Video: the 'lie' of AI startups

A lot of artificial intelligence startups can look huge in a 90-second screen recording and still fall apart in a 9-to-5 workday, and that gap is the center of a recent Spanish-language YouTube critique making the rounds. The video argues that founders are selling “demo value” when buyers eventually need “daily value,” which is a different test entirely. (youtube.com) That distinction is simple: a demo is a magic trick you control, while a daily product is a tool that survives bad inputs, messy data, security reviews, billing fights, and a customer using it 200 times a week instead of twice on stage. Sequoia Capital framed the same pressure in June 2024 when it asked where the revenue was to justify the scale of artificial intelligence spending. (youtube.com) (sequoiacap.com) The money around these companies is real. CB Insights said on July 3, 2025 that artificial intelligence startups made up more than half of all new billion-dollar startups that year, and 1 in 5 new unicorns were building artificial intelligence agents. (cbinsights.com) The adoption is real too, which is why the video does not say the technology is fake. Menlo Ventures reported on November 20, 2024 that enterprise spending on generative artificial intelligence rose to $13.8 billion in 2024 from $2.3 billion in 2023, a jump of more than six times. (menlovc.com) But spending growth does not automatically mean every startup has a moat. Ramp says its Artificial Intelligence Index tracks paid adoption using transaction data from more than 50,000 American businesses, which is useful because it measures who is actually paying instead of who clicked “try demo.” (ramp.com) That is where many startup stories get thin. If the product depends on the same foundation model as 20 rivals, and the customer can switch after a one-month contract, then the shiny interface may be packaging, not protection. (youtube.com) (menlovc.com) The Spanish-language critique goes after founders who talk like they own the intelligence when they may only be renting it by the token. Sequoia made a similar point from the investor side when it wrote that the gap between infrastructure spending and end-user revenue had grown from its earlier “$200 billion question” into a “$600 billion question.” (youtube.com) (sequoiacap.com) A durable software business usually has one of a few hard things: proprietary data, painful workflow integration, regulatory know-how, distribution, or switching costs that make leaving expensive. Menlo Ventures said in its 2024 enterprise report that the application layer was growing faster as companies embedded tools into real workflows, which is where those hard things start to matter. (menlovc.com) That is why “daily value” is a better filter than “viral demo.” A legal assistant that saves a team two billable hours every day, or a customer support tool that resolves thousands of tickets inside an existing system, has a stronger case than a chatbot that gets applause in a sales call and then misses half the edge cases in production. (youtube.com) (menlovc.com) The thread running through the video and the market data is not that artificial intelligence is a fraud. It is that a funding boom can hide the difference between a feature people try once, a demo investors remember, and a product a company budgets for every month. (youtube.com) (ramp.com)