Half my team already uses AI in secret. The other half is terrified. Now what?

Picture the all-hands where you finally said the word "AI" out loud, and you watched the room split in two. One half leaned forward, a little too quickly, as if they had been waiting for permission they had already granted themselves. The other half went still, the way people go still when they think the next sentence might be about them. By the end of the week you had named it to yourself: you preside over a split workforce you never sanctioned. A quiet group is already running its whole job through AI on personal accounts. Another group is frozen with fear. And you authorised neither.

So here is the short answer, before the diagnosis. This shadow AI and uneven adoption is a workforce divide created by a permission failure, not a tooling shortage. The secret users are not first a security problem to police; they are your clearest reading of unmet demand. The frozen half are not slow; they are protecting something. The work is to surface the energy you already have and give it a sanctioned home, then bring the second group across on the back of the first. You can begin with the team you already have. You need a posture before you need a platform.

Why is half my team using AI in secret and the other half terrified?

Start with the half running ahead, because their reasons are sharper than they look. In Ivanti's 2025 research across more than 6,000 office workers, nearly a third of employees who use generative AI at work, 32%, keep it secret from their employer. The motives are revealing: 36% like the "secret advantage" it gives them over peers, 30% fear their job may be cut, and 27% report what Ivanti calls "AI-fuelled imposter syndrome" (the feeling that your competence is borrowed from a tool, so you hide the tool rather than admit the borrowing). Read that twice. The most capable people in your building are hiding their best instrument because the culture has made disclosure feel like a risk rather than a contribution. That is a workforce reading the room accurately.

The other half are responding to a real perception of threat. Harvard Business Review's 2026 analysis of why generative AI feels so threatening describes workers experiencing it as a challenge to three core psychological needs, competence, autonomy and relatedness, and responding with resistance and disengagement. When those same needs are met, the authors note, the very same people embrace AI as a copilot (an assistant that works alongside them rather than replaces them). The fear is rational and human. It signals that the meaning of someone's work feels up for renegotiation, and nobody senior has told them how the story ends.

Why don't I, as the leader, see how wide the divide already runs?

Because the instrument you trust most, your own read of the room, is calibrated wrong on this one. McKinsey's 2025 workplace study put the gap in numbers: C-suite leaders estimate that only about 4% of employees use generative AI for at least 30% of their daily work, when the real figure sits closer to 13%, roughly triple their guess. You stand in front of a wave and read it as a ripple. The secret users widen the gap further, because the people moving fastest are the ones most motivated to stay invisible.

This is the part worth a long pause. The divide is structural, not local to your firm. The World Economic Forum's Future of Jobs Report 2025 (the WEF is the body behind the annual Davos summit; this is its global employer survey) estimates that 39% of workers' core skills will change by 2030, and frames it as a hundred-worker problem: 59 would need training, of whom 29 could be upskilled in their current role, 19 upskilled and redeployed, while 11 are unlikely to receive the reskilling they need. Encouragingly, 85% of employers plan to prioritise upskilling their workforce. The map exists. What most buildings still lack is permission and a path, which is exactly what your frozen half are waiting to receive.

What actually closes the gap: tooling or permission?

Permission, by a wide margin, and the survey data backs it. When WalkMe (a workplace-software firm owned by SAP) polled 1,000 working US adults in 2025, 78% admitted to using AI tools their employer had never approved, while only 7.5% had received extensive training and 23% had received none at all. More than half, 51%, reported conflicting guidance on when and how to use AI. So the tooling already sits in people's hands; what is missing is sanction and instruction. EY's 2025 Work Reimagined survey of 15,000 employees across 29 countries puts a prize on closing that gap: organisations can capture up to 40% more productivity from AI when it rests on a stable talent foundation, yet only 12% of employees say they receive enough training to get there. The companies winning treat this as a human transformation that happens to involve software. The bottleneck is no longer the technology.

So channel the energy rather than policing it. Your secret users are doing your product research for free, in the open market of their own workflow. The move is to make their work sayable.

This is also where the deeper frame earns its place. We talk about Phase Three, and the ladder underneath it matters here. Phase One, the Age of Effort: work hard, get a little more, linear growth. Phase Two, the Age of Scale: build once, sell to millions, exponential growth. Phase Three, the Age of Acceleration: output decoupled from human effort almost entirely, the phase AI unlocks. The split in your building is the early friction of that shift, and it resolves through people, not procurement. Here is the sequence I would run, in order, because order is the whole game here.

1. Grant amnesty, out loud. Name the split you can see and declare the personal-account work safe to surface. The 32% are hiding because the culture taught them to; you change that by going first.
2. Find your hidden builders. Ask who is already getting real lift, and make telling you a status reward rather than a confession. These people become your internal faculty.
3. Give the energy a sanctioned home. A blessed set of tools and a clear line on data turns shadow use into shared use, so the advantage compounds across the whole team instead of one quiet desk.
4. Bring the frozen half across on competence, autonomy and belonging. Pair them with the builders, protect their role identity while it shifts, and let peer proof do what a mandate cannot.
5. Steward the eleven. Of every hundred, some need redeployment, not exhortation. Plan that deliberately and the other 89 trust you more, not less.

Notice what this is, underneath the steps. It is leadership choosing coherence (head and heart pulling the same way) over control. The leaders who win the next decade are the ones who upgrade themselves first, and on this question the upgrade is a single decision: to treat the people running ahead as scouts rather than offenders, and the people standing still as colleagues protecting something worth protecting. Phase Three is mind, and the mind here is yours. Make the secret sayable, and the divide starts to close the moment you treat it as a question of people rather than software.

About the author

Thomas Green

British technology futurist, AI keynote speaker and advisor. Thirty years across enterprise technology and AI strategy, helping leaders navigate the future of work. The futurist who died.