AI agents are about to redraw your org chart

The boxes still show functions, and the lines still show who reports to whom. The work has already left that picture. AI agents in the workplace draft, route and check tasks the chart still assigns to a person.

The constraints the chart was built on

Alexis Krivkovich, Brooke Weddle, Holly Price and Vik Sohoni argue, in McKinsey's 5 October 2026 article AI is changing work. Now it has to change the organisation, that agentic AI changes three constraints that have shaped the modern organisation. Human capacity, expertise and coordination. For more than a century, tasks, jobs, teams, management layers and hierarchies reflected those constraints. The tools do not erase the constraints. They loosen them enough that a leadership team can ask a harder question. What can this company do that was not possible before?

Most organisations are not asking it. McKinsey's 2026 state of AI survey, cited in the piece, finds that 80 percent of respondents say AI has improved their individual productivity. 37 percent report a positive impact on EBIT. 6 percent qualify as AI high performers. Nearly three-quarters of those high performers say they have substantially redesigned workflows. About one-quarter of other organisations say the same. Separate readiness research in the article puts 11 percent of organisations at a reinvention horizon, where they rethink how work gets done and how value is created. The rest are still adding tools to a chart drawn for a different set of limits.

That gap is the argument. A person getting an hour back is a real result. On these numbers it becomes enterprise value only when the organisation around the work changes with it. Workforce plans, organisational design and change programmes, run as separate projects, leave the system that produces the value untouched.

An orchestration chart

The article's operating-model claim is concrete. Companies have organised people into functions and hierarchies, and passed work from one part to the next. A different sequence is available. Start with the outcome. Then assemble the humans and the agents who can deliver it.

A McKinsey survey of more than 700 executives and senior leaders, reported in the same piece, finds that the companies furthest along are more likely to organise cross-functional teams around end-to-end products, customer journeys and business processes. Decisions and accountability sit closer to where value is created. The authors do not offer a single winning structure. The common thread is organising the work end to end, cutting handoffs that do not earn their keep and moving decisions and resources when the priority changes.

Once agents carry a meaningful share of the work, design becomes a question of orchestration. Which humans and which agents should gather around an outcome? What does each of them do? Which handoffs and interfaces are still necessary? An orchestration chart writes those choices down. In some cases, the authors say, agents do more than join an existing team. They remove the need for coordination between people or functions, and the path the work travels changes. Autonomy, access, oversight and escalation then belong on the chart. They are part of how the work is designed, especially when a team is moving quickly.

McKinsey describes one product organisation where pods of eight to ten people became hybrid teams of four to six, supported by agents. The chart for that team shows which work moved to agents, which human roles now span more of the product process, which handoffs disappeared, and where human judgment and accountability still sit. No company in the piece is offered as a finished blueprint. The example is specific enough to copy as a drawing exercise.

Meetings sit on the same drawing. They have been how an organisation transfers information, builds alignment and makes decisions. Agents can track which decisions are made, what work is still open and what needs attention next, so the meeting can spend its time on judgment and commitment.

There is a warning in the client work McKinsey reports. Coordinator roles, the internal jobs that mostly move decisions across the organisation, grew at roughly 1.5 to 2 times the rate of line roles over five years. Agents create a chance to take some of that tangle out. A flatter drawing pays only if the coordination you remove is replaced by clearer decision rights and stronger human judgment. Delete the box and leave the decision unnamed, and the work waits in someone's inbox.

Where the freed capacity goes

Agents can take research, drafting, coding, analysis, monitoring, summarisation, routing and coordination. The job title can stay while the substance of the day moves. The productivity has to be assigned. More output, different work, a saner load, lower cost, or some mix of those. If nobody makes that choice, the hour a person got back disappears into the same queue.

In a conversation reported in the article, Stanford economist Erik Brynjolfsson argues that a plan focused on cutting the workforce misses the larger chance to rebundle work around what people can now do with AI. I would put that choice on the same page as the chart. A headcount saving spends the capacity once. The other uses have to be named before finance books the hour as a reduction.

The feel of the work shifts with the tools. If a system writes the first draft, prepares the analysis or recommends the decision, the person becomes a reviewer of output they did not make. McKinsey expects that shift inside professional, managerial and technical roles. Too little human involvement and people lean on the system past the point where they can check it. Too much, and the role becomes a pile of leftover decisions. The chart should say which parts of a role stay human on purpose.

Career paths are still built on the old drawing. Advancement has meant larger teams, more coordination and a higher layer. If agents absorb status tracking, synthesis, routing and coordination, some of the work that justified those layers shrinks. Managing a bigger team becomes a weaker default path. The article argues for paths that reward craft, judgment, the ability to orchestrate humans and agents, trust with a customer or client and the quality of decisions made with AI in the work. A senior individual contributor who can lead agents around a hard outcome may create as much value as a manager with a large team. The promotion system has to be able to say so.

A second design job sits under the first. If agents take the entry-level tasks through which people have built expertise, today's productivity can become tomorrow's judgment gap. McKinsey's warning is that companies have to rebuild how people learn as the work underneath them changes. Managers spend less time coordinating activity and more time creating practice, feedback and decisions that carry real consequences. An orchestration chart that shows the agents and forgets the apprenticeship is only half a design.

One global life sciences company, in the article's account, has written a boundary in advance of the case-by-case fights. AI may support the work. It should not replace human creativity, expertise, autonomy, agency or oversight. Employees remain accountable for AI-assisted work. A principle like that does not settle every case. It stops the first difficult decision from inventing the rule under pressure.

One outcome and one owner

A single redraw will not hold if every function keeps running its own experiment. The authors describe an agentic mission factory that starts with an outcome, such as funding a good loan faster or resolving a customer issue in one interaction. The business owner brings the outcome, the economics, the constraints and the accountability. The factory brings business, technology, product, design, data, risk, legal, change and HR into the room from the start. Controls sit inside the workflow. Ownership is assigned at launch. Evaluation, permissions, autonomy and the life of the agent are part of the operating model.

That needs a senior leader who can move funding and scarce talent, shape investment and settle competing priorities. The title on the door matters less than those levers. The authors' point is that no existing function has a natural mandate to do this end to end, because the work crosses business, technology, talent and risk.

The readiness numbers explain why the authority has to be real. In McKinsey's readiness research, run from February to April 2026, 70 percent of respondents say they are personally ready for AI, while 27 percent of leaders say their organisations are ready for the shifts an agentic model requires. Personal readiness comes from a panel of 750 employees. The organisational figure comes from 608 leaders in that research. People are prepared to change their own work. The organisation around them is lagging. A factory with no right to move people or money will produce another pilot and another steering pack.

What to draw this month

If you lead learning, innovation or a transformation programme, you can start with one outcome that currently crosses at least three functions, and with the people who actually run it.

1. Name the outcome in a sentence a customer or a colleague would recognise. Approving the loan, closing the books, resolving the case.
2. List the handoffs. Mark which of them an agent could carry, and which a person must still own because the judgment or the relationship sits there.
3. Draw the orchestration chart on one page. Humans, agents, what each does, where a decision stops for a person, and where an exception goes.
4. Assign the freed capacity before anyone books it as a saving. Say whether it becomes more output, different work, a lighter load or a lower cost.
5. Name how a newer person still gets the practice that used to hide inside the tasks an agent now does.
6. Put one senior owner on the page who can move people and money. A committee that only recommends will recreate the coordinator layer you just drew out.

Collective Campus builds high-performance teams for the AI world. Our AI training workshop is a hands-on day for putting the tools into a real workflow, with guardrails for data, accuracy and compliance, and a short plan for the next 90 days. The people who own the outcome still have to draw the chart. The workshop is where they get enough practice with the tools to know which boxes an agent can honestly take.

McKinsey's practical test is the one I would use in the next operating review. Can the organisation make decisions, move information and coordinate work with fewer layers, fewer handoffs and less manual translation between functions? Another licence will leave that test untouched. The chart is the work.