The authors expect both many smaller firms and sprawling AI holding companies. Pay may compress if everyone receives the same good ideas, or diverge if judgment and scope amplify small differences; the outcome is left uncertain.