The future of the firm

This essay is part of a series on a new OpenAI platform for independent voices, and the authors say it reflects their views, not the company's. That's worth keeping in mind. But it is a useful map of how AI could reorganize work, from people who already manage AI agents in their early twenties.

They sketch three ways AI enters the org chart: beneath us, as a basement of AI labor everyone manages; beside us, as AI teams reporting to existing managers; and between us, as an AI layer that watches the organization and advises executives. They argue all three are already happening in small ways.

The 'promoting' scenario is the hopeful one, and we think it deserves a fair hearing. If AI takes over most of the grunt work, a job doesn't have to shrink to the leftover sliver. It can become a different job aimed at something much larger. History offers some support: productivity gains have often been met by bigger ambitions rather than idleness. The authors are careful that this depends on people still adding value and on demand rising to meet supply.

The cost they name, and we'd underline, is the vanishing first rung. Apprenticeship, learning by grunt work, earning your stripes, those are how people learned judgment. The essay treats losing them mostly as a privilege. I'm less sure. Taste and judgment are often built by doing the boring thing badly for a while. If entry-level work disappears before we design a replacement, a generation could be handed management responsibilities without the experience to back them.

The 'layering' scenario is the one I'd flag for anyone who works for a living. An AI consultancy that reads every Slack channel and document and briefs the CEO is described here as a trust-builder that can reward unseen good work. It can be. The authors also admit the structural effect is centralization, more power at the top, and that the same system can enforce a boss's preferred way of working. That is workplace surveillance with a better vocabulary, and its effect depends entirely on who controls it.

Our read: this is a thoughtful essay with an optimistic bias it partly acknowledges. Its best insight is that AI that onboards itself like labor gets adopted faster than tools. That cuts both ways. The easier it is for an executive to deploy, the less say the people beneath them may have.

The authors draw on classic economics, including Ronald Coase's question of why firms exist, to ask whether AI makes firms larger or smaller. They expect both: many smaller firms, and sprawling AI 'holding companies' built around shared assets like factories or robot fleets.

They also ask whether pay gaps widen or compress, concluding it may hinge on whether people mostly provide judgment or follow it.