Companies are eager to be seen adopting AI. Their employees are often just as eager not to be seen using it.
That contradiction has become hard to ignore. Executives face pressure from boards, investors, and customers to explain their AI strategy and show visible progress. The demand for a convincing story has moved faster than many organizations’ ability to decide where the technology is genuinely useful, so public claims about AI adoption can outpace the work happening inside the business.
Meanwhile, employees are already using AI to research unfamiliar subjects, draft documents, analyze information, prepare presentations, test ideas, and shorten tasks that once consumed hours. Much of this use is informal, unreported, and sometimes deliberately concealed.
The hesitation is rational. An employee who discloses using AI may worry that colleagues will discount the result, even when producing it still required experience, judgment, and a precise understanding of the problem. Greater productivity can create its own anxiety. A task completed in two hours instead of two days may be treated not as evidence of skill, but as proof that the employee can absorb more work, that the team needs fewer people, or that the role itself is becoming expendable.
So companies can end up overstating their use of AI to the outside world while underestimating its use within their own walls. Management imagines it is directing a controlled transformation. In practice, much of the experimentation is happening through personal accounts, improvised workflows, and tools the organization has never reviewed.
This is partly a technology problem, but the technology is not the most difficult part. The harder questions concern trust, incentives, workplace culture, and the way companies decide what knowledge work is worth.
Most organizations still struggle to separate effort from value. Time spent, visible activity, and procedural difficulty are treated as evidence that serious work occurred. When AI helps someone produce a strong result more quickly, the company may welcome the efficiency while quietly valuing the work less because it appeared easier to produce.
That contradiction runs through much of modern office life. Companies say they pay for outcomes, but many of their systems still reward visible exertion. An employee who finds a faster method may therefore have good reason to keep it private.
Hidden AI use also creates legitimate risks. Employees may paste sensitive information into consumer tools, depend on systems that have not been tested, or build important business processes outside the company’s approved technology environment. But surveillance and blanket restrictions are unlikely to solve the problem. They may only teach people to conceal their use more carefully.
At Astra Intelligence Labs, we are exploring what a more workable relationship between companies, employees, and AI could look like. Organizations need room for useful experimentation without giving up security, accountability, or quality. Employees need to be able to acknowledge AI assistance without having their expertise dismissed or their productivity used against them.
That raises questions with no settled answers. How should companies value work when the time required to produce it changes dramatically? Where does AI assistance end and human judgment begin? What should employees be expected to disclose? What information does an organization actually need in order to manage risk and understand where value is being created? And how can it gather that information without treating its workforce as a potential security breach?
We are still working through those questions. We are speaking with people inside organizations, examining where the incentives break down, and looking for places where better tools or operating practices could make AI use more visible without making it punitive.
For us, building in public includes discussing the problems that shape our work before they become products. That does not require publishing every hypothesis, method, or commercial idea. It means being clear about the territory we are studying and the reasons we think it deserves attention.
The distance between how companies describe their AI adoption and how employees actually use the technology will create opportunities for new products, services, and operating models. But the gap reveals something more basic: many organizations are trying to introduce a new form of technology while preserving old assumptions about effort, productivity, and value.
Those assumptions may prove harder to change than the software.