66% of Companies Know Human-AI Collaboration Matters but Only 6% Are Doing It

The gap nobody is closing

Two out of three organizations say that designing how humans and AI agents work together is important, according to Deloitte's 2026 Global Human Capital Trends report. Only 6% are making significant progress. That is a 60-point gap between knowing something matters and doing anything about it.

The report surveyed thousands of business and HR leaders worldwide. The pattern it found is consistent across industries. Companies recognize the problem. They are not building the systems to solve it.

McKinsey's State of Organizations 2026 report, which surveyed more than 10,000 senior leaders across 16 countries and 17 industries, found the same thing from a different angle. 92% of organizations plan to increase AI investment. Only 1% consider themselves mature in AI adoption. The money is moving. The organizational design is not.

Tech-only deployments are failing at 1.6 times the rate

Deloitte found that 59% of organizations are taking a technology-focused approach to AI. Those organizations are 1.6 times more likely to miss their ROI expectations than those that also redesign how humans and AI work together.

A European telecom cited in the report added an AI assistant to its customer service workflow and saw productivity rise 5% on its own. After it put 90% of its rollout budget into redesigning the workflows, trust protocols, and training around the AI, that gain jumped to 30%. The technology did not change. The work design did.

This is the core of the problem. Companies are buying AI tools and dropping them into existing workflows. The workflows were designed for humans working alone. When an AI agent generates a plan, a forecast, or a recommendation, someone has to review it, adjust it, and act on it. That review layer is not being designed. It is being improvised by the people closest to the work.

Nobody is measuring the human side

42% of workers told Deloitte that their organizations rarely evaluate AI's impact on people. The measurement gap explains the execution gap. If no one is tracking what happens to the humans in a human-AI system, no one is designing for them.

The trust numbers are worse. Only 16% of knowledge workers trust senior leaders to make the right decisions about people. And 80% of leaders and workers are concerned that their colleagues are using AI to appear more productive than they actually are. The AI is creating a performance measurement crisis on top of the workflow design problem.

When a team member uses an AI agent to draft a project plan in 20 minutes instead of four hours, the manager sees faster output. What the manager does not see is whether the plan accounts for the constraints that only exist in the team's informal knowledge. The speed gain is visible. The quality risk is not.

The PMO is scaling the technology faster than the planning

McKinsey found that 40% of organizations with revenue above $1 billion are now scaling AI agents in one or more functions, up from 27% a year earlier. At the same time, only around 25% of leaders expect AI agents to act as fully autonomous teammates within the next two years. Most still expect AI to work mainly as a support tool in the near term.

That gap matters most for the PMO, the function responsible for deciding how people and resources get allocated to projects. Agents are being scaled into more functions faster than expectations about their autonomy are changing, which means the planning decisions are still largely being made the way they were before any agent existed: by headcount, not by combined human-and-agent capacity.

A PMO that plans capacity in headcount cannot plan for a team where three humans work alongside two AI agents. The agents do not take PTO. They do not need onboarding. They also do not raise flags when a plan conflicts with reality. The staffing model has to account for all of this, and the scaling data suggests most organizations are adding agents faster than they are redesigning the planning function that would absorb them. Only 51% of leaders account for human-machine collaboration in workforce planning at all, according to Deloitte.

The execution gap is also an orchestration gap

Deloitte found a second gap that mirrors the first. 88% of leaders rated the ability to orchestrate people and resources as important. Only 7% reported making actual progress. 85% ranked adaptive workforce capabilities as critical. Only 7% are leading in them.

The pattern is the same across every metric. Leaders understand the problem. The organizations they run are not building the systems to solve it. The reason is structural. Workforce planning sits in HR. Technology deployment sits in IT. Project execution sits in the PMO. Human-AI collaboration design does not sit anywhere, because the role does not exist yet.

What to actually do about it

Before deploying an AI agent on any team, map the workflow it will participate in. Identify every point where a human has to review, validate, adjust, or override the AI's output. Design that review layer as a defined process with clear criteria, not as an informal expectation that people will figure it out.

Measure both sides. Track the AI's output quality and track the human effort required to make that output usable. If the review layer costs more time than the AI saves, the deployment is net negative regardless of what the tool's dashboard says.

Build a staffing model that treats AI agents as capacity, not as free additions. A team of five humans and two AI agents is not a team of five with free help. It is a system of seven that needs to be planned, measured, and managed as a system.

The 60-point gap between knowing this matters and doing it is not a technology problem. It is an organizational design problem. The companies that close it first will have a structural advantage over every competitor still dropping AI tools into workflows designed for humans working alone.