Use.AI survey finds workers using AI look more capable

That matters because the economic payoff from AI at the firm level depends not just on faster output, but on whether managers can still distinguish genuine skill from machine-assisted performance. If they cannot, pay, promotion and succession systems become noisier. Workers who are better at prompting and polishing may rise faster than workers with deeper judgment, while companies risk rewarding output they do not fully understand.

The survey points to a growing disclosure problem inside offices. Some 39% of respondents said they submitted AI-assisted work without saying so, and 30% said they accepted praise for results substantially generated by the technology. In practical terms, that means performance reviews increasingly rely on finished work that may reveal little about who actually did the hard part.

Use.AI chief executive Ihor Herasymov argued companies should not require every AI interaction to be disclosed, saying that would quickly become unworkable as the tools are embedded into daily software. But he said firms need rules for “material” AI use — when a model meaningfully shapes an analysis, recommendation, presentation or code — so managers can assess both the work and the human contribution behind it.

For investors, the issue goes beyond office etiquette. AI spending has become a major capital allocation theme for software and cloud companies such as Microsoft, Alphabet and Adobe, all of which are trying to turn model access into higher productivity and stronger customer retention. Microsoft’s latest filings said AI infrastructure spending is already lifting costs even as adoption of products like Microsoft 365 Copilot grows. If enterprises respond by tightening oversight, the monetization case for productivity software may improve; if they become more skeptical of AI-generated work, adoption could slow or shift toward tools that are easier to audit.

The risk is not that AI eliminates all judgment, but that it masks where judgment is actually coming from. Herasymov said finished results are becoming a less complete measure of skill and that employers will need to test whether workers can explain their reasoning, spot errors in AI answers and make decisions when systems are wrong or uncertain. That would favor workers with stronger “problem framing” skills — the ability to ask the right question and challenge assumptions — over those who simply produce polished outputs.

The broader labor-market implication is that AI is increasingly a credentialing problem, not just an automation problem. The technology may lift output and speed, but it can also flatten the signals employers use to identify readiness for more responsibility. In the near term, that could push companies toward more structured assessments, more disclosure around AI use and more emphasis on judgment-based interviews. For investors, the key question is whether AI becomes a durable productivity multiplier or a source of hidden operational and governance risk.