The Exception Is the Job: What the Next Org Chart Actually Looks Like

The organizations being built right now produce serious output with almost no human structure inside them. Agents run the standard path, and the humans handle what breaks. That inversion, from humans doing the work and machines assisting to machines doing the work and humans handling the exceptions, is the largest change to corporate structure in my working lifetime.

I have been asking one question since I was a kid, and I have never really stopped: is it possible to do this faster, better, and at higher quality than the version that came before. It started with software and small electronics experiments, taking things apart to see whether the design was actually the best available answer or just the one that shipped. The habit never left. And at some point I stopped applying it only to systems and started applying it to organizations, which are just systems with people inside them.

The first time that question cost me something was when I started my first technology company and decided to build it as a distributed team. This was many years before remote work was normal, before anyone had a word for it. We had people in Toronto, in Nova Scotia, in Houston, in Cancun, in Costa Rica, and in Peru, and we were delivering complex enterprise solutions to clients who mostly assumed the team was down the hall. There was no Zoom. There was no Slack. The tooling was primitive by any standard you would apply today, and coordinating a distributed services organization across that many time zones with what existed then was genuinely hard. But it worked. We delivered, repeatedly, and the geographic location of the people doing the work turned out to be almost irrelevant to the quality of the output.

I am not telling that story for nostalgia. I am telling it because the structural bet underneath it, that you do not need to co-locate people to produce excellent work, was contrarian then and is invisible now. Every company is global today. Teams span continents as a matter of routine and nobody considers it a strategy. That is what happens to a structural insight once it is proven: it stops being an advantage and becomes the floor. Which means the interesting question is never the last structural bet. It is the next one.

Here is the next one. What does an organization look like when the constraint is no longer where the people are, but how few of them the work actually requires?

I think we are heading toward organizations that deliver high-value output to clients with the minimum viable human structure inside them. There is a director, a CEO, someone accountable and setting direction. Below that, increasingly, are teams of agents doing specialized work: not one general assistant, but many narrow, competent processes each owning a defined slice of the workflow, running continuously, writing back into real systems rather than producing suggestions for a person to execute. The humans are not gone. They are relocated. They sit where judgment is actually required, which is at the edges, on the cases the system was not designed for.

That relocation is the whole thing, and it is why I would put it in the title. In a conventional company, exception handling is the annoying residue of the process, the thing you push down to whoever has the least leverage. In an agent-run company, exception handling is the job. The standard path is automated by definition. What remains for people is the ambiguous, the novel, the high-stakes, and the genuinely new. The org chart stops being a hierarchy of execution and becomes something closer to a map of where human judgment is load-bearing.

The market is already moving there faster than most people realize. Gartner projects that by the end of this year up to forty percent of enterprise applications will include task-specific agents, against less than five percent a year earlier. That is not gradual adoption. And on the company-formation side, Y Combinator has started using the term “20x company” for startups producing with three to five people what used to take twenty or more. When the unit of output stops scaling with headcount, every assumption baked into how we structure firms comes loose.

The hard part is that this transition is far more difficult for established companies than for new ones, and the reason is structural rather than cultural. A mature company’s processes are not arbitrary. They are the accumulated product of decades of systems, SOPs, controls, handoffs, and approvals, and every one of those was designed around the shape of a human doing the task. Retrofitting agents into that architecture produces exactly what you would expect: the machine gets bolted onto a workflow that only exists because people used to do it, and you automate the artifact instead of the outcome. It is the same failure I have written about before, starting from the tool instead of the result.

Which is why the most revealing thing I see large corporations doing right now is setting up very small, light teams to build purely AI-native processes from scratch, running in parallel, aimed at producing the same output the existing organization already produces with its systems and its people. Read that carefully, because it is an admission. They are not improving the existing process. They are quietly rebuilding it clean, next to itself, because they have concluded that the old architecture cannot be incrementally converted. That parallel-build pattern is the clearest signal available of where this is going.

And it points to a real, temporary, and large advantage for anyone starting from zero. A new company gets to design its architecture around agents from the first day, with no legacy process to honor, no system of record shaped by a headcount plan, no organizational muscle memory to fight. It can define the outcome first and build backward, which is the only way I have ever known to build anything well. The incumbent has to do the same work while carrying the old structure on its back.

I want to be honest about what this breaks, because the version of this argument that ignores the cost is not worth reading. If exception handling is the human job, exceptions require judgment, and judgment is built by doing the ordinary work first. Freeze entry-level hiring because agents now handle the standard path, and you get an inverted pyramid: a generation of seniors capable of handling the hard cases, and nobody behind them who learned how. That structural debt does not come due immediately. It comes due in three to five years, and it will be very expensive and slow to repair. The organizations that get this right will redefine what entry-level work is rather than deleting it. The second problem is measurement. Most dashboards were built to count human activity, and they quietly stop meaning anything when the execution is agentic. Velocity can rise while quality falls and the chart will not tell you.

None of this is a prediction about technology. It is a prediction about structure, and structure is the part most people underestimate because it is invisible until it fails. I have made this bet once before, when I argued that a distributed team could deliver work as well as a co-located one, and the thing I remember is how ordinary that idea became once it was proven. The same will happen here. Building a company around a small core of humans and a large fleet of agents will look radical for a short time and obvious shortly after. The advantage belongs to whoever reads the structure early, defines the outcome clearly, and builds for it before it becomes the floor.