Changing the Shape

Organizations are increasingly redesigning work around what AI makes possible, reconsidering team size, roles, management layers and how much execution can move from people to agents. All of those choices matter, and together they can produce an organization that looks very different from the one it replaces.

Meta appears to have gone further down this path than most.

Reuters reported in August on an internal program called Project OT, short for Organization Transformation, that took shape at a Meta leadership retreat in January 2026. According to a planning document reviewed by Reuters and people familiar with the work, the company explored replacing product development teams of ten to twenty specialists with pods of three to five people, moving away from conventional titles toward a broader builder role, reducing management layers, and having agents perform much more of the daily work under human supervision.

This was meaningful organizational redesign. Meta was exploring what its organization could become when AI was capable of taking on significantly more of the work.

The program was expected to unfold in two waves. Meta proceeded with roughly eight thousand job cuts in May, while a second wave planned for November was cancelled before the first began. Reuters also reported that Meta paused an initiative intended to track the transformation and allowed some engineers who had moved into new AI-focused roles to return to their previous teams.

Reuters reported that the reversal came as internal data raised questions about whether the technology was delivering. Code changes to Meta’s software rose 220 percent year over year, while changes producing new or upgraded features users could actually see rose 36 percent. The company has disputed some characterizations of the program.

The picture is still incomplete, and it is difficult to draw firm conclusions about what worked and what didn’t. What the example does give us is a useful way to examine how deeply an organization has to redesign when AI changes who, or what, can act.

Meta’s plans addressed many of the elements we traditionally associate with organizational design. It reconsidered how many people belong on a team, which roles are needed, how many layers sit between employees and leadership, and how work should be divided between people and agents.

Agency introduces another layer of design.

Once an AI agent can perform meaningful work, the organization has to decide what that agent is actually empowered to do. It needs to establish where human judgment enters, how much discretion a person has to challenge or redirect what the AI produces, who can intervene as conditions change, and who remains answerable for the result. It also has to consider whether the people working alongside AI have enough context and authority to exercise judgment when it matters.

Those decisions shape the experience of work just as much as team size or reporting lines do.

They also have to come earlier.

If agency were the starting point, the design process would begin by understanding the outcome the organization is trying to create and the capabilities available across both humans and AI. From there, it could determine where agency should sit, where it needs to be shared, what requires human judgment, what AI can act on independently, and what accountability has to remain human.

Only then would questions about team size, roles, management layers and headcount have enough context to answer well.

A three-person pod may be exactly right for one kind of work and completely wrong for another. Removing a management layer may give the people closest to the work greater agency, or it may remove the person who had the context and authority to resolve an exception. Giving an agent more execution may free a person to exercise more judgment, or it may leave that person responsible for an outcome they no longer have enough proximity to understand.

The structure alone cannot tell us which outcome we will get.

This is where I think organizational design is changing most profoundly.

For much of the history of modern organizations, the human was implicitly the agent. Work could be divided into roles, authority could be attached to positions, decision rights could move through a hierarchy, and accountability could ultimately be traced back to a person.

AI introduces another source of agency into that arrangement. It can increasingly interpret, recommend, create, decide and act, while people are simultaneously moving into different forms of contribution around it. Agency becomes distributed across humans and AI, and it can shift between them depending on the work, the conditions and what happens next.

Designing for that requires understanding the movement of agency before deciding the shape that should contain it.

Meta’s experience is still unfolding, and we don’t know whether designing from agency would have changed the decisions it made or the outcomes it experienced.

But it gives us a useful example of how much an organization can change while still leaving a deeper design question unresolved.

When the capacity to act changes, the organization has to be designed around that new reality.