The org chart is about to get weird

Open your company's org chart. Every box is a person. Every line is a reporting relationship between people. The whole diagram rests on an assumption so deep nobody ever needed to state it: work is done by humans, and accountability flows through humans.

That assumption held for the entire history of management. It's now developing cracks, and the cracks are showing up in the most bureaucratically boring places imaginable. Access requests. Approval chains. The question of who gets blamed when something goes wrong.

Questions that sound absurd until they land on your desk

An agent drafts a customer-facing document. Whose name goes on it?

An agent needs CRM access to do its job. Who approves that, and under whose identity does it operate? Your IT team has a process for provisioning humans and a process for provisioning software. An agent that reads the sales channel, joins document threads, and messages people when it finds problems is neither, or both.

Two teams rely on the same agent's output. When their priorities conflict, who does the agent listen to? Nobody has ever had to ask who a spreadsheet reports to. People are starting to ask it about agents, in those words, half-joking and then not.

An agent hands work to another agent, which hands it to a third, and the final output causes a customer complaint. Who owns the incident? The person who kicked off the chain? The one who reviewed the output? The team that configured the agents in the first place?

Performance review season arrives. A manager's team shipped three times more than last year, but a large share of the execution wasn't human. How do you evaluate the manager? How do you evaluate the individual whose main contribution was superb delegation to agents?

None of this is science fiction. Versions of all of it are being improvised right now inside companies that adopted AI faster than their org design could absorb. Improvisation is the wrong tool here, because these are all the same question wearing different costumes: how does accountability work when some of the work isn't done by a person?

The tree can't hold what's coming

Org charts encode three different things at once, and we rarely bother separating them. Authority, meaning who can decide what. Accountability, meaning who answers for what. And information flow, meaning who needs to know what.

For humans, bundling them into one diagram mostly works because all three travel together with the person. Your manager decides, answers, and routes.

Agents break the bundle. An agent can execute without holding accountability, and it must never hold accountability, full stop. No serious organization will accept "the AI did it" as an answer to a regulator or a customer, and any company that tries will deserve what happens next. An agent can route information without having authority over anyone. And an agent can serve five teams at once, which a tree-shaped diagram literally cannot draw. Ask where agents go on the org chart and the honest answer is that they don't. Not on that chart. The chart is the wrong data structure for what teams are becoming.

The patterns showing up early

Nobody has the final answer, but the improvisations are converging on a few shapes, and they rhyme with things organizations already know.

The most common one could be called the accountability sandwich. Every piece of agent-executed work gets a human on both sides: one who delegates and defines the standard, one who reviews and accepts or rejects. Often the same person. Agents do the work in the middle, sometimes several of them passing it along a chain, but accountability never leaves the bread. It's boring and robust, which is why it's quietly becoming the default among teams that have thought about this for more than five minutes.

A second pattern treats agents as shared services, the way legal or design ops serve multiple teams rather than belonging to one. That solves the multi-team problem and creates a new one, because now someone has to own the service, its standards, and its failure modes. Expect a job title for this within two years. It will probably have "operations" in it.

The third pattern is slower and deeper: roles reorganize around decisions instead of tasks. Job descriptions stop reading "writes reports, maintains dashboards" and start reading "owns pricing calls under $50k, accountable for pipeline data quality." Task lists were always a proxy for what we actually wanted from people. When agents absorb the tasks, the decisions are what's left, and it turns out the org chart of decisions is more stable than the org chart of tasks ever was.

The harder problems are the human ones

Performance reviews built on individual output become nonsense when output is a blend of the person, the agents they direct, and the agents those agents call on. What you actually want to measure shifts toward judgment: did this person delegate the right things, catch the errors, escalate the right calls? Measurable, in principle. Measured by almost no rubric that exists today.

Career ladders have a quieter problem. The traditional deal was that juniors do execution and seniors do judgment, with execution as the training ground. If entry-level execution migrates to agents, where do juniors learn? Companies will need to deliberately preserve learning-by-doing, or in five years they'll discover they have no bench of people who understand the work deeply enough to review it. I'd nominate this as the most underrated organizational risk of the whole transition, and I notice almost nobody in a leadership role is talking about it yet.

If you run a team, you don't need a grand reorg to get ahead of this. You need a record of what work is delegated to agents and who owns each delegation, so that the day something goes wrong you're managing an incident and not a scandal. You need one sentence, agreed in advance, about when agent work requires a human decision. And you need a shared answer to what "good enough to ship" means, because vague standards plus fast execution equals fast garbage.

The org chart got weird once before. When companies first computerized, there were sincere debates about where "the computer department" belonged and who was responsible when the machine made an error. It took a decade, several spectacular failures, and eventually new roles and new diagrams. We're at the start of that decade again. The chart isn't dying. It's about to stop being a tree.