[bisby]

There is also the caveat that some companies tried things that work now, but didn't work back then. I don't remember the specific company, but something like Uber was tried back then. and it just didn't work. "Internet is available" is not the same as "internet is ubiquitous" and some ideas require the ubiquity for it to work out, even if the execution was otherwise fine. "LLMs are pretty neat" is not the same as "AGI is ubiquitous." So there are some AI products that people will try, that will fail horrendously, and it won't be because it's a bad idea or executed poorly, but it's because the AI behind the idea isn't ready yet.

[danenania]

Yeah that's true, but I would still categorize it as an execution problem. If your product (or your AI feature) doesn't work because the models aren't ready yet, it means you failed at design and implementation imho. It doesn't mean there's no plausible AI integration that would make the product more valuable.

It's like if you choose the wrong database and it makes your product slow. It probably doesn't imply you shouldn't be using any database; you just need a different one.