There is a version of the AI transformation conversation I keep hearing from senior leaders, and it goes roughly like this: AI can surface dependencies, summarise context, reduce the cost of coordination — and if that’s true, maybe we don’t need all the layers we built. Maybe we can finally flatten the org, get closer to the work, move faster.
What they don’t say out loud is that they built the org that needs flattening.
John Cutler named this directly last week: the same people who benefited from riding the ZIRP wave into safer, more insulated leadership positions are now talking about removing bureaucracy, reducing layers, getting back into the details. “The people who helped build the system are now turning around and saying the system is out of control.”
He is right. And the question that matters is not whether AI can help. It is whether the people asking for its help are willing to do what it actually requires.
How the kingdoms got built
The General Manager model made sense on paper. Give a leader a team, a budget, a clear area of ownership, and business accountability for outcomes. They move fast because they have everything they need. They make better tradeoffs because they own the consequences.
What the model underestimated was the geometry problem. As organisations grew, the number of dependencies between autonomous units grew faster. Each unit optimised for its own outcomes. The seams between units became the most important and least owned part of the business — a dynamic Conway’s Law predicts with uncomfortable precision.
The coordination burden did not go away. It got redistributed downward. The people closest to the work — the product managers, the engineers, the analysts — spent their time discovering dependencies that nobody had planned for, negotiating exceptions across team boundaries, routing around political obstacles that existed because no leader wanted to acknowledge shared accountability.
I have been in those rooms. The org chart said one thing. The actual flow of work said something completely different. And the people absorbing the gap between the two were always the ones with the least formal power to change it.
What AI can and cannot do
AI can genuinely help here. Cutler is fair about this. It can reduce information friction. It can summarise meeting notes, surface patterns, help smaller teams hold more context than they previously could. If an organisation has already done the hard structural work — clear ownership, matched decision rights, leaders willing to prioritise across units — AI becomes a useful multiplier.
But AI cannot do the structural work. It cannot clarify who decides when two teams disagree. It cannot neutralise the incentive misalignment between a business unit leader whose bonus depends on their metrics and a platform team whose work enables everyone but gets credit from no one. It cannot resolve contested ownership. It cannot create the organisational slack required to actually coordinate rather than just communicate.
These are not information problems. They are power and incentive problems. And the reason they persisted through the ZIRP era is that the people with the power to fix them were often the people who benefited from leaving them unfixed.
Coordination theatre
Cutler introduces a concept I find worthy of entering the org design vocabulary: coordination theatre: the appearance of working together without the friction of actually deciding together. Agents sending summaries to other agents. Copious documentation flying across Slack in every direction. Well-formatted async briefs that cover every angle except the one that requires someone to make a call and own it.
This is the failure that AI makes extremely easy to fall into. If you were already inclined to avoid the uncomfortable conversation — the one where two teams have incompatible priorities and someone senior has to choose — you now have a whole new toolkit for creating the impression of collaboration while the actual decision keeps getting deferred.
The front line absorbs this too, just as it always has. The coordination burden that used to arrive via org design — too many handoffs, too many dependencies, too many approval layers — now arrives via async abundance. More messages, more threads, more agents generating more output that someone, somewhere, has to actually read and act on.
The volume changes. The problem doesn’t.
What actually has to come first
I wrote recently about AI transformation being a sequencing problem. The organisations making progress are not the ones with better models. They are the ones who stopped waiting for everything to be ready and started with a narrow, clearly owned scope.
The same logic applies here. AI-enabled org flattening is possible, but the sequence matters. You cannot use AI to skip the structural work. You have to do the structural work first, or AI simply accelerates the existing dysfunction.
That means clearer end-to-end ownership. Decision rights that actually match the accountability being asked for. Senior leaders willing to make the tradeoffs across units that they have historically avoided making. Smaller teams with real autonomy, not nominal autonomy inside a fiefdom structure that constrains every significant decision anyway.
None of this is AI’s job. All of it is leadership’s job.
Cutler’s line stays with me: “AI can amplify a coherent operating model. It cannot substitute for one.”
The organisations that will use AI well in the next five years are not the ones with the best prompts or the most agents. They are the ones that built something coherent enough for amplification to mean something — where AI makes the right things go faster rather than making the wrong things go faster with better formatting.
The kingdoms we built were optimised for a world of abundant capital and loose accountability. That world is gone. AI is not going to restore it. It is going to make it more visible for what it always was.
Leave a Reply