Not long ago, “shadow AI” looked like a win for workers.
People quietly used tools like ChatGPT and Claude on personal accounts. They finished tasks in minutes that once took hours. An MIT study found that over 90% of employees used personal AI tools at work. Yet only 40% of companies had official access in place. Workers moved fast. Companies lagged behind.
Managers called it a risk. Workers called it progress. Now the story has shifted.
A global survey by WalkMe shows a clear pullback. More than half of workers skipped company AI tools in the past month and did tasks by hand instead. Another third did not use AI at all. That means about 80% of workers are avoiding or rejecting tools their companies paid to deploy.
This is not about tools failing. It is about trust.
Workers worry about what happens when AI works too well. They fear errors. They fear exposure. They fear being replaced. So they step back.
Why Multimillion-Dollar AI Investments Are Sitting Idle?
At the same time, companies keep spending. Digital budgets rose 38% to over $54 million on average. Yet 40% of that spend falls short because people do not use the tools.
Leaders and workers see different worlds.
Only 9% of workers trust AI with complex decisions. Among executives, that number is 61%. That is a wide gap. Leaders think tools are ready. Workers do not.
The same gap shows up in tools. Nearly 90% of executives say their teams have what they need. Only 21% of workers agree.
They are not describing the same workplace.
Some critics say the results are not surprising. Economist Steve Hanke points to weak productivity data. If AI worked as promised, output would rise fast. That has not happened yet. Use is shallow. Impact is limited.
That view matches what companies see on the ground.
Dan Adika, CEO of WalkMe, asks CIOs how many employees use AI in real work. The answer is often below 10%.
His explanation is simple. AI is like a fast car. Companies bought the car. But most people do not know how to drive it.
Some lack skills. Some lack context. Some lack systems that connect AI to real tasks. Without these, the tool sits idle.
Others use a similar image. Brad Brown from KPMG compares AI to a Formula 1 car. It performs well, but only with a skilled driver. Without that, it adds little value.
The cost of this gap is clear.
From Shadow Users to the Left Behind
Workers lose about 51 days a year to tech friction. That is almost two months. At the same time, workers who use AI well save about an hour a day. The gain and loss almost cancel each other out.
The old shadow AI problem still exists. Many workers still use unapproved tools. But a deeper issue has emerged.
Some workers are not sneaking around rules. They are not using AI at all.
This is not always resistance. In many cases, it is lack of support. A third of workers have never used AI tools. They report low training and high anxiety. They were not guided. They were left out.
Others resist on purpose. They take pride in their work. They see flaws in AI output. They do not trust it with important tasks. In some ways, this looks like quiet quitting. People do what is required, no more.
There is also a policy gap. Many companies want to punish shadow AI. Yet few explain the rules. A third of workers do not know which tools are allowed. Most have never been warned.
This creates confusion. Workers fill gaps on their own.
Some leaders now see shadow AI in a new light. It shows where official tools fall short. It shows where workers need help. The real issue is not access. It is adoption.
Closing the Gap Between AI Promise and Practice
Companies that succeed will focus on the handoff between humans and AI. They will decide when a person leads and when the tool supports. Trust grows in that balance.
Some firms are building paths to get there. KPMG groups workers into levels: builders, makers, and power users. Each level has clear skills and goals. The aim is to move people up step by step.
This is not about raw talent. It is about guidance, practice, and safe space to learn. Workers need to know what AI is good for. They need to test it without risk. They need time to build skill.
Even skeptics change with experience. Hanke once banned AI use. Now he uses it daily as a research tool. It saves time. But he checks the results. He knows its limits. That mix of use and judgment is key.
The lesson is simple. AI can help. But only if people know how to use it.
The gap today is not between humans and machines. It is between promise and practice. The companies that close that gap will move ahead. The rest will keep paying for tools that sit unused.
