We run engineering in India, design and operations in Accra, and client work out of London. People assume this is a cost decision. It started as a capability decision and it has stayed one, and the thing that makes it work is not the geography — it is being honest about what actually has to happen in the same room.
Distance is not the problem people think it is
The received wisdom is that distributed teams lose something in the handoffs. That is true, but it misidentifies where. Timezone gaps are survivable; most work does not need a synchronous answer. What does not survive distance is ambiguity. A vague brief in one office becomes three different builds in three others, and you find out two weeks later.
So the discipline that makes this work has nothing to do with tooling. It is that anything crossing a boundary has to be written down to a standard that would survive the author being unavailable. Not a ticket that says "improve onboarding". A statement of what is true now, what should be true after, and how we will know.
That is a tax. It is also the thing that makes an AI-orchestrated team possible at all, because a model is an extremely literal colleague. If a brief is too vague for a competent operator in another timezone, it is far too vague for an agent.
Where the agents actually go
We do not put agents where judgement lives. We put them where volume lives.
The pattern that keeps recurring: a human decides the shape of the thing, agents do the wide part in the middle, a human checks the output against the original intent. First-pass drafting, triage, extraction, the tedious comparison of one document to another at a scale nobody would staff for. The team gets smaller in the middle and stays exactly the same size at the two ends where the thinking happens.
The failure mode is putting an agent at one of the ends. Every time we have seen it go wrong — ours or a client's — someone automated a decision rather than a task, and the error did not surface until it was expensive.
Lean is a consequence, not a goal
I want to be careful here, because "lean AI-orchestrated team" has become a way of saying "we replaced people", and that is not what this is.
Smaller teams are a consequence of removing the work that should never have been human work. The operators we keep are doing more of the thing they were hired for and less of the thing they tolerated. That is a better job, and it is also why they stay — which matters enormously in a market where good operators have options.
The group runs on this. Academy-trained operators go into Studio engagements alongside agents, and the ratio of what a small team can hold is the actual product. Not the models. The arrangement.