The Next Three Years of AI-Native (And Who Owns the Coordination Layer)

Six months of bringing CRAFT public. What we got right about coordination, what we would revise, and the question that decides the next three years: who owns the layer the business runs on.

Skip: Land the plane

We started making CRAFT public at the start of the year, and we have spent the last ten weeks of it making one claim ten different ways. AI's value is in coordination, not automation. The business that wins is the one that installs the operating system, not the one that buys the most tools. We made that case from the category claim, the customer service framework, the owner-operator advantage, and two Field Notes drill-downs into businesses that share nothing on the surface and everything underneath.

Back in Series 2 we wrote a piece called Three Months In, where we looked backward at what the CRAFT operating system had actually done after a quarter of running it. This is the mirror of that piece. Instead of looking back at a quarter, I want to look forward at three years, because we now have enough engagements behind us to say something about where this goes that is not just a prediction. It is a pattern we are already watching repeat.

In three years, AI-native stops being a thing companies announce and becomes a thing they simply are.

Here is the short version, and then Chuck and I will earn it. In three years, "AI-native" stops being a thing companies announce and becomes a thing they simply are, the way "we use the internet" stopped being a strategy slide around 2003. The interesting question by then is not whether you adopted AI. Everyone will have. The question is who owns the coordination layer your business runs on. That answer is being decided right now, quietly, in choices most operators do not realize they are making.

Chuck: What the Reshuffle thesis got right, and what I would revise

The intellectual spine of this whole series was borrowed. Sangeet Paul Choudary's Reshuffle thesis argues that AI's real economic payoff is coordination, not automation, that the value moves to the layer reconciling messy, fragmented inputs into coordinated action across people, tools, and decisions. Six months ago I thought that was the most useful frame anyone had put on this. Having now built coordination layers inside real businesses, I still do, with one revision.

The model was never the scarce thing. The agents were the easy part.

What the thesis got right is the location of the value. Every engagement we have run confirms it. The model was never the scarce thing. The agents were the easy part, in the wealth management firm and the field service company both. The hard, valuable, defensible work was building the layer that held context and enforced how the business wanted to coordinate. That is exactly where Choudary said the value would land, and it is landing there.

What I would revise is the speed and the smoothness. The thesis can read as though coordination is a property you unlock once the AI is good enough. It is not. It is a thing you build, deliberately, against real organizational friction, and then earn adoption for one skeptical expert at a time. The reshuffle is real. It is just slower and more human than a clean systems diagram suggests, because the coordination layer is not only technical. It is a set of decisions a business has to make about itself, and most businesses have never written those decisions down.

Skip: When AI-native stops being an initiative

The tell that a company has actually become AI-native is that it stops talking about it.

You do not see steering committees for electricity.

Right now AI lives in a transformation initiative. There is a budget line, a steering committee, a vendor shortlist, a slide that says "our AI journey." All of that is the scaffolding of something a company is doing to itself, from the outside in. It is temporary by definition, the way "going digital" was temporary. You do not see steering committees for electricity.

In three years, for the companies that get there, the AI will not be a project. It will be the substrate the business runs on, invisible the way your accounting system is invisible until it breaks. The dispatcher will not think of herself as using an AI tool any more than she thinks of herself as using a phone. The work will just flow through a system that holds the context, proposes the action, routes the judgment to a human, and measures the outcome. The companies that arrive there first will not be the ones that ran the biggest transformation. They will be the ones that treated it least like a transformation and most like plumbing, installed once and then relied on.

Chuck: The maturity model, honestly

If you want to know where you actually are, here is the ladder we see businesses climb. No jargon, just the stages.

Most companies are on the first rung: scattered tools. They have bought AI features that each work alone and none of which share context. This is the eleven-tools firm from two weeks ago. It feels like progress and produces almost none, because the tools cannot coordinate.

The second rung is a connected layer. The context lives in one place, and the existing tools start behaving like one system. This is where the real gains begin, and it is where most of our engagements spend their first months. It is unglamorous and it is the whole game.

The third rung is governed autonomy. The system does not just hold context, it acts inside written boundaries, with the gate enforcing what it can touch and the telemetry proving whether outcomes improved. The human is in command of the judgment, not the keystrokes.

The fourth rung, the one almost nobody is on yet, is when the operating system spans the whole business rather than one workflow, and improving it is a normal part of how the company runs, not a project it launches. I want to be honest: we have helped clients reach the third rung. The fourth is still partly theory, and any consultant who tells you they have a fleet of clients living on it is selling you the brochure, not the building. We are figuring out the fourth rung in the open, the same way we have brought the rest of CRAFT public.

Skip: Who owns the layer

So here is the question that actually decides the next three years, and it is not a technology question.

If a vendor owns it, you have rented your own operating system, and the leverage sits with whoever can turn it off.

When the coordination layer becomes the thing your business runs on, ownership of that layer becomes the most important relationship you have. If you own it, your data, your workflows, your agents, the written record of every decision about what to automate and what to protect, then AI compounds into a durable advantage that is yours. If a vendor owns it, you have rented your own operating system, and the leverage sits with whoever can turn it off.

This is the part we have been deliberate about all series. Clients own the layer: the context, the rules, the agents, the infrastructure. We keep the method, the reference architecture, the reusable components, and the telemetry model we bring to build it. Both of those are true at once, and the line between them is the most important thing to get right in any engagement, because it determines who holds the leverage in three years. The firms that will matter in this market are the ones that build you a layer you own, not a dependency you cannot leave. The AI-transformation crowd that runs a workshop and hands you a slide deck is not even in this conversation. They were never building the layer. They were describing it.

Chuck: What we are still figuring out

I do not want to end a series about honesty with a victory lap, so here is what is genuinely unsettled.

We do not yet know how far governed autonomy goes before the cost of writing and maintaining the boundaries outweighs the gain. We do not know how these layers behave when two of them, built by two different firms for two different parts of a business, have to coordinate with each other. We are watching the maturity model's fourth rung get built in real time and we are learning where it bends. Anyone claiming certainty about the three-year picture is guessing with confidence. We would rather guess out loud and correct in public, which is the whole reason this newsletter exists.

What we are sure of is the direction. The value is in coordination. The advantage goes to whoever owns the layer that coordinates. And the businesses that win the next three years will be the ones that treated that layer as the thing worth owning, while everyone else was still shopping for tools.

That is where Series 4 picks up. We are going to stop talking about the operating system in the abstract and start opening up the parts of it we have only hinted at: the maturity model in detail, the reusable architecture underneath these engagements, and the honest economics of building a layer you own. Less thesis, more blueprint.

Thank you for reading this far, and for six months of it. That is not a sign-off line. We genuinely did not expect this many of you in the room when we started writing about what was actually changing in software delivery, and the conversations in the comments have shaped the work more than a comment section usually does.

AI that runs your business. Not the other way around.