Most leaders' assessment of what AI can do for their business was formed somewhere between six and twelve months ago. It was based on a demo they saw, a chatbot they tried, or a vendor presentation they sat through. That assessment is now shaping decisions about hiring, technology investment, and competitive positioning. And for most, it is significantly wrong.
I have started calling this the glorified Google trap. It is the pattern where a leader forms a mental model of AI based on early or limited exposure, and that model then determines everything: what they try, what they ask for, what value they see, and what they recommend to their teams. The model feels like an informed assessment. It is not. It is a snapshot that froze the moment they stopped experimenting.
Recently, I sat with five leaders at the same professional services firm, same tools available, same level of seniority, and asked each to rate how reliable they thought AI was for their work. The answers ranged from 3 out of 10 to 9 out of 10. If it is not the technology, what is driving these differences?
The leader who rated it 3 had been using AI exactly the way you use a search engine: type a question, assess the answer, move on. She described her experience of AI as, essentially, a glorified Google search. She knew there had to be more to it but could not see what that was from inside her current way of using it. The leader who rated it 9 had been using AI to process financial data, draft communications, and build structured workflows. She had learned to provide context, iterate, and shape her requests. A third leader rated it 6-7 but acknowledged the number was based on what she had heard, not what she had tried. Her concern was not about AI itself but about her team losing the analytical skills she had spent years building.
Three leaders, three fundamentally different beliefs about what the same technology is, how it applies to their work, and what value it can deliver for their business. Each belief was rational given their experience, but it was also incomplete. And the incompleteness was invisible to the person holding it, creating a trap that is difficult to escape without intervention.
Every one of them shifted.
In a single structured session, the leader who saw AI as a glorified search engine started treating it as a professional she could brief with context, constraints, and goals. She realised that the interaction was no different from briefing a colleague: you provide background, objectives, and constraints. She had been skipping all of that because her mental model said it was a search engine. Her prediction of AI's reliability for her next task jumped from 3 to 9, not because the tool changed, but because her concept of it did. The sceptical leader discovered that AI was most useful for work she had not considered: not her technical domain, but structuring her own thinking as a leader. After working through a real management challenge, she reflected that AI had not told her anything she did not already know. It had helped her articulate what she could not have structured on her own.
In both cases, the mental model shifted. Not because anyone explained what AI can do, but because each leader encountered the gap between their model and the reality, on their own work, in a way that made sense to them.
This matters more now than it did six months ago.
Last week, Anthropic released Claude Opus 4.7, a model that can plan and execute multi-step tasks, maintain context across sessions, verify its own outputs, and work coherently for hours. Harvey, the legal AI platform used by more than half the top 100 global law firms, integrated it immediately, reporting that the model now correctly distinguishes between assignment provisions and change-of-control provisions in contracts, a task that has historically challenged frontier models. The same model can analyse a firm's financial data and surface patterns that would take a team days to find manually, or prepare a leader for a difficult conversation by drawing out the strategic considerations they had not yet articulated. This is not a better search engine. It is a different category of capability. A leader whose mental model says "AI is a search engine I check" will never encounter any of this, because their model does not generate the behaviour that would reveal it.
The research explains why the trap is so hard to escape from. Nielsen Norman Group found that people who use AI frequently are actually less likely to discover new capabilities than newcomers. They tested early, found limitations, and stopped pushing. Their confidence that they understood AI's boundaries was evidence-based. It was also out of date. The trap is not a failure of curiosity. It is the reasonable result of forming an assessment, finding it useful enough, and never revisiting it while the technology moved underneath.
This is what makes the glorified Google trap so expensive. A leader stuck in it does not just miss efficiency gains. They miss entire categories of value that their mental model makes invisible. They cannot see AI as something that augments their own executive judgment, because their model says it retrieves information. They cannot see it restructuring how their teams deliver work, because their model says it automates tasks. They cannot see it changing how their firm creates value, because their model says it is a productivity tool. The trap does not make leaders slower. It makes them blind to the strategic question altogether.
Five leaders, same firm, same tools. Reliability ratings from 3 to 9. The gap was not the technology. It was the belief each person held about what the technology is. And every one of them updated that belief in a single session, not by learning more about AI, but by using it on work that actually mattered.
The question worth sitting with is not whether AI is ready for your business. It is whether your picture of AI is current enough to make that judgement. And with each release, the cost of getting that answer wrong goes up.