The agentic organization: rewiring the company, not the workflow
McKinsey's State of Organizations 2026 surveys over 10,000 executives and lands on a blunt message: experimenting with agents is easy, and almost nobody has rebuilt the organisation that would let them matter.
TL;DR
- McKinsey's State of Organizations 2026 draws on more than 10,000 senior executives across 15 countries and 16 industries.
- Nearly two-thirds of enterprises have experimented with AI agents; fewer than one in ten have scaled them to tangible value.
- 53 percent of leaders still expect AI mainly as a support tool over the next one to two years; only 25 percent expect agents acting as autonomous teammates.
Writing in Bloomberg this week, McKinsey's global managing partner Bob Sternfels compressed the firm's research into one sentence: the hardest challenge is not adopting AI, it is rewiring your company around it. The underlying evidence sits in The State of Organizations 2026, and it describes a gap that should feel familiar by now, at a scale that should not.
The experimentation trap
Nearly two-thirds of enterprises have piloted AI agents somewhere. Fewer than 10 percent have scaled them into anything that moves a business outcome. That ratio, roughly six experimenters for every scaler, is the agentic version of the pilot purgatory documented for generative AI since 2023, and it has the same root cause: agents get added to an organisation designed for a workforce that no longer looks like the one actually doing the work.
An agent dropped into an unchanged process inherits the process's problems and adds its own.
The expectation data explains the inertia. A 53 percent majority of leaders see AI as a support tool for the next one to two years, and only a quarter expect agents operating as autonomous teammates. If leadership plans for a copilot era, it builds copilot-era structures: individual licences, human-shaped role descriptions, approval chains that assume every action has a human author. Meanwhile the technology, as McKinsey documents, is already carrying out multistep processes, from customer interaction through transaction processing to follow-up coordination.
What the scalers do differently
The report's through-line matches what every serious study this year has found, from work-chart thinking to frontier firms: the winners move from scattered use cases to operating-model redesign. In practice that means designating end-to-end workflows, not tasks, as the unit of transformation; assigning clear ownership for agent behaviour and its failures; and rebuilding role descriptions around directing and evaluating machine work. It also means confronting the barriers executives actually name, with regulatory and legal concern, organisational friction, legacy infrastructure, and missing strategy well ahead of model capability.
Getting out of the six-to-one ratio
- Pick one end-to-end workflow with a measurable owner and rebuild it around agents, rather than running ten more scattered pilots.
- Decide your agent operating rules once, centrally: what agents may do autonomously, what requires sign-off, and who answers when one is wrong.
- Fund reskilling in the same decision as deployment. A rewired process without rewired roles is a pilot with better marketing.
The organisations that treat agents as an org-design problem are pulling away from those that treat them as a procurement item. The ratio says most are still in the second camp, which is precisely the opportunity.
Related reading
- McKinsey boss: AI's next big mission is rewiring your workplace (Bloomberg)
- From org charts to work charts: organising for AI
Source: The State of Organizations 2026, McKinsey & Company.