You’ve just sat through a demo. The slides were clean, the examples crisp, and the person presenting clearly knew their material. Fifteen minutes later, back at your desk, something felt off. There was no mention of a specific problem it solved, no before-and-after, no number. Just capability, neatly presented.
That feeling is worth attending to. What you watched might have been genuine progress. For a founder who has delegated an AI mandate, the difference is not always obvious, and the stakes for getting the read wrong are higher than they appear.
What is AI theatre?
AI theatre is what you get when an initiative is judged on how it presents rather than what it produces. Tool counts, licence deployments, polished demos, usage reports, all held up as evidence that AI is working in the business, with no named business outcome attached. The term has been in active use in implementation planning circles since around 2023, and the failure mode it describes is well documented.
The signs are consistent across business types. The update sounds confident. The tools are real. The team has been busy. But when you ask what problem the work has solved, the answer gets vague quickly. That vagueness is the tell.
Part of what keeps the pattern alive is that AI vendors and internal teams both have reasons to show activity. Licence purchases are easy to track. Demos can be scheduled in an afternoon. Measurable business outcomes take time to emerge, sometimes 12 to 24 months by the estimates of implementation practitioners, and in the interim something has to go into the report. Activity fills the gap that outcomes haven’t closed yet.
Why does this matter for a founder?
A founder who has delegated an AI mandate does not have the time or the technical depth to audit every initiative. That is exactly why a confident update with nothing underneath it is the most dangerous kind. You accept it, the team relaxes, and the moment passes. The gap between what AI is supposedly doing and what your results show keeps growing.
BCG’s 2025 research found that AI usage across organisations was rising while measurable business impact remained flat. Analysis of AI implementation puts the pilot failure rate at around 95% of initiatives that never reach P&L impact. Adoption increases; outcomes do not follow at the same pace.
For a founder, the exposure goes beyond wasted budget. Research into AI risk disclosures found that reputational risk was the top AI concern for 38% of S&P 500 companies surveyed. Boards expect AI to produce something tangible. A founder who cannot distinguish a genuine AI update from a performative one will struggle to manage that expectation when the questions arrive.
Where will you actually meet it?
AI theatre shows up in three predictable places in a founder-led business. The internal demo presents a tool working well in a controlled setting, often on a task that was already handled adequately. The progress update arrives heavy on adoption figures and usage statistics, with no business outcome attached. The licence announcement presents a purchasing decision as evidence of AI commitment rather than as a means to a specific end.
The tell in each case is the same. Activity language has replaced outcome language. “We have onboarded thirty users” is an activity. “We have reduced invoice processing time by 40%, which has freed the finance team for two additional days per month” is an outcome. Both can be true at the same time. But the first cannot be offered as a substitute for the second.
The drift happens gradually, particularly in teams under board pressure to show something. A demo can be arranged in a week. A licence can be procured in a day. Confirming that an AI initiative has produced real business value typically takes months, and teams learn, sometimes without noticing, to report what is available rather than what is meaningful.
When to push back and when to let it pass?
A founder who challenges every AI update will exhaust the team and slow the work. The test is whether an update can name the business problem it addresses and explain how you would recognise when the work has solved it. If the answer is clear and specific, let it pass. If the update sounds confident but the answer goes vague, that is the moment to push.
A useful test comes from Addepar, an investment management platform whose executive team has written about how they evaluate AI initiatives before committing to them. The test asks whether an initiative would still matter if it did not use AI at all. If the answer is no, AI has become the point of the initiative rather than a means to a business end. If yes, the AI is a component, and someone on the team should be able to name the outcome it is working toward.
Three questions carry this into a real conversation without making it confrontational. The first is to ask what success would look like in concrete terms six months from now. Putting an outcome on the table does not imply the current work is wrong. The second is to ask what would actually be lost if the initiative paused for a month. The answer reveals whether the work addresses a live problem or fills a gap in the reporting. The third is to ask for the single metric being used to evaluate progress. One number, not a dashboard. The question forces specificity, and if there is no clean answer, the work has not been designed around an outcome.
What does real progress actually look like?
Genuine AI progress has two recognisable signatures. The team is tracking leading indicators, adoption rates, prompts run, time saved per task, alongside lagging indicators, revenue, margin, or capacity figures that matter to the business. High usage paired with flat business results is the classic AI theatre signature. Where progress is real, the two converge, and someone in the team can show you the specific work that moved the numbers.
A genuine update also names a specific outcome without being prompted. Consider a BD team that has cut proposal turnaround from five days to 48 hours because an AI drafting tool reduced first-draft time by 65%. The AI did the work. The business change is what the update makes visible. A founder receiving that report knows something real happened.
Tracking where a programme actually stands requires watching two different measures. Implementation guides distinguish between trending ROI, the early signals that an initiative is working, and realised ROI, the financial outcomes that confirm it. A healthy AI programme tracks both. A theatre programme tracks only the trending side and presents it as evidence of progress. Knowing the distinction gives a founder a practical filter for every update they receive.
The founder’s advantage here comes from asking outcome-focused questions and noticing when the answers go vague. AI theatre persists in owner-managed businesses because no one in the chain between the board and the team insists on connecting activity to outcomes. That insistence is a founder’s job. It requires no understanding of the underlying technology to apply.
