Stop Performing AI — Start Actually Using It

There's a strange new performance art sweeping through offices worldwide: people pretending to use AI.

A June 2025 Howdy.com survey found that 16% of U.S. workers admit to faking AI usage to meet employer expectations. Another 22% feel pressured to use AI tools even when they're not sure the tools actually help. Meanwhile, nearly half of enterprise leaders now track how often employees open AI tools as a formal performance metric — not whether those tools produce better work, just whether people click on them.

Welcome to AI Theater: the growing gap between the appearance of AI adoption and the reality of it.

I spent the past several months researching this phenomenon — digging into studies from Harvard Business School, Stanford, BCG, IBM, and others — and what I found is both alarming and clarifying. The biggest risk most knowledge workers face today isn't that AI will replace them. It's that institutional panic about AI will drive them into performative adoption patterns that waste time, leak data, and burn them out.

The Fear Engine

The roots of AI Theater trace directly to fear. The IBM CEO Study (2025), surveying 2,000 chief executives, found that most AI spending is driven by FOMO — not by demonstrated return on investment. Only 25% of AI initiatives delivered expected ROI, yet investment is expected to more than double because leaders are terrified of falling behind. The ABBYY AI Trust Barometer put it bluntly: 63–64% of executives say fear of competitive disadvantage is their primary driver for AI investment.

This fear cascades downward. Sixty percent of executives in a 2024 Forbes survey planned layoffs, with "inability to keep up with AI" cited as a key reason. Seventy-five percent of CEOs fear their companies will fail within five years without AI. When your boss's boss is making decisions from a place of existential dread, it's no surprise that the workplace response is more about optics than outcomes.

The Three Faces of AI Theater

The performativity crisis manifests in three dangerous ways.

First, measurement theater. When leaders evaluate employees on AI tool frequency rather than AI-generated value, they create perverse incentives. People optimize for appearances. They open ChatGPT, paste in something trivial, and check the box. The work doesn't get better — but the dashboard turns green.

Second, shadow AI. When management mandates AI usage but fails to provide secure, approved tools, employees improvise. They paste confidential data into free-tier AI tools on personal accounts. IBM's 2025 Cost of a Data Breach Report found that shadow AI now accounts for 20% of all reported breaches, adding roughly $670,000 to average breach costs. The mandate creates the very security disaster it was supposed to prevent.

Third, burnout spiral. Here's the statistic that stopped me cold: the Upwork Research Institute found that 88% of highly productive AI users report severe burnout — nearly double the rate of less frequent users. These power users are twice as likely to consider quitting. Unlike social media FOMO (fear of missing others' exciting lives), AI FOMO is directed inward. When you see what AI can do, you don't just fear competitors — you confront the terrifying gap of your own unrealized potential. The ceiling becomes infinite, and the pressure never lets up.

The Productivity Paradox Nobody Talks About

Beneath the Theater is a real and well-documented paradox. AI genuinely boosts individual task performance — the landmark BCG-Harvard experiment with 758 consultants showed up to 40% quality improvement on well-suited tasks. But across studies, 77% of employees say AI has actually added to their workload, not reduced it. Time is consumed reviewing AI outputs, learning new tools, correcting mistakes, and managing the cognitive load of deciding when to trust or override the machine.

This is the AI productivity paradox: task-level gains get absorbed by increased work intensity, cognitive overhead, and the organizational failure to redesign workflows around the new technology. Companies layer AI onto existing processes and wonder why people are exhausted.

Cyborgs, Centaurs, and Self-Automators

So how should humans actually work with AI? A fascinating 2025 Harvard Business School working paper identifies three collaboration species:

Cyborgs blur the boundary completely — AI is woven into every step of their work. They develop powerful new skills but risk propagating AI errors when oversight slips.

Centaurs deliberately divide labor: they handle the strategic, creative, and judgment-intensive work themselves, and delegate bounded, routine tasks to AI. In empirical tests, Centaurs achieve the highest accuracy.

Self-Automators hand everything to AI and walk away. They save the most time but produce the worst quality and develop no new skills.

The research is clear: be a Centaur, not a Self-Automator. The winning pattern isn't "use AI for everything" — it's knowing precisely when to use it and when to trust your own judgment.

What Actually Matters Now

If AI Theater is the disease, what's the cure? I've been calling it anti-theater competence — the ability to distinguish genuine cognitive leverage from symbolic technology adoption. It comes down to four things:

1. Start messy. Stop waiting for the perfect AI tool, the perfect training program, or the perfect use case. Research shows that when management explicitly encourages AI experimentation — even without formal training — adoption jumps from ~15% to 55%. The antidote to perfectionism paralysis is permission to produce a forty-out-of-a-hundred first draft and iterate from there.

2. Be a Centaur by default. Don't automate everything. Don't avoid everything. Deliberately choose which tasks benefit from AI delegation and which require your unaugmented human judgment. Review AI output critically. Say no to AI when it makes your work worse.

3. Learn context engineering, not just prompting. The real skill isn't writing clever prompts — it's architecting the entire informational environment the AI operates in. That means curating knowledge bases, structuring retrieval systems, and articulating your intent, values, and constraints clearly. This is a learnable discipline, and it's where the competitive advantage is heading.

4. Measure outcomes, not activity. If your organization tracks how often people use AI but not whether AI-augmented work is actually better, faster, or more accurate — you're running a theater, not a strategy. Push for metrics tied to decision quality, error reduction, and genuine business impact.

The Broken Rung We Can't Ignore

One more thing that keeps me up at night: a 2025 Stanford study found that employment for young software developers (ages 22–25) dropped roughly 20% from its 2022 peak. Entry-level IT roles in the UK declined 46% in a single year. If junior workers can't get their foot in the door because AI handles the "grunt work" they would have learned from, we sever the pipeline of expertise development. AI amplifies expertise — it doesn't create it from nothing. This "broken rung" problem demands serious policy attention before it becomes irreversible.

The Bottom Line

The old map — built on industrial-era specialization, passive news consumption, and fear-driven compliance — doesn't work anymore. The organizations and individuals who will thrive aren't the ones adopting AI fastest or loudest. They're the ones adopting it most authentically: with clear-eyed assessment of what works, honest measurement of what doesn't, and the discipline to resist performing enthusiasm they don't feel.

Stop performing AI. Start actually using it.

This post is based on my longer research article, "Discarding the Old Map: Cognitive Leverage, Workplace Evolution, and the Anti-Theater Effect in the Age of AI," which includes full citations from Harvard Business School, Stanford, BCG, IBM, McKinsey, Pew Research, and other sources.

Download Research PDF