Putting AI inside the existing organization is adoption. Redesign begins when the organization itself changes around the new shape of work.
McKinsey reports that 80% of respondents say AI has improved individual productivity, while only 37% report positive EBIT impact and just 6% meet its definition of AI high performers. That is not proof that organization design is the missing variable. It is a strong reason to ask why individual gains do not automatically become enterprise gains.
AI changes work first.
Jay Galbraith’s Star Model argues that organization design is more than structure. Strategy, structure, processes, rewards and people have to align.
Henry Mintzberg approaches the problem from another direction. Organizations differ not only in boxes and reporting lines, but in how work is coordinated: direct supervision, standardized processes, standardized outputs, standardized skills and mutual adjustment.
AI changes the economics of several of those mechanisms at once. It changes how work is decomposed, how expertise is accessed, how information flows and how much one person can direct.
Automating the old organization.
Imagine a company designed around a set of constraints. Experts are scarce. Information moves slowly. Managers coordinate because individual workers cannot see the whole system. Jobs bundle many tasks because coordination between people is expensive.
Then AI changes some of those constraints. Research becomes faster. Information becomes easier to retrieve. Some expertise becomes more accessible. One person can direct more execution. Cross-functional work can move differently.
The common response is to insert AI into the existing boxes: same roles, same approvals, same meeting cadence, same handoffs, same measurements, same team boundaries. But faster.
That is not redesign. It is acceleration of the existing operating model.
I can see the misalignment in my own work.
One person can now direct far more execution than the role was originally designed around.
The unit of work has moved. What once looked like writing one message can now look like preparing and statusing an entire staged wave, with research, verification and structured handoffs happening before I touch the exceptions.
But the surrounding organizational mechanisms do not automatically move with that capability.
That is the kind of misalignment Galbraith helps make visible. The technology changed the process. That does not automatically change structure, decision rights, measures or roles around it.
Galbraith’s warning is still useful.
The Star Model says an organization cannot be redesigned through structure alone.
Change one substantially and the others eventually experience pressure to move too.
AI is already changing processes and capabilities. If structure, measurement and role design remain frozen, misalignment is predictable.
What became possible?
The first question is not how many tasks AI can automate. It is whether AI changes what the organization can compete on.
Faster response? More personalization? Broader account coverage? Shorter product cycles? More analysis per decision? Expert-level support available to more employees? A different service model?
If AI creates a genuinely new capability, strategy should determine which capability matters.
Without that decision, AI adoption fragments. Every function optimizes locally. The organization becomes more productive without becoming more coherent.
Where should authority sit now?
Structure is not simply who reports to whom. It determines where power and decision authority are located.
If an employee can access expertise that previously required several handoffs, does the old approval path still make sense? If one manager can direct a larger amount of execution through AI, should their role remain designed around supervising individual tasks? If specialists can contribute through shared systems rather than permanent team assignment, should expertise stay trapped inside functional silos?
AI does not automatically imply flatter organizations, fewer managers or smaller teams.
It changes the conditions under which those design choices were originally made. Then the organization has to reconsider them.
Mintzberg helps explain why.
Mintzberg’s organization-design work focuses heavily on how work is coordinated.
- direct supervision
- standardization of work processes
- standardization of outputs
- standardization of skills
- mutual adjustment
Different kinds of work rely on different mixtures.
AI can change the relative cost of those mechanisms. A workflow encoded as repeatable procedure can shift work toward standardization of process. A strong evaluation system can allow more work to be managed through standardization of output. Grounded AI can make specialized knowledge more accessible. Complex, novel work may still depend heavily on mutual adjustment between people.
The same company may need more than one design.
A regulated repeatable workflow may benefit from standardization and clear roles. A research team working on an uncertain problem may need small expert groups and constant mutual adjustment. A sales organization may standardize account preparation while leaving relationship strategy decentralized. A product organization may automate testing heavily while keeping product direction highly collaborative.
AI makes hybrid design more likely, not less.
The mistake is assuming one company-wide “AI operating model” should dictate every kind of work.
Follow the work, not the boxes.
Galbraith describes processes as the way information and decisions move through the organization. That flow often crosses the org chart.
AI can change those lateral flows before formal structure changes at all.
Where is information being recreated? Where does work wait for a handoff? Where does one function own information another function repeatedly needs? Where does an AI system make a traditional coordination meeting unnecessary? Where does AI create a new dependency that did not exist before?
If the process changes but the organization still rewards and manages the old process, friction appears.
AI can remove coordination work and create integration work.
AI may reduce the need for people to manually collect updates, assemble information, format reports, route routine work and restate context.
That is coordination work.
But removing coordination does not eliminate the need for integration. Someone still has to decide which objective matters, which tradeoff wins, which function has priority, what the organization will standardize and where local variation remains useful.
Stop paying for the old unit of work.
Organization design breaks quickly when the workflow changes but performance measures do not.
If AI allows one employee to produce ten times as many artifacts, rewarding artifact volume becomes less meaningful. If AI takes over routine execution while the human role shifts toward judgment, quality and integration, measuring only throughput creates the wrong behavior.
Microsoft’s own framing points in the same direction: as human-agent work shifts from Author to Editor to Director to Orchestrator, the unit of work and the human contribution change with it.
Measurement should follow the work.
Redesign roles after redesigning work.
Jobs are bundles of tasks, responsibilities, relationships and decision rights. AI rarely changes every component at once.
One part of a role may automate. Another may become more valuable. A third may move to another function. A fourth may become an entirely new coordination responsibility.
That means redesign should begin with the work. Then rebuild the role around what remains.
McKinsey’s 2026 analysis describes AI as changing three fundamental organizational constraints: capacity, expertise and coordination.
The unit of work is moving.
Signal 020 used Microsoft’s software-engineering progression as one concrete example:
As that unit expands, the human role moves from creating individual pieces toward evaluating, specifying and orchestrating larger bodies of work.
You do not need to believe every profession will follow software engineering exactly. The organizational question still holds.
If the smallest meaningful unit a person owns becomes larger, should their role stay the same? Their manager’s span? Their performance measures? Team boundaries? Decision rights?
Maybe. But that should become a design decision rather than an assumption.
The org chart is a lagging indicator.
The org chart is attractive because it is visible. Boxes. Lines. Titles.
But most organizational change happens underneath it first.
Work moves differently. Information moves differently. Decisions move differently. Expertise moves differently.
Only later do titles and reporting lines catch up.
A company can already be operating as a different organization while continuing to display the old one.
Start with coordination, not headcount.
If I were redesigning a function around AI, I would not begin with: How many people should this team have?
What work changed?
Which tasks disappeared, expanded or moved?
How is work coordinated now?
Direct supervision, standard process, output standard, expertise or mutual adjustment?
Where does information flow?
Which handoffs remain necessary and which are artifacts of the old model?
Where does decision authority belong?
Did execution move without authority moving? Did expertise become more distributed?
Only after those questions would I change headcount, team design or reporting lines.
AI does not redesign the company for you.
There is a temptation to describe all of this as inevitable. AI arrives. Organizations flatten. Teams shrink. Managers become orchestrators. Functions dissolve.
Maybe some of those things happen in some organizations.
Organization theory gives us a better answer:
AI changes the constraints. It does not remove the design problem.
The redesign test.
A company has not redesigned around AI simply because employees use AI, agents exist, automation increased or a new AI team appears on the org chart.
I would look for something deeper.
- Has the unit of work changed?
- Have roles changed around that new unit?
- Have processes and handoffs changed?
- Have decision rights moved?
- Have performance measures changed?
- Has the organization changed how it coordinates different kinds of work?
If the answer is no, the company may have adopted AI without redesigning around it.
The organization is not the boxes.
Galbraith’s lasting contribution was to make organization design larger than structure. Mintzberg’s was to show that organizations are defined partly by how their work is coordinated.
AI makes both ideas newly relevant because the technology is changing work faster than many organizations are changing around it.
The redesign challenge is not: How do we fit AI into the organization we already have?
That question begins with strategy, moves through coordination, changes processes, reshapes roles and only then earns its way onto the org chart.
AI changes work first.
The organization has to decide whether to follow.
Jay R. Galbraith — Star Model
Henry Mintzberg — Structure in 5’s, Management Science, 1980
McKinsey — AI is changing work. Now it has to change the organization, October 5, 2026
Microsoft WorkLab — Author / Editor / Director / Orchestrator framing, 2026