Entrepreneur Evidence Theatre

Is there a customer who needs it, will they pay for it, and is there a business worth building around it?

Five years ago, if an entrepreneur walked into an Angel meeting with a functioning product, it told you something. Someone had invested considerable time, money or technical skill to build it. The product wasn’t proof there was a market, but it represented substantial work and commitment.

Today, an entrepreneur can arrive with a polished application built over a weekend.

The two products may look much the same. What sits behind them can be very different.

That distinction came to mind reading The Year AI Came For Us: Teaching Entrepreneurship Will Never Be The Same, an account of what happened when generative AI met the Lean LaunchPad entrepreneurship class. We were very familiar with the Lean LaunchPad approach from our decade living in California and our involvement with the TechCoast Angels, where the discipline of customer discovery, testing assumptions and finding product-market fit was part of the language of startups and early-stage investing.

For 15 years, the Lean LaunchPad followed a consistent model. Students began with hypotheses about a problem, a product and potential customers, then got out of the classroom and tested those assumptions. Over ten weeks, teams might conduct more than 100 interviews, reporting each week what they thought, what they did, what they learned and what they would test next.

The product evolved alongside the evidence.

In the Spring 2026 class, that sequence changed. Teams that once would have arrived with slides, sketches or wireframes turned up with working apps, digital twins and sophisticated prototypes. Products that previously might have appeared near the end of the course were there on day one.

Initially, the instructors saw this as progress. If students could reach the prototype stage immediately, perhaps they could learn and iterate faster.

Instead, they discovered that building faster didn’t necessarily mean learning faster.

Some students used AI-generated research in place of customer conversations. Others showed their products before properly understanding the underlying problem. Interviews became reactions to a proposed solution rather than investigations of customer needs.

The article uses a useful term for what emerged: evidence theatre.

Evidence theatre occurs when the visible signs of entrepreneurial progress are convincing while the evidence supporting the business remains thin. The product works, the website looks professional, the presentation is polished and the market research has been summarised.

All are useful. None demonstrates that customers have an important problem, that this product solves it, or that enough customers will pay enough to support a business.

AI has made that distinction more important because many of the things we once associated with startup progress are becoming inexpensive to produce.

From MVP to IUP

The Minimum Viable Product has long been central to startup thinking. Properly used, an MVP isn’t simply an early version of a product; it is an experiment designed to test an assumption.

There was also a practical discipline built into the process. Software took time and money to develop, so there was good reason to think carefully before building it.

AI changes that calculation. If a prototype can be created over a weekend, why not simply build it?

Often, that’s exactly the right thing to do. Cheap prototypes allow entrepreneurs to test ideas that previously would have been too expensive to explore.

The problem arises when the prototype itself is mistaken for evidence.

The Lean LaunchPad instructors began calling some of these creations Initial Untested Products, or IUPs.

It’s a useful distinction. An MVP is intended to test something the entrepreneur wants to learn. An IUP may demonstrate little more than the fact that the product can be built.

Its value comes from what happens next.

Customer discovery or customer confirmation?

There was another unexpected effect. You might assume that cheaper products would be easier to abandon. Instead, students sometimes became attached to what they had built.

A sketch still feels like an idea. A polished application feels like a product.

Once that happens, customer discovery can quietly become customer confirmation. Instead of asking customers to explain their problems, behaviours and existing workarounds, entrepreneurs demonstrate their solution and ask what people think.

The information produced is quite different.

“That’s interesting” isn’t the same as “I need this”, and “I’d probably use it” isn’t the same as paying for it.

Good customer discovery therefore depends on searching for evidence that might contradict the idea, not merely collecting reactions that support it. AI hasn’t changed that principle. It may simply have made it easier to overlook.

The bottleneck is moving

For many technology startups, building the product was once a major constraint. Software development was expensive, engineering talent scarce and a meaningful prototype could take months.

AI is reducing that constraint. As building becomes easier, identifying worthwhile problems, understanding customers, finding distribution and deciding what deserves to be built become relatively more important.

In other words, judgment becomes more valuable as production becomes cheaper.

Competition changes as well. The same tools available to a startup are available to competitors and increasingly to customers themselves. A feature that once required months of engineering may be relatively easy to reproduce.

Customer discovery therefore needs to address another question: if customers value this product, what makes the resulting business difficult to replace?

The answer might be proprietary data, distribution, integration into customer workflows, specialist expertise, relationships, network effects or brand. Whatever it is, defensibility is another assumption requiring evidence rather than a claim to be added to a pitch deck.

A better measure of progress

AI may also require us to reconsider what entrepreneurial speed means.

The ability to produce prototypes quickly is useful, but the number of things built isn’t necessarily a measure of progress. A better measure may be how quickly uncertainty is reduced.

Have we established that the problem matters? Do we know who experiences it most strongly and how they solve it today? Has somebody paid? Have we discovered why others won’t? What assumptions have turned out to be wrong, and what did we change as a result?

These outcomes are less visible than a polished prototype, but they represent genuine learning.

Lessons learned

The experience described in The Year AI Came For Us suggests several practical lessons.

Building and learning should be treated as separate activities. AI can accelerate the first without necessarily accelerating the second.

A prototype should be regarded as an experiment, not evidence in itself. Its purpose is to answer a question about customers, behaviour, pricing, distribution or another important assumption.

Customer discovery becomes more important as development becomes easier. Entrepreneurs should actively look for disconfirming evidence rather than accumulating favourable reactions.

Defensibility also needs to become part of discovery earlier. When products and features are easier to reproduce, founders need evidence for why the business itself will be difficult to replace.

Most importantly, we may need a more modest definition of entrepreneurial progress.

A working product once told us quite a lot. Increasingly, it tells us something narrower: that the product can be built.

The harder questions haven’t changed.

Is there a customer who needs it, will they pay for it, and is there a business worth building around it?