The Roman legion was an information system.
It was built to move orders from a center to a frontier and reports back again, across distances that took weeks to cross. The mechanism was nested hierarchy: “consistent spans of control”, and at every level a named individual whose job was “to pass orders down and send information back up the hierarchy.”
It worked extraordinarily well, for a specific reason — a human being was the fastest available way to move a decision from one place to another.
The observation
The talk’s claim is that most companies today are still built this way: organized “like a Roman legion where human beings are the conduit for information flowing up and down.”
That is not a criticism of managers. It is a description of a structure that was correct under a constraint — and the constraint has changed. The org chart is a solution to a communication problem, and most of the communication problem it solves no longer exists.
The design has outlived its reason, which is the most ordinary way for a good design to become a bad one.
Why the productivity framing fails
Here is the part most organizations get wrong, and the talk names it exactly.
A year ago, the case for AI was productivity: “co-pilots, making engineers 20% more productive, adding co-pilots to workflows, shipping more software.” The talk calls this “a broken way of thinking” and borrows a sharper image for it — “we’re basically just like taking the old way of working and adding like a more powerful engine onto it.”
A faster engine in the same chassis produces a faster version of the same thing. If the structure exists to move information between people, and the reason it had to move between people has gone, then making each person faster leaves the structure — and its cost — entirely intact.
This is why so many AI programs report enthusiasm and no change. They were never aimed at the structure. They made the existing shape quicker, and the existing shape was the thing that needed to move.
What the alternative demands
The talk’s alternative is to “reimagine like what a company is and how it acts”, which is easy to say and worth being honest about.
It is harder than buying tools, and it is harder for reasons that are not technical:
- It touches the org chart, which means it touches people, and no vendor can do that on an organization’s behalf.
- It cannot be piloted in a corner. A pilot that changes nothing structural is a demonstration, and demonstrations are how organizations spend two years being enthusiastic.
- It requires knowing how the work actually runs — not the documented process, the real one, including the hidden hours and the handoffs nobody owns.
Which is why the honest version of this is not a strategy document. It starts by observing the work, because a structural change made against an inaccurate map is worse than no change.
The test to apply
Not did we deploy AI. That question can be answered yes by an organization where nothing has changed.
Ask instead whether the way the organization operates is different — whether work that used to consume people is now done by systems, whether the handoffs that existed only to move information still exist, and whether the people who were the conduit are doing something else now.
If the answer is that everyone has a faster tool and the shape is unchanged, then what was bought was an engine, and the chassis is still Roman.
Totomoko re-architects how organizations work so that AI does the job — inside the systems they already run, on an architecture built for portability: model routing and a data layer the organization owns. Built that way, swapping a vendor is a configuration change rather than a migration.