Ilusions about AI Adoption at Firms - Maya or not?
Why Organizational AI Transformation May Mostly Be Performance Art
If I were to walk into a senior executive-level meeting at a firm today (May 2026), I would likely hear the same speech and in the same tone. AI strategy. AI roadmap. AI readiness. AI maturity. Pilots. Platforms. Foundational models. The PPT slide will be clean, and the vendor logos will be in the right places. And I will not be surprised if on page eleven, there is an arrow that bends sharply upward and points at “the future.”
Now, if I ask the everyday manager or even the executive, the harder question. What decision in your company is genuinely better this quarter than last? Which customer is being served in a new way? Which workflow has been rebuilt, not just re-skinned? The room is likely to go quiet.
This is not because leaders are lazy or dishonest. It is because most of what we now call “AI transformation” is what the Advaita tradition (an ancient branch of Indian Philosophy) calls Maya. A convincing appearance of something real. Not the real thing itself. The performance is sincere. The transformation is mostly a mirage.
I am writing this article to walk through why it is happening and what to do about it. As a psychologist who studies leadership and a researcher who works on AI in HR, I keep seeing the same “six, seven” illusions play out across companies on three continents. They are not random. They are built into how we are approaching this. And they are quietly costing firms the very advantage they think they are buying.
Illusion 1: Adoption metrics are not adoption.
The Hawthorne studies of the 1920s taught us that people behave differently when they know they are being watched. Modern AI dashboards have taken that idea and put it on every screen in the office. License counts go up. Prompt volumes go up. Vendor revenue goes up. Boards see green dashboards. Analysts raise their ratings. Underneath all of that, market surveys show the same comfortable pattern. A small group of firms has moved meaningfully along the maturity curve. The rest are stuck in what the data politely calls “emerging” and what I would, less politely, call dressed-up standing still. The signal is loud. The transformation is faint. This is not really a measurement problem. We are measuring the costume, not the actor.
Illusion 2: The ideation bubble
Here is the part that should worry every innovation officer.
When teams use mainstream LLMs and search engines to generate ideas for proposals or new initiatives at work, they may end up in what are known as “ideation bubbles”. Dense clusters of similar, conventional ideas. The tools are built to surface the most likely content, which is a polite way of saying the most obvious content.
In plain psychology terms, the algorithm has taken our laziest mental habits and given them a beautiful interface. We feel more creative because the output is louder. We are actually less creative because the range of what we produce has shrunk.
Now picture every competitor in your industry using similar models, trained on overlapping data, drawn from the same corners of the internet. Standing out gets harder. And the system itself is quietly doing the squeezing. The end state is a quiet, well-lit plateau of sameness. An economist would call it a mediocre equilibrium. I would call it a flat line in a nice font.
The fix is counterintuitive. Hack the tool against itself. Force odd prompts. Pull analogies from fields outside your industry. Treat the LLM as a sparring partner whose first answer you should usually refuse. If everyone is fishing in the same pond, your edge is in the rivers nobody else is wading into.
Illusion 3: You are chasing the wrong dragon.
Most AI roadmaps I read are fixated on two things. Foundation models and copilots. These are the showy, conference-friendly, easy-to-explain parts of the AI moment. For most firms, they are also the least distinctive and the least profitable use of AI money. The unsexy truth, which practitioners who have actually shipped systems with real P&L impact will tell you in private, is that domain-specific personalization and recommendation systems are still among the most profitable and most underused AI applications. They quietly move top-line numbers by margins any CFO would notice. They do not get written up in The Economist. They compound.
Worse, the dominant story confuses AI with AI-driven business model innovation. The original disruption theory is clear on this point. Technologies are not disruptive on their own. Business models are.
The firms that “reimagine” and “reinvent” their way of doing business, their operating model, workflows, and organizational design — the ones who start from a blank slate rather than just taming their way into transformation — are more likely to reshape industries. They are the ones who use AI economics to design offerings that big, established players initially find unattractive. Too cheap. Too narrow. Too niche. Until those offerings own the market.
So, ask the uncomfortable question. Is your AI strategy building a new value network, altering your operational model and workflows, or is it making your existing one slightly more efficient? One builds a moat. The other deepens a puddle.
Illusion 4: The quiet hollowing of judgment
This is the illusion that bothers me most as a psychologist. AI decision-support is sold as a way to make your people sharper. In practice, in many firms, it works the other way around. It quietly takes the thinking out of their hands. Frontline staff are asked to defer to scores, suggestions, and confidence intervals from systems they cannot open up and inspect. Mid-level judgment, which is the connective tissue of how organizations learn, slowly weakens.
Self-determination theory tells us that autonomy, competence, and relatedness are not nice extras. They are the foundation of motivation.
When an AI system quietly encodes the firm’s risk preferences and dresses them up as neutral analytics, you are not giving your people more power. You are concentrating control at the top while wearing down the psychological capital that makes the organization resilient.
The org chart looks the same. The actual place where decisions are made has shifted toward a model that no one fully governs. This is not a tooling problem. It is a governance problem in disguise.
Illusion 5: It is not a tech project. It never was.
If I could rewrite one slide in every AI strategy deck, it would be the one that puts the program under the CTO. AI transformation is not a tech project with change management bolted on at the end. It is a redesign of how people and machines work together, with technology added near the end. The research on adoption work is detailed on this.
What holds firms back is not capability. It is what economists call an adjustment cost. Roles, workflows, incentives, identities, accountability structures, and the political question of who decides what. Firms that refuse to redesign those things will not get AI value, no matter how impressive the model.
I have started telling clients something blunt. AI is a re-negotiation of the social contract inside your firm. If your program is not led by people who can hold that conversation and live through it, you do not have an AI program. You have a procurement event.
What to actually do
Let me close with three things to do, because abstract ideas without action are their own kind of theater.
First, separate signal from substance in your own dashboards. Stop reporting adoption. Start reporting decisions changed, cycle times altered, and revenues you can trace back to a system that did not exist eighteen months ago. If the answer is uncomfortable, that is the point.
Second, deliberately design against algorithmic sameness. Build ideation routines that force people to diverge. Use AI to generate the consensus answer. Then ask a human to argue the opposite. Variance is the new alpha.
Third, move the AI mandate from pure tech to organizational design. Put it under people who can credibly redesign roles, governance, and value flows. Treat models as the last decision in the chain, not the first.
The firms that will look like winners ten years from now are not the ones with the loudest AI announcements today. They are the ones quietly doing the unsexy, structural, deeply human work that no vendor sells and no analyst rewards.
Maya is convincing. Just remember what the word actually means.