I keep seeing people use AI to take minutes. I do it too. 

Meetings recorded, notes generated, reports drafted in a fraction of the time they used to take. Everyone’s pleased with themselves and I understand why: it’s the obvious use case, the safe one,  and it gets stuff done.

But what is that stuff that is getting done? Meeting notes and reports are the raw material of bureaucracy and centralisation. Where information travels up from the frontline to wherever decisions actually get made. We’ve just been handed a tool that could dissolve the reason the journey needs to happen at all, and we’re using it to make the journey faster.

We’ve got a technology capable of reshaping organisations, and we’re getting it to do our admin.  Worse still,  ‘AI workslop’ is a growing phenomenon. 

We’ve been here before. In the mid 2010s I watched the public sector fall in love with digital transformation and mostly waste it — not because the technology failed, but because we copied the wrong thing. We looked at Amazon and Google and took the last layer, the interface, the self-service portal: without understanding the layers underneath. We built cost-per-transaction graphics that treated every interaction with a complex human being as a unit to be minimised, and we called it efficiency. Digital transformation was a big ambition, assembled with borrowed logic that didn’t fit, and then applied to the wrong model. 

This time the danger is the opposite. Nobody’s building a grand AI transformation programme. The failure this time is smaller and quieter: aiming too low, at exactly the moment the tool is becoming capable of more.

The traditional organisational model of functional specialisation, centralised processing, and standardised service delivery wasn’t an arbitrary choice. It was the rational answer to a specific constraint. Capacity was scarce and expensive, so you concentrated it centrally and pushed standardised process out to the frontline. 

With artificial intelligence, that world is disappearing. 

So a question to ask is why do the organisations we work in look the way they do? 

And what should they look like tomorrow? 

Ronald Coase was an economist who, in 1937, asked a deceptively simple question: if markets are so good at allocating resources, why do firms exist at all? Why doesn’t all work just get bought and sold job by job, rather than organised inside companies with employees and managers?

His answer was that using the market has hidden costs such as finding the right person, negotiating terms, writing contracts, checking that the work was completed properly. Once those costs get high enough, it’s cheaper to just employ someone and tell them what to do than to start from the beginning every time you need something done.

Economists call this a transaction cost, and it’s the logic sitting underneath almost every large organisation: centralise the thing that’s expensive to coordinate any other way.

A narrow focus on efficiency in the public sector forced more centralisation.

Local libraries, local offices, local counters closed one by one, because a face-to-face conversation in a real building is expensive, and a phone queue or a website isn’t. Direct access to a person became indirect access to a call centre, then indirect access to a form. Underneath it all sat a genuine belief, borrowed from the private sector in the 1980s and 90s, that services would be better if they were run more like businesses — standardised, measured, centrally managed for efficiency and consistency.

Nobody set out to say local was bad. Local was always the thing everyone said they valued. It was just expensive, and centralised was cheaper — so cheaper won, again and again, until “efficient” quietly became a synonym for “further away.” We didn’t lose local provision because we stopped believing in it. We lost it because we never found a way to make it affordable at scale. It was an irrational answer to a genuine cost problem. 

It’s not just cost that forces centralisation, it’s the desire for consistency. Consistency of reporting, consistency of results, consistency of line management.

Gary Hamel pointed out that “A small organisation may have one manager and 10 employees; one with 100,000 employees and the same 1:10 span of control will have 11,111 managers. That’s because an additional 1,111 managers will be needed to manage the managers.” Bureaucracy doesn’t grow because someone wills it to. It metastasises, because a layer that exists to make sure the layer below it is consistent eventually needs its own layer watching over it. 

The Watchers watch. But who watches the Watchmen? More Watchmen. And so on and so forth.. 

However, AI can do more than take notes. It can watch, manage, coordinate, and look for exceptions. It can allow for consistency when we need it, and can perform tasks faster than 11,000 managers. 

So the big question is: if AI can collapse coordination costs, what happens to the boundary and shape of the organisation?

If organisations exist to reduce the cost of coordination, then a dramatic reduction in those costs should change the structure and size of the company itself. As coordination becomes cheaper, the need to internalise it through hierarchy diminishes. Some are already predicting an extreme version of this.  Marc Bara writes that “The organisational structures that currently exist to manage cognitive coordination face obsolescence not because they perform poorly, but because the problem they solve is becoming inexpensive enough to eliminate the need for dedicated infrastructure.”

This is not just theory — some hierarchies are already measurably flattening. Ewens and Giroud document that U.S. public firms flattened their hierarchies following AI adoption, reducing management layers across a sample of more than 3,100 firms shifting toward flatter structures with proportionally fewer mid- and senior-level employees.

AI can be aimed directly at the cost that justified centralising in the first place. 

If a place-based team can get expert-level analysis, diagnosis, or judgement without waiting two weeks for a team in a central office, then what is the centre for? 

I don’t think this means blowing up the central office. Some things are centralised for reasons that have nothing to do with the cost of coordination — safeguarding escalation, fraud, consistency of risk decisions. What I think it means is a genuinely different question to the one most of the public sector is currently asking. Not “how do we use AI to perform our existing business more efficiently?” but “what did we centralise because coordination used to be expensive — and does that expense still hold?”

The price that made it right to centralise might be disappearing.

Whilst artificial intelligence doesn’t eliminate the need for a centre it fundamentally redefines its purpose. 

Rather than managing the centre becomes enabling. 

Rather than seeking consistency the centre can promote local variation and innovation. 

Rather than requiring managers to do work about work, AI can do that work for them , or eliminate the need for that work entirely. 

Right now though, AI is merely doing the bidding of bureaucracy and the paperwork of centralisation.

Photo by Immo Wegmann on Unsplash