McKinsey’s 2026 State of Organizations report finds that AI adoption is widespread, but meaningful results remain rare. Closing the gap will require redesigning work around people—not simply adding technology.
AI has entered the workplace. Transformation, however, is still waiting in the lobby.
That is one of the clearest messages in McKinsey’s The State of Organizations 2026. Based on a survey of more than 10,000 leaders across 16 countries, the report examines how technology, economic uncertainty, and workforce changes are reshaping large organizations.
Its headline numbers reveal a striking disconnect:
- 88 percent of organizations are experimenting with AI.
- 81 percent report no meaningful bottom-line gains.
- 86 percent of leaders say their organizations are not well prepared to use AI in day-to-day operations.
Access to AI is no longer the main problem. The hard part is turning tools and pilots into a better way of working.
The pilot phase has run its course
Many organizations introduced generative AI through isolated use cases: a writing assistant here, a service bot there, perhaps a coding copilot for the technology team. These experiments can save time, but they rarely change how an organization operates.
McKinsey argues that companies must move toward AI-enabled operating models. That means examining an entire workflow—from the initial request through every approval, handoff, decision, and customer outcome—and deciding how people and technology should work together.
This is a much larger undertaking than purchasing software.
If a process contains unnecessary approvals, duplicate data entry, unclear ownership, and departmental bottlenecks, adding AI may simply help the organization produce confusion faster. Automation creates lasting value when paired with simplification.
Two-thirds of the leaders surveyed describe their organizations as overly complex and inefficient. Traditional responses such as restructuring, cutting costs, or removing management layers are producing diminishing returns. The greater opportunity is to improve how work flows across organizational boundaries.
For automation practitioners, the lesson is simple: automate the process you need, not the process you inherited.
“Automating a broken process doesn’t transform the work. It simply helps the organization produce confusion faster.”
AI agents are teammates, not magic employees
The report places considerable emphasis on agentic AI—systems capable of planning and completing multistep work with some autonomy.
Agents could coordinate activities across systems, handle routine decisions, assemble information, and move processes forward without waiting for a person to initiate every step. McKinsey sees particular potential in areas such as finance, HR, procurement, and IT.
Yet leaders’ near-term expectations are more modest than the hype suggests. Only one in four expects AI agents to operate as autonomous teammates within the next one or two years. Most expect AI to remain primarily a support tool for routine tasks and basic decisions.
That caution is reasonable. Agents do not eliminate accountability. Someone must define their objectives, control their access, review their performance, manage exceptions, and remain responsible for consequential decisions.
The useful question is not, “Which jobs can an agent replace?” It is, “Which responsibilities belong to a person, which can belong to an agent, and where must the two collaborate?”
Employees must be part of that discussion. They know where exceptions occur, where customers need judgment rather than speed, and where a supposedly simple process depends on unwritten knowledge. Treating workers merely as recipients of automation wastes the expertise required to make it successful.
Productivity needs a destination
Leaders are under intense pressure to raise productivity. Forty-three percent of respondents call productivity growth a top priority, while 61 percent report high pressure to deliver further gains.
AI is expected to help. Fifty-five percent believe an AI-capable workforce can produce exponential productivity gains. Leaders also anticipate better access to information, less administrative work, and improved decision-making.
Those benefits are plausible—but only if organizations decide what to do with the time automation releases.
Without a plan, saved time can disappear into extra meetings, growing workloads, or vague efficiency targets. A people-centered automation strategy makes reinvestment explicit. Recovered capacity might go toward better customer service, skill development, quality control, creative problem-solving, or simply a more sustainable workload.
Productivity should not be measured only by the number of tasks completed or positions removed. It should also reflect whether work becomes more useful, reliable, humane, and capable of producing better outcomes.
The biggest barriers are organizational
The leading obstacles to AI adoption are not purely technical. Respondents cite concerns about AI itself, including bias, intellectual property, and effects on jobs; regulatory and ethical risks; and organizational challenges such as silos and change management.
Leadership is another constraint. One in six surveyed organizations lacks a clear C-suite owner for AI adoption, while only 14 percent say leaders consistently champion experimentation with a clear strategy.
Buying tools without establishing ownership produces predictable results: disconnected pilots, uneven rules, unclear accountability, and employees who are asked to adopt systems they do not understand or trust.
Successful adoption depends on some rather ordinary capabilities: a clear business problem, visible sponsorship, usable tools, reliable data, appropriate governance, workflow-specific training, and feedback from affected employees.
None is as exciting as announcing an AI agent. All are more likely to determine whether that agent delivers value.
AI fluency must extend beyond the experts
As human–AI collaboration expands, employees throughout the organization will need to understand what the technology can do, how to guide it, when to question it, and when human judgment must take over.
That does not mean turning everyone into a machine-learning engineer. It means building practical competence around real work.
Training should use actual workflows rather than generic demonstrations. Employees need opportunities to experiment safely, evaluate results, identify failure modes, and participate in redesigning their jobs.
Organizations must also address incentives. If workers believe sharing their expertise with an automation project will simply help eliminate their positions, resistance is not irrational. Trust grows when leaders communicate honestly, provide credible paths to reskilling or redeployment, and involve employees before decisions are finalized.
The real work starts now
The experimentation era must give way to deliberate transformation.
A practical starting point is one important end-to-end workflow. Map its delays, repetitive tasks, decisions, exceptions, and handoffs with the people who perform the work. Remove unnecessary complexity before adding technology. Decide where automation can assist, where an agent might act, and where a person must remain responsible.
Then measure quality, employee experience, and customer outcomes alongside speed and cost.
The central finding of The State of Organizations 2026 is not that AI will transform every company. It is that possessing AI and being capable of transformation are very different things.
Technology can make work faster. Creating better work still requires people.
Automation for the people—not instead of them.
Leave a Reply