PPaint/Ikon Images
Dry rot. A Potemkin village. The Wizard of Oz.
What do those things have in common? In each case, they may look good on the surface, but it’s only an illusion. Wood afflicted with dry rot looks just fine until the tree it’s in topples down. Grigory Potemkin is said to have tried to impress Catherine the Great by creating a prosperous Crimean village constructed only of building facades, masking the real town’s crippling poverty. The Wizard of Oz looked great and powerful, until the cowering man behind the curtain was revealed.
Could artificial intelligence do the same to organizations — create an impressive, seamless exterior even as capability completely falls apart inside?
This alarming possibility was raised by the AI experts we interviewed for a joint Anthrome Insight-Axialent study on AI’s impact on behavior and culture inside organizations. Even as those experts uniformly cited AI’s potential to transform work for the better, they cautioned that AI adoption risks unintentionally creating an organizational mirage: workplaces that appear highly capable while people’s real skills quietly erode beneath polished AI-generated output.
Perhaps even more worrisome: When we can no longer reliably tell who truly knows what, the interpersonal trust that teams depend on to function also crumbles.
What are the early signs that the mirage effect is already forming in your organization — and what does it look like when it takes hold? How can leaders avoid the mirage effect and make sure that in the AI age, their people and organizations are as capable as they appear?
Capability Mirages: The Early Signs and Troubling Possibilities
An AI-driven illusion of competence, hiding the absence of genuine understanding, is already appearing in some organizations, the AI experts told us. Stephanie Antonian, founder and CEO of AI product development company Aestora, explained how this can play out in dangerous ways for leaders and organizations: “The upside [of AI tools] is that everyone can produce a level of work that’s pretty good for basic tasks. … It looks pretty good. But then you don’t know what’s underneath it, how resilient that piece of work is, or whether it’s going to give you an additional liability.”
After all, AI is not necessarily an improver of work but, rather, an amplifier. As AI Business Impact CEO Gábor Szórád observed, “At the end of the day, if you are a fantastic software engineer or a great manager, AI allows you to do more with the same energy. If you’re a bad one, you’re just going to create more crappy instructions, longer ones, more bad ideas. It just magnifies whatever you put in.”
The tech vendor marketing narrative stating that AI makes individuals more capable just compounds the mirage issue. In some situations, AI does enable the production of better short-term outputs — but it also breaks the historical link between strong output and strong capability, causing teams to lose sight of who actually possesses strong skills.
Moreover, AI itself doesn’t know when it’s out of its depth, said Amir Michael, professor of accounting and deputy executive dean for executive and professional education at Durham University. Likewise, Antonian observed, people who aren’t capable in a particular subject area can’t spot where AI-generated work has gone wrong — unlike people with subject-matter expertise. They may pass on bad output to others who also can’t tell the difference.
The mechanism through which professional mastery has always been built — productive struggle, error-based learning, the slow accumulation of genuine judgment — may be quietly bypassed before many people realize what is being lost.
But there is a second, less visible problem that, in the long run, may be the more dangerous one: Not everyone is self-aware enough to notice their own skills eroding. There is no single individual — among managers, colleagues, or clients — who can do an accurate, real-time read on how capable an organization is, overall. Individual skills and collective capability could weaken long before anyone notices, and there is absolutely no guarantee that even the best-functioning AI could begin to fill the gap. Leaders should be especially concerned about losses in the sophisticated, critically necessary “muscle of critical thinking,” said Albert Durig, cofounder and partner at Triviam Consulting.
This situation results in a direct and damaging consequence for organizational trust. Teams have traditionally functioned based on knowing, or at least being able to calibrate, who knows what. That calibration determines whose judgment should be relied upon, how managers identify who is ready for greater responsibility, and how organizations know what they are truly good at. When AI makes that calibration unreliable, trust erodes with it. When AI-assisted work is later discovered to have been misrepresented, the trust collapse tends to be swift: “The impact on trust is 0 [doubt] to 100,” said Elisa Farri, vice president at Capgemini Invent Management Lab.
Concerned yet? We all should be.
But if leaders act now, they can avoid organizational dry rot — and maintain true capability fueled by humans and machines alike.
How to Preserve Human Capability and Team Trust
Let’s explore what the experts we spoke with had to say.
1. Choose purpose and business goals over an AI-first mentality.
It’s trendy to announce that your company is thinking “AI first.” But it’s not what these experts would recommend. Antonian put it directly: “When you go AI-first, you have already told your organization it’s not human-first.” That signal, once received, is hard to unsend — and at a moment when people are already anxious about their relevance and job security, it can quietly erode the trust that makes teams function.
One potential outcome of an AI-first future is an unpleasant inversion of the roles of human and machine, said one senior AI executive at a Fortune 500 company. They see a concrete risk that people might let AI do the reasoning, interpreting, and responding only to become “transactional tools” themselves.
“AI can be the lead,” Michael noted. “That’s the problem. As long as AI is your follower — it follows your requests, your orders — we’re fine. The time that AI jumps to be your lead, that’s the downturn.”
AI-first thinking, in the view of our experts, causes people to lose perspective on the utility of AI as a tool — and to deploy AI in comically inappropriate settings. Remember the old saw “If you’re a hammer, everything looks like a nail”? That applies to AI-first: “You don’t walk around the house, holding the biggest drill that you have, asking people if they need their coffee stirred,” said Andrea Jones-Rooy, a data scientist, organizational researcher, and visiting associate professor at New York University’s Center for Data Science.
Even when AI is used for more seemingly appropriate ends, such as measurement and KPIs, Durig noted, it can cause a dominance of measurement over meaning. That leads to “a performance culture without purpose,” he said.
On the flip side, when purpose comes first, enabled by AI tools, our experts see the potential for true progress and even stark disruption. With AI enablement, “Small groups of people that get together for a specific purpose may outperform corporations because they are more nimble, flexible, fast-moving,” said AI entrepreneur Thierry Kahane. The key, in Kahane’s vision, is cohesion around a goal versus a technology.
What concrete leadership steps fuel this mission? First, you should check your own rhetoric. If discussion of AI is eclipsing dialogue around business outcomes, the conversation is framed incorrectly; people are more likely to become passive, let their skills slip, and quietly lose faith in their own relevance. Similarly, if the only voices you’re hearing in the AI conversation are those of people who are passionate about the technology, the balance of business-purpose versus tool is likely off.
Finally, it’s critical to routinely audit how people are operating AI tools on the ground. Is AI making decisions that humans should be making? To ensure that an organization is operating “business first,” leaders and teams must exercise constant vigilance around day-to-day AI use. Otherwise, human capability is destined to slide … and we won’t know until a black swan event happens that AI, trained on typical data, is ill-equipped to handle it.
2. Leaders should model AI usage specifics.
If you don’t want people to switch off their judgment, critical thinking, and intrinsic motivation, you must teach them how to engage with AI as a sparring partner rather than as a delegation tool, Farri said. People need to engage in active, back-and-forth interaction with AI —interrogating its suggestions, pushing back, and cocreating outputs.
And this behavior needs to start at the top. When leaders model active, curious, judgment-led AI use and do so visibly, this signals what the organization actually values far more powerfully than any guidance document can, Szórád said. “The project sponsor needs to be the CEO. The CEO needs to use AI daily,” he said.
That’s important advice at a time when many organizations have given employees only the barest clarity on how to utilize AI day to day. This may be well intentioned on those organizations’ parts; perhaps their leaders don’t want to stifle employee creativity. But thoughtful guidance on AI usage can walk the tightrope of specifying behavior without shutting down exploration. For example, organizations can distribute highly generic but intelligently framed prompts — “When I say _____, what am I not thinking of that I should consider?” — that keep the human in the driver’s seat.
The AI realm is still new enough that people genuinely need to be steered away from the wrong behaviors: Telling people what not to do is just as important as telling them what to do, Farri said. Caveats must be clearly communicated, she added. For example, you might say “The more flawless an AI output looks, the more ruthlessly you should stress-test it.”
3. Build human capabilities first, AI augmentation skills second.
MIT Sloan School of Management postdoctoral researcher Isabella Loaiza offered a deceptively simple principle that organizations are widely ignoring: “You need to learn first and then use a tool to supercharge your abilities.” One executive asked us to imagine what could happen when the sequence is reversed: “When someone with 20 years of professional experience uses AI to amplify their impact, that works. But what happens when a 20-year-old’s first interaction with work involves AI from day one? How do we develop that person at the same speed and depth? We will create a talent gap that will be hard to close.”
This has direct implications for how organizations design onboarding, early-career development, and role progression. One question you should ask now is not “How can we use AI to accelerate this person’s output?” but “What does this person need to genuinely understand before AI can help them go further?”
Performance measurement systems need to take this phenomenon into account too — not just asking “Is the work good?” but “Does this person understand it, and are they growing through producing it?” Measuring human-centered outcomes requires human-centered metrics, which most organizations have not yet built.
4. Restore accountability, and use it to rebuild trust.
Accountability is challenging in the best of times. Knowledge work, particularly, has long been slippery to ascribe: That’s due to both messy over-collaboration and, frankly, some bad, illegitimate-credit-taking behavior within teams. When you don’t know who does what, it’s hard to hold anyone accountable. Add in AI, and the accountability muddle gets even worse, with direct consequences for both skills and trust.
When a team member takes credit for work that an AI clearly performed, intra-team trust is naturally eroded, several people commented. But worse — and central to the issue of AI mirages — if a team is not sure how work is getting performed (that is, what mix of AI and human capability, exactly, is being deployed), it quickly becomes impossible for colleagues to understand each other’s skills. Without that visibility, the natural ways that teams maintain their collective skills will break down. More seasoned team members and managers won’t know who to coach.
The trust consequences extend further than most organizations realize. Kahane described a dynamic that he said is almost universal: employees deliberately concealing their use of AI to ascribe productivity gains to their own efforts — thus protecting manager and peer perceptions of their human performance. (Researchers are beginning to see this dynamic in academic settings as well. Students are concealing their AI use due to perceived taboos.
Daniel Strode, a professor at the IE School of Human Sciences and Technology, pointed out the vicious cycle this could create: Employees hide AI use; leadership expects efficiency gains that don’t materialize; neither side communicates effectively; and AI initiatives collapse under the weight of accumulated mistrust. The irony is sharp: The very tool meant to help organizations perform better becomes the source of the opacity that prevents them from understanding how they are actually performing.
Now you might say, “Shouldn’t that be OK? If the AI can do the work capably, who cares?” The issue, though, is that the AI can do the work capably until it cannot. And as previously noted, neither the humans involved nor the AI know those specific capability limits, which are masked by the polish of AI outputs. Losing early-warning signs of capability gaps only makes the eventual emergencies more dire. The solution here can be classified as “simple but not easy.” Cedric Wells, head of IT innovation and new technologies at Gorilla Glue, framed it practically: “Set clear ownership and verification norms.” Who produced the work? What role did AI play? Who is accountable if it is wrong?
Today’s AI technology has raised important questions about the basics of how we get work done — especially around who (human or technology) has the capability to do what. Individual-skills erosion, and collective capability erosion, could damage organizations irreparably. If AI tools are both contributing to human deskilling and concealing that fact, they could speed the collapse of many organizations. The experts we spoke to were alarmed, but they were also hopeful that leaders will step up to the challenges.
If leaders coach teams to use AI thoughtfully, accountably, and with real purpose, the mirage could become the reality: Organizational capability could meaningfully increase and human skills could grow too. We can be our best selves if we lead technology and are not led by it.