Insights
Everyone bought the software. Nobody trained the people.
Ask an employee where they learned to use AI at work, and the honest answer is usually "nowhere, I just started clicking around." Only 39% of people globally who use AI at work have gotten any AI training, according to Microsoft and LinkedIn's Work Trend Index 2024, a survey of 31,000 knowledge workers across 31 countries. A separate, larger study — KPMG and the University of Melbourne's Trust, attitudes and use of AI: A global study 2025, with more than 48,000 respondents across 47 countries — landed on the exact same figure, independently. Two research teams, two different years, one number. That's not a coincidence. That's the baseline.
Nobody was planning to fix it, either
The Microsoft and LinkedIn data doesn't stop at "untrained." It also found only 25% of companies were planning to offer generative AI training that year — meaning roughly three out of four employers had no plan to teach anyone anything, even as their people were already using the tools. Training wasn't deprioritized in some deliberate tradeoff. It mostly wasn't on the list at all.
This is the part that gets missed in AI budget conversations, and it's worth being precise about the distinction: budgets skewing toward tools over training is a money problem — dollars flowing to licenses instead of people. What's described here is what happens on the ground once that money has already been spent: real employees, doing real work, with no instruction, quietly making their own arrangements.
The arrangements people make on their own
They don't stop using AI. They just stop waiting for permission. The KPMG study found 70% of employees use free, publicly available AI tools rather than anything their employer provides — against 42% who use an employer-provided tool. Those two figures add up to more than 100%, which means a meaningful share of people are using both, largely without anyone steering the choice. Only about two in five report their organization has any policy or guidance on generative AI at all. And the training figure shows up again here, unprompted: 39% report having received any form of AI training — the same number Microsoft found, from a completely different sample.
That's the environment: a tool available to almost everyone, guidance available to almost no one.
What that environment produces
The same KPMG study measured the downstream effects, and the pattern is consistent rather than scattered:
- 66% of employees have relied on AI output without critically evaluating it
- 56% have made mistakes in their work because of AI
- 57% have used AI non-transparently — presenting its output as their own or concealing that they used it at all
Read those three together and a shape emerges. It isn't "some employees misuse AI." It's a majority operating without the judgment training would have given them, a majority who've already been burned by trusting output they shouldn't have, and a majority who've learned, correctly, that admitting they used AI carries some risk — so they don't. None of that is a character problem. It's what happens when you hand someone a capable tool, skip the instructions, and hope for the best.
Training is where the capability actually lives
Software sits on a server. Capability sits in a person's head — what they know to check, when they know to push back on an answer, when it's faster to just do it themselves. Untrained employees don't lack access to AI; the numbers above make clear they have plenty. What they lack is judgment about it, and judgment doesn't arrive by osmosis. It gets built, deliberately, the same way any other job skill does.
This is the part of AI adoption we spend the most time on, because it's the part almost everyone skips: not a slide deck on what AI is, but a workshop built on a team's actual work, followed by a check-in at 7, 30, and 90 days to see what actually stuck. If you're staring at a team that's already using AI — well or badly, you may not know which — a short briefing is the honest place to start. Details are on the Team AI Upskilling page, and it's one part of the Carbon Copy Method.
Source: Microsoft and LinkedIn, "Work Trend Index 2024" (31,000 respondents, 31 countries); KPMG and the University of Melbourne, "Trust, attitudes and use of AI: A global study 2025" (48,000+ respondents, 47 countries). For the money side of this same gap, see Your AI budget is backwards; for how we close it, see Team AI Upskilling.