V.A. Graicunas computed it in 1933: a manager with six direct reports isn't managing six relationships — counting cross-relationships and group permutations, he's managing 222. Add a seventh report and it's 490. That geometric explosion is why span of control settled at five to nine people, and why every large organisation is a pyramid. The arithmetic left no choice.
The term that drops out
AI agents have no relationships to maintain. No morale to protect, no rivalries to referee, no 1:1s, no career conversations. Between six agents there are zero cross-relationships that need a manager's attention. Graicunas's geometric term — the thing that made wide spans impossible — drops out of the equation entirely.
So what limits how much one person can run?
The new constraints
Two things, and only two:
How many distinct briefs you can write well. Every delegated stream needs a definition of done, curated context, and constraints. Specification is the new supervision.
How much output you can verify. Delegated work returns, and someone has to judge it against the brief — the verification inversion guarantees this can't be skipped.
Span of control was limited by relationships. Span of specification is limited by briefs and review bandwidth — and both are learnable.
Why this changes careers, not just org charts
Both constraints compound in your favour. A brief, once written, is reusable — the tenth weekly competitor digest costs minutes to delegate, not hours. Verification gets faster as your definition-of-done discipline improves. A high-DQ operator quietly runs a span of ten, twenty, fifty concurrent workstreams — an org chart that classical management math says requires a department and three layers.
This is the precise mechanism behind the claim that output should stop correlating with hours. It's not hustle. It's that the 1933 equation no longer applies to you — if your briefs are good enough.