Busy, Performative, and Going Nowhere: The Anatomy of “AI Theater”

The C-level or boardroom meets every single week with great confidence. There’s a slide deck; it was ten slides in Q1, and now it has ballooned to twenty. There’s a use-case matrix, color-coded by department. There’s a Copilot license rollout process that the IT department is incredibly proud of.

Someone from Legal asks about governance.

Someone from Finance asks about the return on investment (ROI).

Everyone nods their heads in approval. The meeting ends exactly on time.

And what’s actually live, meaning in production? A RAG (Retrieval-Augmented Generation) demo that a mere three people use, and a pile of Copilot licenses that nobody is measuring.

If this sounds familiar, you’re not alone. In fact, that is precisely the core of the problem: you are not alone. A striking majority of companies calling themselves “AI-powered” today are playing a part in some version of this exact theater.

We’ve seen this before

Every organization has fallen into the trap of rewarding the appearance of being busy rather than actual delivery. It’s the comfortable illusion of a packed calendar where being constantly occupied sneaks in and takes over actually being productive.

AI Theater is simply the modern, multi-million-dollar upgrade of this old habit.

It looks great on paper. Budgets get approved, fancy pilot projects are announced at all-hands meetings, everyone is assigned to a new project and sometimes press releases about “AI transformation” are even published. It keeps the CEO and the board feeling safe because everyone is working on it. Yet, nothing has actually been put into production. It’s busy. It’s performative. And it is fundamentally going nowhere.

The show is real, but the results aren’t

A 2025 MIT study on enterprise AI adoption put a hard number on how common this reality actually is, drawing from dozens of executive interviews and an analysis of hundreds of public AI deployments: Only about 5% of AI pilot programs achieve rapid revenue acceleration. The vast majority grind to a halt and deliver almost no measurable impact on the profit and loss statement. In other words, that board committee isn’t an anomaly. It pretty much represents the median.

This pattern reveals itself with the same clarity when you drop down a level to what people are actually doing during their day. The same research uncovered a rapidly growing “shadow AI” economy: In over 90% of firms, employees are using personal AI tools on the side, even when the officially sanctioned pilot project has failed. Think about that for a second. The theater isn’t a technology problem. This is, in the most literal sense, a production problem.

The structural flaws behind the curtain

Once you can name AI Theater, the natural next question is what’s actually holding the curtain up. AI Theater is a symptom, the visible, boardroom-facing symptom of something we call Non-Adoption: a company that has genuinely tried AI, spent the budget, run the pilots, and still hasn’t put anything into production at scale. Most of the mid-market is living here right now, and it stems from highly consistent structural flaws across companies. Think of it less like an unsolvable mystery and more like the components of a poorly designed mechanism.

1. The foundational error: Platform trap.Teams try to build a massive infrastructure before they ship a single real use case. “First we need the right stack, the right models, the right governance layer, the right data architecture.” Six months later, zero use cases are in production, and the CEO is asking what the money actually bought. It’s the equivalent of setting up a factory and trying to spin the assembly lines before designing a single product to manufacture. A company doesn’t need a massive AI platform. It needs a shipping pattern. The platform, if it emerges at all, should emerge from working use cases. The process never works from the top down.
2. The coordination error: Fragmented architecture.Every department independently wants its own AI toy. Sales wants a deal-desk assistant, support wants a ticket summarizer, engineering wants a code reviewer, marketing wants, well, exactly this. Ten half-built experiments are in flight, none of them finished, each pulling in its own direction with no shared architecture connecting them. Picture ten understudies rehearsing ten different plays in ten different rooms, with no opening night scheduled for any of them. Everything feels like it’s “in progress,” but nothing ever gets finished.
3. Spinning the wheels: Features vs. Systems.This is the one that does the most damage, because from the outside it still looks like the wheels are turning. Most AI initiatives bolt on temporary features: a chatbot here, a summarizer there, a Copilot seat for the developers. These features look great in the shop window, but they don’t lift the operational load of the organization. A system is different. It’s what you get when AI is actually woven into how work flows through the company, compounding on your own data and your own workflows, getting a little better every quarter instead of staying static. Features are rented. Systems are owned. Right now, most companies are paying rent to tools that are compounding on those platforms’ AI capabilities, not on the company’s own internal processes.

The scale of this confusion isn’t small. Gartner expects over 40% of agentic AI projects to be canceled by the end of 2027, due to escalating costs, unclear business value, or inadequate risk controls, and has started calling out “agent washing” (the rebranding of ordinary chatbots and basic scripts as agents) noting that only around 130 of the thousands of vendors calling themselves agentic AI companies actually meet the bar. A lot of what gets bought as a system is, on inspection, just a simple feature wearing a system’s name tag.

Bring together a weak foundation, completely disconnected departments, and a framework that doesn’t touch actual business workflows, and the result is exactly what we started with: a highly active, very modern-looking organization that cannot actually move. That is the exact reality hidden behind a steering committee slide deck.

Why this isn’t a competence problem

Here’s the part worth sitting with. None of these three failures come from a lack of talent or effort. They come from a set of understandable, well-intentioned decisions, made in the wrong order. Wanting the infrastructure to be perfect before you build on it feels responsible. Letting every department pursue its own AI project feels like creating an environment of empowerment. Shipping a visible feature feels like progress, because it is right in front of everyone’s eyes.

That’s exactly what makes AI Theater so persistent. It’s built out of decisions that look correct in isolation. The failure only shows up in aggregate, a year later, when someone asks what actually shipped and the honest answer is a demo, a license, and a very good slide deck.

Where you actually are matters more than you think

Here’s the part that should create urgency rather than despair. Where a company sits on the AI adoption curve in 2026 is roughly where it will sit in 2029. Every quarter, more of the work inside a software business gets done by AI instead of people: first chat, then assistants, then agents, then increasingly autonomous workflows. Companies that are shipping and compounding today, accumulating their own AI capability one production use case at a time, are not going to be a little ahead of the pack in three years. They’re going to be structurally ahead, and that’s not a gap you close later by simply buying a better tool.

That’s what makes AI Theater dangerous. It feels like motion. It generates enough activity to satisfy a board update. But the clock underneath it keeps running exactly the same as it does for the company doing nothing at all.

Exiting the theater

None of this is an argument against AI. Yet, it is an argument against pretending. The fix isn’t a bigger platform or a longer roadmap. It’s picking one real use case, shaping it to your actual data and workflows, building it with engineers rather than consultants who hand you a deck, and proving business value before you scale to the next one.

Ship one thing.

Prove it moved a real number. Then ship the next.

That’s a fundamentally different posture than “let’s run a pilot and see.” It also happens to be the only posture that produces something you can point to a year from now, besides a very tidy slide deck.

The organizations still running AI Theater in 2029 won’t be the ones lacking ambition. They’ll be the ones that never noticed the AI theater show. The first step out is simply naming what’s happening, and then refusing to mistake motion for progress.

If your organization is somewhere between “we have a pilot” and “we’re not sure it’s actually working,” that’s a familiar starting point and there’s a way through it that doesn’t start with rebuilding your entire stack. Talk to Nebul.

*Sources

(1) The GenAI Divide: State of AI in Business 2025, MIT NANDA (Networked Agents And Decentralized Architecture), July 2025; as reported in Fortune, “MIT report: 95% of generative AI pilots at companies are failing,” August 2025.

(2) “Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027,” Gartner, Inc. press release, June 25, 2025.