A new workforce study puts a number on something a lot of managers have suspected but haven't measured: employees are using AI a lot, but most companies have not told them why.
The study, run by the employee experience firm Culture Amp, pulled data from about 112,000 employees across 123 companies. Most workers said they are encouraged to try AI at work. But close to half said nobody has clearly explained how it connects to the company's actual goals.
That gap matters more than it sounds. Understanding of the risks involved in using AI is high among these same employees, and it climbs the more someone uses it. The problem is direction, not carelessness.
There is a pattern hiding underneath, and it is a little uncomfortable for leadership. The people using AI the most inside these companies are the executives themselves, more than any other level of staff. Yet they are the ones failing to explain the plan to everyone below them.
Here is the part that should worry anyone paying for AI tools across a workforce. Employees who use AI heavily say they feel far more productive. But workload numbers barely move, no matter how much someone uses the tool.
A separate study of digital workers helps explain why. People estimate AI saves them about 11 hours a week, but then spend close to 6 and a half hours a week checking the AI's work, fixing its mistakes, and feeding it the background information it needs. Researchers now have a name for this: babysitting the bot.
This lines up with a widely cited report from MIT that found the overwhelming majority of company AI projects are not producing a measurable financial return yet, even though the tools are everywhere. Individual employees feel the benefit personally. The company, as a whole, often cannot see it on the balance sheet.
There is a second, quieter problem building alongside this one. Employees' sense of what career opportunities exist inside their company has dropped sharply over the past year, the single biggest one-year change in the entire dataset. Satisfaction with career development itself has not moved at all.
People are not upset about the training they are getting. They are unsure what jobs will even exist to grow into. And HR teams are not equipped to answer that question either.
A separate survey of senior HR leaders found that most are struggling to properly judge whether someone is actually skilled with AI, because their evaluation tools were built for a world before AI existed.
The lesson for any company paying for AI tools is simple. Adoption numbers are not the same as results. If workload is not dropping and nobody can say what got removed from someone's plate, the return on that spending is still an open question.
The fix is not pushing more people to use the tool. It is picking specific tasks to hand over completely, tracking whether that time gets reinvested somewhere useful, and telling employees honestly what the plan is, even when the plan is still being written.