AI agents are joining your workforce. Your workforce isn’t ready

Deloitte’s August 2026 survey found 43% of leaders expect agentic AI to disrupt workforces within 18 months, rising to 72% over 2-3 years, but only 25% are prepared and just 5% consider processes highly ready. Guy Couillard argues the old model of training is inadequate: a course can explain what an AI tool does but cannot recreate the pressure of deciding whether to trust it. Teams need realistic simulation environments where failure is useful before they deploy for real.

Executives are making a dangerous mistake with AI. They are preparing employees to use new tools when they should be preparing them to work differently. As AI agents begin to recommend, decide, coordinate, and act, the real transformation is not happening inside the software. It is happening inside the organization. If leaders do not prepare people for that shift before deployment , they are not managing an AI transformation. They are gambling with it.

Deloitte’s August 2026 survey of 501 U.S. business and IT leaders found that 43% expect agentic AI to significantly disrupt their workforce s within 12 to 18 months, rising to 72% over two to three years. Yet only 25% said their organizations were prepared or highly prepared in workforce readiness, and only 21% said the same about business processes. Just 5% considered their processes highly prepared for AI agents. The ambition is enormous; the readiness is not.

I have spent decades watching organizations introduce powerful technologies and make the same error. They focus on whether the system works and underestimate whether people are ready to work differently. AI makes that error far more dangerous because these systems are moving beyond assistance. An agent can execute a task, influence a decision, trigger another process, and hand work to another agent. That changes who does what, who decides, and who is accountable.

Employees must know when to accept an AI recommendation, when to challenge it, and when to stop it. When one department lets an agent optimize its own work, leaders must understand what that decision does to finance, operations, sales, customers, and the rest of the enterprise . The person responsible cannot shrug and say, “The AI did it.”

This is why I believe the old model of training is fundamentally inadequate. A course can explain what an AI tool does. It cannot recreate the pressure of deciding whether to trust it when the decision has consequences.

People need to experience the new way of working before they are asked to perform it for real, and they should experience this as a team, not as an individual.

That means giving teams realistic, risk-free environments in which they can run business processes, collaborate across functions, work alongside AI agents, who themselves become team members, make decisions, and see the consequences. Simulation is powerful precisely because failure is useful when nothing is actually at stake. Competition changes the experience. People test, adapt, remember, and learn. Leaders, meanwhile, can see where teams overtrust AI, hesitate unnecessarily, or fail to recognize a downstream impact.

Get this right, and the upside is substantial. People become faster because they know what to delegate. They become better decision-makers because they understand where human judgment matters. Departments work from a shared view of the business rather than optimizing isolated tasks. AI becomes a multiplier of human capability instead of another expensive system sitting on the sidelines.

Get it wrong, and the opposite happens. Companies spend heavily on agents that employees do not trust or understand. Workflows become more complicated instead of simpler. People experiment in live environments, burning through costly AI usage while learning by trial and error. Errors travel faster because machines can execute them faster. Investment rises while adoption stalls.

Deloitte ‘s research makes the choice stark. Seventy-five percent of surveyed leaders said human collaboration with AI agents creates more value than automation alone, while half said their organizations are not adequately investing in AI-related workforce transformation. Nearly two-thirds are reevaluating their business models, yet only about one in five say they are prepared to redesign processes for autonomous work.

So, before you approve another AI deployment, demand proof that your workforce is ready for it. Put employees into realistic AI-enabled scenarios before they are responsible for real outcomes. Make them challenge the system, intervene when it fails, and confront the consequences of their decisions. Measure whether they can actually operate in the new environment, not whether they completed a training course.

Do not put AI into the hands of a workforce that has only been taught how to use it. Prepare people to question it, manage it, and take responsibility for what it does. Workforce readiness should not be the final step before deployment. It should be the condition that makes deployment possible.