By now, I’ve attended enough AI steering committee meetings to spot a trend. There is always one person in the room (usually the one who is most familiar with the tools) who ultimately decides how quickly the rest in the room can move. This decision is made discreetly and without malice. They rarely intend to gatekeep anything. To put this simply, everyone else has mastered the art of waiting for their approval before acting.
There is more to that pattern than just one meeting. According to a 2026 study by Gartner, only 20% of businesses think their employees are truly prepared for AI. The majority of discussions on that gap typically revolve around a skills narrative: insufficient time, training, or tools. A more subtle issue that lies behind the skills gap is overlooked; it has less to do with people’s knowledge and more to do with how important they have begun to feel.
Contribution and Confidence Are Two Different Things
People are usually bad at evaluating their individual contribution to a team effort, and this is a rather consistent problem. When you ask your teammates to independently calculate the amount of a shared project that each of them personally contributed, the results frequently sum up to well over 100%. It has nothing to do with honesty and everything to do with how memory functions: everyone recalls their own effort in vivid detail, and everyone else’s in outline.
When AI is added, the effect becomes even more noticeable. Those who are the fastest to pick up new skills also develop a sense of control over the outcomes those instruments affect. This tendency is called psychological ownership. When applied sparingly, this ownership is exactly what you want: it’s what makes someone genuinely care about a project’s success rather than just the number of tickets. But if you go a step further, the same motivation that protects excellent work also protects one’s involvement in its production.
What the Numbers Actually Show
We ran a survey among 2,000 professionals working with AI at Sombra, since we wanted to see how widely that instinct had spread. And the scale was surprising even to us. 79% of respondents said they believed they were a key driver of innovation at their organization: 37% “definitely,” and another 42% “a lot.” 64% went even further and said they thought that if they personally left tomorrow, innovation at their company would slow down.
We named this gap between that self-image and what people actually deliver “AIrrogance”. The term might be a little deceptive because no one in our data was bragging. What we were really picking up was more subdued: a personal belief that the work depends on you specifically, rather than on your team simply having someone with your skills. According to other coverage of the findings, creativity appears to have an ego problem.
What followed was an uncomfortable finding. Turns out that the people most convinced they were driving innovation were the ones slowing it down. 22% of respondents had pushed back AI projects because they didn’t see enough potential in them. 21% cited data that wasn’t available in time. 17% acknowledged slowing down a project without ever officially blocking it.
Also, 70% of businesses in the entire sample said they have experienced this form of internal resistance. This rarely feels like sabotage from the inside. Even if it pulls on the very innovation they think they’re carrying, it feels like caution, and someone being careful with something they truly care about.
The Bus Factor Problem
Engineering teams have a name for the risk of depending too heavily on one person: the bus factor, the number of people who’d need to disappear before a project stalls. Even if the one person keeping things together is a superb worker, a manager should be concerned by a bus factor. Most teams are aware that they need to keep an eye out for it in documentation and code. But almost no one applies the same reasoning to conviction, to the idea that a project is only progressing because one individual consistently determines that it is worthwhile.
AIrrogance provides its own version of that risk. But instead of tribal wisdom, conviction is concentrated in one person. Even with the best of intentions, a single strong opinion silently becomes the gate that every initiative must go through.
What I’d Actually Do About It
All of this does not preclude those who are slowing things down from voicing concerns. A faulty AI rollout should be slowed down by good pushback that is based on truly missing data or genuinely weak evidence of usefulness. When caution ceases differentiating between “this specific implementation has a real flaw” and “I’m not yet convinced, and my not being convinced is reason enough,” — this is where problems arise.
The teams I work with have found a few ways to distinguish between these two. Before they pause a project, it’s good to make complaints clear and verifiable by asking what specific proof would address this issue and when. A bench full of missing data should be given more weight than an ambiguous “I don’t see the value yet” statement.
Second, don’t always hand the domain expert role to whoever is objectively best at it. We rotate it deliberately at Sombra, even when it means putting someone less experienced in the lead. It costs us a bit of speed on any single project, but it means we’re never one resignation away from losing the only person who understood how something worked.
Third, take the person away from the checkpoint. If an AI project truly needs the approval of one expert before moving forward, note why and include the name of another person who may reasonably provide that approval within a quarter. If you can’t think of a second name, you’re probably experiencing the good old bus factor problem.
And fourth. It’s the test I actually use on myself: think of the last big idea at your company that wasn’t yours, and ask how much you did to help it happen. If you have to think for more than a few seconds, sit with that. The strongest innovation cultures I’ve seen tend to have a lot more shared responsibility and fewer prominent individual contributors. These contributors are dispersed widely enough that no one individual, no matter how competent, becomes a prerequisite for progress.
People will keep reading Gartner’s 20% figure as a training gap, and training is part of it. But I’ve watched enough rollouts now to know the harder fix has less to do with teaching more people to use the tools faster, and more to do with getting your fastest learners to notice when their own progress has stopped standing in for everyone else’s.
Credit: Source link
