Restructure for Human-AI Agent Collaboration

TL;DR: The state of organizations is shifting toward human-AI agent collaboration, and the restructure has to happen at the level of the org chart, not the tool stack. McKinsey’s 2026 research names collaboration between humans and AI agents as one of nine forces reshaping companies, and argues the future structure is a network of small teams supervising agents around shared end-to-end outcomes. The blocker is institutional knowledge trapped in departmental silos. A semantic layer that connects those silos is what lets people and agents actually work from the same facts.

Most companies are adding AI agents to an org chart that was drawn for a different problem. The boxes still map to departments. The lines still mean “who reports to whom.” Then a procurement agent needs a contract that lives in the legal team’s workspace, and the whole arrangement stalls at the first handover.

That mismatch is the real subject of McKinsey’s State of Organizations 2026, a report built on a survey of more than 10,000 senior executives across 15 countries and 16 industries. This piece walks through what the research actually says about human-AI agent collaboration, why a structural restructure beats another pilot, and how a shared knowledge layer makes the new model run.

What does the State of Organizations 2026 actually say?

The report identifies three tectonic forces reshaping organizations: the infusion of AI and automation into how work gets done, intensifying economic and geopolitical disruption, and evolving workforce expectations (McKinsey, 2026). From those forces it draws nine concrete shifts.

Two of the nine are the heart of the restructure question:

• Collaboration between humans and AI agents. The report treats this as a new working model, not a feature you switch on.
• Rewriting the future of shared services with AI. Functions that used to be back-office cost centers become candidates for redesign around agents.

The framing matters. McKinsey’s emphasis moved from the short-term resilience of its 2023 edition to sustained productivity and long-term performance, with technology and AI placed “at the core of organizational transformation” (McKinsey, 2026). That is a structural claim. It says the change belongs in how the company is built, not in a single team’s toolset.

Why the old org chart breaks under AI agents

A companion McKinsey article, The agentic organization, is blunt about the cause. Traditional organizations are built around functional silos. Even digital companies with cross-functional product teams are limited by handovers and human team-size constraints like the two-pizza team and Dunbar’s number (McKinsey, 2025).

An agentic team is the proposed building block: a small group of multidisciplinary people who own and supervise the AI workflows underneath them. McKinsey reports that a human team of two to five people can already supervise an “agent factory” of 50 to 100 specialized agents running an end-to-end process such as onboarding a customer or closing the books (McKinsey, 2025).

That ratio changes the shape of the company. The recommendation is direct: organization charts based on hierarchical delegation pivot toward agentic networks, or “work charts,” based on exchanging tasks and outcomes (McKinsey, 2025).

Most companies are nowhere near that today. McKinsey estimates that 89 percent of organizations still operate in the industrial age, 9 percent run agile or product-and-platform models from the digital era, and only 1 percent act as a decentralized network (McKinsey, 2025). The gap between where companies are and where the agents need them to be is the work.

Shared intelligent services: the unit of the restructure

This is where the two shifts connect. If agentic teams own end-to-end outcomes, then the functions feeding those outcomes can stop being departmental silos and become shared intelligent services instead.

A shared intelligent service is a cross-functional capability where a small human team supervises agents that run a process for the whole organization, rather than each department maintaining its own disconnected copy. McKinsey points to banks already running mortgage and compliance work with “agent squads,” and to one global bank whose “agent factory” handles know-your-customer processes with ten agent squads (McKinsey, 2025). Those squads do not respect the old departmental lines, and that is the point.

The restructure, then, is not a reorg of people into new boxes. It is the move from siloed functions to shared services that humans and agents operate together.

The blocker nobody reorgs away: context

Here is the constraint that an org-chart redraw alone will not solve. Agents are only as good as the context they can reach. McKinsey makes this explicit in its technology recommendations: organizations need to “wall in proprietary organizational context, institutional knowledge, and nonpublic data” as a source of competitive advantage, on architecture that separates that knowledge from the underlying vendor landscape (McKinsey, 2025).

That is a knowledge problem before it is a structure problem. If a shared intelligent service spans what used to be five departments, the agents running it have to read across five tools that were never designed to talk to each other. A flat work chart on paper does not give them that. A connective layer does.

A semantic layer is a unified map of an organization’s entities — people, documents, projects, customers, tools — and the relationships between them, queryable as one graph rather than a dozen disconnected systems. It is the substrate a shared service runs on. Without it, every agent re-derives context from scratch and every human still plays messenger between teams. This is the gap SemanticOS is built to close: a knowledge-graph and AI-search layer that connects fragmented enterprise tools into one operational view, so people and agents reason over the same institutional knowledge instead of separate copies.

A concrete example: Vantage Health restructures claims

Take Vantage Health, a mid-size insurer. Its claims operation runs the way most do: intake sits in one system, clinical review in another, fraud checks in a third, and customer communication in a fourth. Each team has its own queue, its own context, and its own habit of emailing the next team to ask what happened upstream.

Vantage wants a claims agent to draft adjudication summaries. The pilot looks good in a demo and falls apart in production, because the agent cannot see the clinical note that justified an earlier exception. It lives in a tool the agent was never connected to. So a human goes and finds it, by hand, the way they always did.

The restructure that fixes this is the one McKinsey describes. Vantage forms a small claims pod, four people, who supervise a set of agents covering intake through communication as one end-to-end shared service. Underneath, a semantic layer links the claim, the policy, the clinical notes, the prior exceptions, and the customer history into a single graph. Now the adjudication agent traverses that graph in one query, the fraud-check agent reads the same entities, and the human pod steers outcomes and handles the edge cases where agents fail. The departmental handovers that used to define the work are gone, because the silos under them are connected.

The structure made the collaboration possible. The semantic layer made the structure work.

Key takeaways

• The State of Organizations 2026 names human-AI agent collaboration as one of nine forces reshaping companies, and frames it as a structural change, not a tool rollout.
• The recommended restructure replaces functional silos and hierarchical org charts with networks of small teams supervising agents around end-to-end outcomes.
• Shared intelligent services are the unit of that change: cross-functional capabilities humans and agents run together, where departments used to each keep their own copy.
• An org-chart redraw is not enough. Agents need institutional knowledge they can actually reach, which means a semantic layer connecting the tools the old silos kept apart.
• Only about 1 percent of organizations operate as a decentralized network today, so the distance between most companies and the agentic model is the real project.