I thought this was a pretty good analysis. And for comments along the lines of "Most start-ups are doomed", the article acknowledges this in the first sentence

I thought this was a pretty good analysis. And for comments along the lines of "Most start-ups are doomed", the article acknowledges this in the first sentence ("The statement that most AI startups are doomed can be fairly mundane. After all, most startups are doomed, just by the numbers.") but then goes on to make an argument specific to these AI startups. I do, though, believe the author missed a large class of AI startups that I think will likely succeed in the "Wait, so what IS defensible?" section: startups that focus like a laser on very specific, semi-niche workflows where things like UX and compliance are critical. My best example of this so far is Harvey.ai, whose tagline is "Generative AI for Elite Law Firms": 1. First, elite law firms have lots of money to spend, and they'll spend it if they see an ROI. 2. There are plenty of Web 2.0 startups who won primarily because of first-mover advantage. I mean, Docusign wasn't exactly amazing, world-changing technology, but they became synonymous with "legal signatures over the internet" such that they became the default for this use case. 3. Obviously something like "generative AI for elite law firms" has tons of compliance concerns around it. If Harvey.AI can address that, it's a huge wn. As another example, I know of some big financial firms/banks that have giant committees around anything remotely label-able as "AI" because there are so many compliance concerns around AI (a system that gives you an answer with no visibility into how that answer was generated is anathema to the "everything must be auditable" mindset of the financial world). Again, Docusign is a good analogy here, because so much of their initial work was not in tech but ensuring that there was a legal framework (in many jurisdictions) that would deem internet signatures valid. My overall point is that a UX that is highly tailored to specific, profitable use cases can still win out.