**Since 2006, I’ve tested 13 business ideas. ** That works out to roughly one new business every 18 months. Only one became a meaningful success.
The economics of experimentation have since changed.
Over the past eight months, I’ve built more than 10 applications, roughly one every 24 days. The increase is not the result of suddenly having more ideas. AI has compressed the time and cost required to turn an idea into something people can actually use.
That matters because building products is has become so easy, that, arguably, the hardest problems in business in 2026 is determining whether anyone values it.
Markets are difficult to predict from first principles. Founders can conduct interviews, analyze competitors, estimate market size, study search behavior, and construct detailed financial models. These methods improve decision-making, but none substitutes for observing what people actually do when presented with a product.
The creator economy illustrates the problem.
Adobe estimated that there were 303 million creators globally in 2022. Goldman Sachs, using a narrower definition, estimated a global population of roughly 50 million creators and found that only about 4% qualified as professionals earning more than $100,000 per year (Adobe, 2022; Goldman Sachs, 2023).
The estimates differ because the definitions differ. The economic pattern does not.
Participation is widespread. Significant financial success is concentrated.
Attention is scarce and the distribution is difficult. And above all, creating something does not establish demand for it.
Startups face the same constraint.
CB Insights analyzed more than 100 startup post-mortems and identified lack of market need among the most frequently cited reasons for failure. Its widely circulated analysis found “no market need” in 42% of the cases examined (CB Insights, 2018).
That statistic requires some precision. It does not mean that 42% of all startups fail because there is no market need. The research examined a particular sample of failed companies, and companies could report more than one reason for failure.
The useful point is narrower: founders repeatedly invested substantial resources into products for which sufficient demand never materialized.
This is the economic purpose of the Minimum Viable Product.
Eric Ries formalized the concept in The Lean Startup. An MVP is not simply a crude or incomplete version of a product. Its purpose is to generate validated learning about customers with the least necessary investment of effort (Ries, 2011).
That changes the objective of early product development.
Before scale, the first objective is reducing uncertainty.
Asking one's self:
- Does the problem exist?
- Do people care enough to try the solution?
- Do they return?
- Will they pay?
- Can they be reached at an economically viable cost?
These questions are better answered through behavior than speculation.
This is where AI changes the operating model.
Software development historically imposed a substantial cost on experimentation. Even relatively simple products could require weeks or months of engineering, design, debugging, deployment, and iteration. That cost naturally encouraged founders to spend considerable time deciding which ideas deserved to be built.
The problem was still there, but it did require greater diligence.
AI coding tools reduce that constraint.
Today, what is needed is a higher rate of experimentation.
If an entrepreneur can test one serious product hypothesis every 18 months, a failed hypothesis is expensive. If the same entrepreneur can test one every few weeks, failure becomes information acquired at a much lower cost.
The relevant metric therefore shifts from How much can I build? to How quickly can I obtain credible evidence about demand?
This is why I now build so frequently.
I do not need to know in advance which idea will work. I need a process that exposes ideas to the market quickly enough that weak assumptions are discarded before they consume significant time and capital.
Build the smallest credible version.
Ship it.
Measure what people actually do.
Keep what survives contact with the market. Change or discard what does not.
Then run the process again.
AI has made building cheaper, but demand remains unpredictable. So build it fast and let the market decide.
For the curious
Adobe. (2022, August 25). Adobe “Future of Creativity” study: 165M+ creators joined creator economy since 2020. Adobe. https://news.adobe.com/news/news-details/2022/adobe-future-of-creativity-study-165m-creators-joined-creator-economy-since-2020
CB Insights. (2018). Why startups die. CB Insights Research. https://www.cbinsights.com/research/why-startups-die/
Goldman Sachs. (2023). The creator economy could approach half-a-trillion dollars by 2027. Goldman Sachs. https://www.goldmansachs.com/insights/articles/the-creator-economy-could-approach-half-a-trillion-dollars-by-2027
Ries, E. (2011). The lean startup: How today’s entrepreneurs use continuous innovation to create radically successful businesses. Crown Business. https://www.penguinrandomhouse.com/books/210088/the-lean-startup-by-eric-ries/