the AI org chart

The term "AI org chart" has started showing up in conversations about how companies should adopt AI. In the conversations I've seen, people using it mean something different from what I built, so here's what I mean by it, why I built it the way I did, and why the shape of it has mattered more than I expected.

An AI org chart is an agent system built to look like a company. There are departments. Each department has a head. Each head has specialists under it who do the actual work. There's also one more layer sitting on top of all of it: a routing agent, something like a chief of staff, that reads every request that comes in and sends it to the right department. That routing agent never does the work itself. It only decides who should.

I run my company through one. Nordic Design Collective is about two and a half years old. The agent system underneath it, the one I call NORD, is much younger. I started building it in December of 2025, about seven months ago. Today it's more than fifty agents spread across seven departments, with an executive layer sitting above the whole structure.

why I actually built it

The org chart shape wasn't the goal when I started. Visibility was the goal.

I was running a lot of work in parallel with no real way to see what was happening across all of it. Work was getting done, I just couldn't see it clearly enough to know what needed my attention. I needed a way to see at a glance what work was being done, and in the beginning, the simplest way to do that was to look at what agents were running.

A department name and a specialist name tell you more at a glance than a raw log line does. "The frontend specialist is working on a client project" tells me something instantly. A log entry with a timestamp and a function name doesn't tell me anything until I stop and read it closely.

That's the part I think gets missed when people hear "AI org chart" and picture a diagram somebody drew for a slide deck. I didn't build the structure and then bolt a dashboard on top of it afterward. The structure is the dashboard. The same shape that routes the work is the same shape that reports on the work. I got the visibility for free, because I built the org chart to route work, not to look good in a deck.

building it backward

I think a lot of companies are trying to shove AI workflows into their already existing operations, bolting them on top of whatever systems they have, however messy or improvised those systems might be, or no real system at all. They add a tool here, an assistant there, each one wedged into operations that were never designed with any of it in mind. That ultimately never works well, I think. The bolt gets added, the workflow underneath stays exactly the same, and from what I've seen, it's hard to point to what actually got better.

There's a different way to do it, and it's the way I built NORD. Analyze the company first, before writing a single agent. Break every workflow down into what the company actually does, not what an org chart says it does on paper. Build a complete, honest picture of the operation. Then build systems where the AI augments the human work that's already happening, instead of asking the humans to bend around a shape the AI dictates.

That's what I did with NDC. I didn't ask where AI could slot into our existing org chart. I built the whole operating structure around one question: how lean can our team be while still operating at the capacity of a large agency? That question shaped everything downstream of it: which departments existed, what each specialist was responsible for, what got automated first, and what stayed firmly in human hands.

From what I've seen, most of the public conversation about "AI org charts" runs the opposite direction. It's enterprises asking where agents fit into an org chart that already exists, built for a company that already exists, staffed by people who are already there. That's a reasonable question for a large company to ask. It's just a different question than the one I was answering. I built the structure first, for myself, inside a real business with real revenue and real clients. The human team, about ninety hours of combined weekly capacity across four people, works alongside the structure I built, not on top of a structure someone else designed for a company like mine.

why the same shape works almost anywhere

My read is that departments map onto almost any business, because the same handful of problems keeps showing up no matter the industry. Something has to get made. Something has to get sold. Someone has to support the people who bought it. Someone has to handle the money. Someone has to take the incoming requests and route them to the right person. None of those are specific to my company. They're the operating problems I ran into at NDC, and I'd expect some version of all five inside almost any operation.

That's why the shape travels. When a company needs a new workflow, an org-chart-shaped agent system doesn't need a new architecture to hold it. It needs a new specialist dropped into an existing department slot. The routing agent already knows how to send work to that department. The department head already knows how to review what comes back. Adding a workflow ends up looking like hiring, not like re-engineering the company.

That's who this helps first: solo operators and small teams, the people who feel the weight of the meta-work the hardest because there's nobody else around to hand it to. Visibility and capacity show up together, in the same structure, at the exact size where both are scarce.

The shape doesn't stop paying off once a company grows past that stage, either. The departments, the routing, the review layer, none of it needs to be rebuilt as the company scales. It just needs more specialists inside the departments that are already there.

the diagram isn't the hard part

None of this works because of the diagram. A chart with boxes and arrows doesn't do anything on its own, and neither does an agent system that only has the org chart shape and nothing underneath it.

What actually holds it together is the layer underneath the shape: written rules that define how each department is supposed to operate, automated checks that catch it when an agent drifts from those rules, and memory that persists across sessions so the system doesn't forget what it already decided. One hard discipline holds everything else together: the routing agent never does the work itself. It only decides who does.

Without clear escalation paths and real enforcement, agents start stepping on each other, the same way a human org chart falls apart when the reporting lines exist on paper but nobody actually manages anyone. The org chart gets you the shape. It doesn't get you the discipline. Build both, or the shape is just decoration.

The shape isn't only for companies my size. It works at any size, because it's answering the same three problems at any size: visibility, capacity, and where the next piece of work should go. Start with the problem that hurts the most, give it a specialist, and let the shape grow from there.