The room had everything it needed. The AI tool was live. The licenses were paid. The training had been delivered. Leadership had sent the memo.
Nobody was using it the way it was intended. Six months in, usage had plateaued at a small fraction of the organization. The teams that had adopted it were the ones that would have found a way to move faster regardless. Everyone else had quietly returned to how they worked before.
I've been in that room more than once. Different organizations, different tools, same dynamic. And the question that eventually surfaces (if the leadership team is honest) is always the same: is this an AI problem, or is it something else?
It's something else. It's always something else.
What AI actually does to an organization
Not a technology explainer. A behavioral one.
AI compresses the time between a decision and its consequences. Assumptions that used to take months to test can now be tested in days. Gaps between what leadership believes is happening and what is actually happening become visible faster. Processes that worked because nobody could see how slow they were suddenly look very slow.
This is uncomfortable in a specific way. AI doesn't create organizational dysfunction. It makes existing dysfunction faster and more visible.
The organization that had unclear decision rights before the AI rollout still has unclear decision rights after. The team that couldn't agree on priorities hasn't resolved that disagreement. The manager who hoarded information because information was power hasn't stopped doing that. The AI has just removed the cover that slow processes used to provide.
What looks like an adoption problem is usually a clarity problem. What looks like a technology gap is usually a behavioral one.
Why organizations reach for the tool anyway
Buying a tool feels like taking action. It's measurable, budgetable, defensible. It produces a project plan, a rollout timeline, a utilization dashboard. It doesn't require anyone to admit they don't know something. It doesn't require anyone to change their status or their role.
"We have AI deployed across the organization" sounds like progress. It's a statement a leader can make to a board, to a press release, to themselves.
The harder observation: the same organizations struggling with AI adoption also had difficult agile rollouts. They had digital initiatives that produced a lot of activity and not much change. They had lean programs that ran for two years and then quietly stopped. The tool changed each time. The underlying dynamic didn't.
This isn't a failure of intelligence or ambition. It's a structural problem. Iansiti and Lakhani made this point clearly in Competing in the Age of AI: the bottleneck for most organizations is not the technology. It's the operating model. Organizations are optimized for stability. The same properties that make a large organization reliable (consistency, hierarchy, process adherence) also make it resistant to the kind of behavioral change that makes new tools actually work.
What actually changes organizations
This is the part that has to come from experience rather than theory. Across engagements in organizations ranging from global banks to government agencies to energy companies, a few patterns repeat.
The decision that nobody will name. In every stuck initiative, there is a decision that the organization is circling but not making. Not a technology decision. A people decision, or a structural one, or a strategic one that someone with authority is avoiding. Until that decision is surfaced and owned, everything downstream stalls. The AI rollout is downstream.
The person who already knows. There is almost always someone in the middle of the organization who understands exactly what's wrong and has been unable to say it clearly to anyone who can act on it. Finding that person, and giving them a path to be heard, is often the most valuable thing an outside advisor can do. The knowledge is already there. The channel isn't.
The thing that had to stop before the new thing could start. Real change requires removing something. A process, a meeting, a metric, a habit that is actively preventing the shift. Organizations that only add (new tools, new roles, new initiatives) without removing the things that contradict them tend to produce exhaustion, not progress.
The 90-day moment. There is a point, usually in the first 90 days of any genuine change effort, where it either becomes real to the people doing the work or it doesn't. If it doesn't land in that window, the initiative continues on paper and dies in practice. Executives move on to the next priority. Middle managers return to the patterns that kept them safe. The tools remain deployed and unused.
None of these patterns are about AI. They predate AI by decades. They will outlast whatever comes after it.
The trap of the current moment
The specific failure mode right now is this: leaders are treating AI adoption as a deployment problem when it's a change problem.
They're measuring tool usage, not behavior change. They're running training programs, not examining incentive structures. They're reporting on AI coverage to their boards, not on whether decisions are actually improving.
Some leaders know this and are using the AI rollout as cover for the change conversation they're not yet ready to have. The activity creates the appearance of momentum. The hard questions get deferred.
What this costs is not obvious immediately. It becomes obvious about 18 months in, when the results don't match the investment and someone needs to explain why.
A different set of questions
Not prescriptive steps. A reorientation.
What decision is not being made? Before deploying anything, name the organizational decision that the AI is supposed to improve. If you can't name it specifically, you're not ready to deploy at scale.
Who has something to lose from what this reveals? Every AI initiative surfaces something: a capability gap, a legacy process that shouldn't exist, a management layer that isn't adding value. Name who stands to be exposed. That's where the resistance will come from, and it's where the change work actually lives.
What has to change in how people work together, not just what tools they use? AI works in organizations where people can ask hard questions, disagree productively, and update their beliefs when they're wrong. Amy Edmondson's research on psychological safety shows this is not a culture accident. It's a leadership output. It fails in organizations where status depends on appearing certain. No tool fixes that. Leadership does.
The organizations that will actually benefit from AI are not the ones with the most sophisticated stack. They're the ones that have built the internal capacity to change: to surface difficult questions, own real decisions, and move when the evidence points somewhere uncomfortable.
That capacity is built by people. It's maintained by culture. It's enabled, occasionally, by technology.
AI in an organization that can change is a multiplier. AI in an organization that can't is just another expensive thing that didn't work.