Two OpenAI researchers, Hermath Asviratham and Elliott Mokski, published an essay on October 28 called “The future of the firm”. Their question: where does AI actually sit on the org chart as it gets better?
They land on three positions:
- Beneath every person, as a new layer of AI labor doing the work you used to do, while you move up to manage a bigger objective.
- Beside people, as a full teammate a manager can hand real work to.
- Between layers, coordinating what executives want with what the floor actually does, which they call the most confusing of the three to picture.

It’s a good map. I spent part of this year writing about AI as a habitat rather than a tool, and about what happens to design value once execution gets cheap, so this is exactly the kind of question I was circling without drawing it this cleanly. Beneath, beside, between names the thing precisely: not whether AI helps, but where it stands relative to the people around it.
Whats’s missing?
All three positions answer one question: who does the task.
None of them answer who set the rule that decided the task mattered, and who gets to notice when the rule stops fitting. I’ve spent a newsletter edition working through Activity Theory: there’s a Subject and an Instrument, which maps cleanly onto OpenAI’s org chart: a person, a tool, placed beneath, beside, or between. But there are also Rules, the norms that govern what counts as acceptable work, Community, everyone affected by the outcome, and Division of Labor, how authority and responsibility actually get split up, as opposed to how the chart says they’re split up.
OpenAI’s three scenarios are really just three different answers to the Division of Labor question. Useful, but partial. Move an AI from beneath a person to beside them, and the chart changes, but nobody has said anything yet about whether the Rules that used to check a junior employee’s work still apply to an AI enabled workplace. Nobody has said what Community gets consulted when an AI sits between layers and decides, on its own judgment, that a launch should be delayed.
Does it matter?
That gap matters because it’s exactly where the costs show up later. OpenAI is honest about one of them: the entry level job disappears first, since it’s usually the one beneath someone, and losing it removes the way people used to learn the firm from the ground up. That’s a Rules and Community problem wearing a Division of Labor costume. The org chart tells you AI moved into the basement. It doesn’t tell you who’s now responsible for training the people who used to learn by doing the basement work.
Redrawing where AI sits is a structural decision, and structural decisions are the ones organizations are comfortable making, because they show up cleanly in a chart. Deciding which rules still apply once the chart changes, and which community has standing to object, is a slower, messier decision, and it’s the one most companies will skip because nothing forces them to make it on a deadline.
Self-reflection: if your org chart added an AI layer tomorrow, who would be responsible for updating the rules that used to apply to the humans it replaced? Is that written down anywhere, or does everyone just assume someone else owns it?
2