I first learned to name this gap in a graduate course on adaptive leadership, partway through a career spent managing technology change inside a large international organization.
The final assignment was supposed to be a memo: explain the concepts to my own team. What I actually turned in was a diagnosis of why our transformation wasn’t landing. On paper, the project was a success: delivered on time and on budget. Sit with the team actually doing the work, and a different picture emerged – morale eroding under task-switching, siloed handoffs, and hours lost to coordination instead of the technical work itself: no time to swarm a hard problem, no room for critical thinking, just pressure for a quick fix.
Both readings were accurate. They simply weren’t describing the same organization, and that same gap is opening up again now, in a different form, around AI.
Senior leaders rely on multiple channels to understand what’s happening across an organization: town halls, surveys, dashboards, and advisors who synthesize the noise into something that can be acted on. Those systems were built for a world where change moved through hierarchy: up from the field, filtered at each level, arriving at the top in a form clean enough to inform decisions.
AI doesn’t move that way. Its adoption often begins with individual and team experimentation and spreads through channels outside the formal ones leadership depends on to see where that is actually happening.
Research on generative AI adoption across seven European software companies by Kai-Kristian Kemell of Tampere University found that adoption does not always follow a top-down path, with AI use sometimes emerging from the employees themselves. This creates a visibility gap – not because AI creates it, but because it reveals which systems already had the sensing infrastructure to catch a shift early, and which didn’t.
The question is not whether AI is being adopted within an organization, but whether leaders can see where and how this change is taking place, before it has already reshaped the work.
By the time an AI working group is meeting, something else may already be in motion. In a regional office, a coordinator quietly built a prompt that cuts her most time-consuming task from four hours to 40 minutes. In a technical division, two analysts started sharing a workflow they developed together: nothing official, no ticket raised, no approval sought. In an operations team, a junior staff member became the person everyone goes to when they want to know what AI can actually do.
None of this necessarily appears in an executive briefing. None of it came through the strategy process. It arrived the way change often arrives: not from the top, but from the edge. From the people closest to the work, with the most immediate pain, and the least to lose by trying something new.
The initiative doesn’t create adoption; it arrives late to something already in motion. Done well, it gives that motion structure. Done poorly, it gets in the way.
The question is not whether this kind of experimentation is happening within the organization. It’s whether anyone at the top can see it.
Why Organizations Struggle to See What’s Changing
This is the difficult part, and it has nothing to do with competence or loyalty.
The people who help leaders read the room are often excellent at synthesis: taking complexity and translating it into something that can be acted on. They know which signals matter, which voices carry weight, and how to move information through the hierarchy without losing its meaning.
That role is also changing shape. Anna Shipman, writing on executive AI literacy, notes that most boards still lack the fluency to sit with an uncertain account of what AI is actually doing inside their organization. When leaders can’t hold that complexity, the people translating it upward come under pressure to flatten it into a fast, uncontroversial answer instead of the messier one — not a failure of middle management, but what happens when people are asked to protect leaders from what they don’t understand, rather than help them understand it.
That skill set assumes information moves in one direction: upward, through filters, until it’s usable. AI adoption doesn’t move that way. It moves laterally. It moves quietly. It lives in individual judgment calls and informal peer networks and workarounds that nobody names out loud because nobody asked. It doesn’t raise its hand. It doesn’t file a report. The problem now is not that the formal signal is wrong. It’s that it is probably incomplete.
There’s a further complication that leaders least want to hear: staying quiet about a misalignment is often the more rational choice, not the less honest one.
John Cutler, who writes on product leadership, argues that organizations are optimized to “avoid confronting deep-seated tensions or ‘elephants in the room’ at the leadership level,” leaving front-line teams to work around them pragmatically.
Research on workplace silence confirms the pattern: James Detert and Amy Edmondson found that employees hold widely shared beliefs about when speaking up is risky, beliefs that predict silence even when the suggestion would help the organization. Raising a misalignment rarely reads as neutral: it’s a judgment on work other people built, and the person who names the gap absorbs that cost personally while any benefit accrues to the organization as a whole.
So the people best positioned to see the gap between strategy and actual work are also the people with the least reason to say so out loud – which is why the briefings can be accurate and still incomplete, and why that gap is where strategy quietly fails.
The cost is not immediately visible. That’s what makes it dangerous. Leaders move forward on the strategy. They invest in the infrastructure their teams recommended. They formalize the approach, communicate it across the organization, and build governance around it. And what has been formalized is the old picture.
The coordinator now has to use the official tool, which doesn’t work as well as her own. The analysts have nowhere to put what they built. The informal expert gets no acknowledgment and stops volunteering what she knows. The organization hasn’t just missed the adoption already happening – it has interrupted it.
It is also where the real risk sits. Governance that exists only on paper does not track tools nobody logged, access nobody revoked, or workarounds nobody reviewed, and that unattended surface, not the AI itself, is what eventually costs an organization its trust and its security.
I saw the same blind spot from the other side of my own portfolio: a unit whose engagement scores looked strong on paper – people proud of the organization, satisfied with their colleagues. What the score never asked was what kept the unit standing: one person’s knowledge with no real backup, work piling up until each budget cycle’s end and then absorbed all at once, priorities set by whoever’s request carried the most weight. None of that was a failure the survey could have caught; it wasn’t designed to ask those questions.
AI can help expose these gaps by surfacing patterns formal processes tend to miss – but technology alone can’t fix a visibility problem. Leaders have to look beyond the formal picture they’re given, or an underperforming initiative gets blamed on readiness or training rather than on whether leadership ever had a complete picture of what was happening on the ground.
What Leadership Requires Now
The answer is to broaden the sources of information leaders draws on, and keep widening it as the environment keeps changing. According to Cornelia Choe and Marshall Goldsmith, leaders can’t manage a complex environment on the strength of their own judgment and a single line of sight, however good both are.
This matters for AI specifically because adoption was never just about automating a task – it changes how people work, how they interact with each other, and increasingly with the AI itself. Leaders rarely have visibility into that shift, and without it, guardrails are guesswork: governance built blind is either too loose to catch the real risk or rigid enough to strip out the value it was meant to protect. Seeing what’s actually happening is the precondition for governing it well, not something added once the initiative is underway.
For AI adoption specifically, that means building a second channel. Not a parallel hierarchy. Not a shadow working group. Something simpler and more disciplined: a deliberate practice of going closer to the edge.
In practice, this means small recurring habits rather than a new process: managers asking “what did you try this week?” in existing one-on-ones, peer-led show-and-tells where colleagues walk through their workflows, and reverse mentoring pairings – junior staff demonstrating the tools, senior leaders sharing back the organizational context and judgment those tools need to land well.
Leadership has always required knowing when the map no longer matches the terrain, and going to look at the terrain directly even when the map is well drawn. AI isn’t arriving through formal strategy – it’s arriving through people. The leaders who get this right will be the ones who find out what’s already happening, and build from there.
That’s the discipline adaptive leadership gave me a language for, long before AI made the pattern this visible: protect the people who see the gap first, and give the work back to the ones who can actually close it.
© IE Insights.
