from Hacker News

There's a lot of what one might refer to as "vertical AI" products being built by startups. Use an LLM in a specialized flow to accomplish some unique task that a general model itself can't just do yet on its own. Usually these involve some combination of specific prompt engineering and chaining together into a flow, external software tools like a vector database, etc.

One could call them "GPT wrapper startups" or whatever I suppose. YC seems very bullish on them at the moment, judging by their recent videos and compositions of accepted companies. But I remember a time when YC was very bullish on cryptocurrency/blockchain projects as well (and notice how they're not even mentioned in the latest Request for Startups?). I know it's early but I don't think we've seen any big hits from wrapper startups - though please feel free to share good counterexamples.

So aside from speed of execution, and perhaps putting the effort into personal relationships and sales to get the ball rolling, what do these startups really have to offer that's defensible?

I've seen first-hand a big company I consult for reject several of these 'wrapper startup' sales proposals because myself and others (correctly so) advocated that a few of the internal engineers could build a good enough product clone themselves with GPT. I.e. a bit of simple "intrapreneurship", which is easier now than ever.

Not to mention, whatever "agent" a startup is working on now, almost definitely 1 year from now a foundation model company (OAI, Anthropic, etc) will just release their brand-new Operator/Computer Use that will blow it out of the water. Unless maybe the startups glue and hack together enough integrations that it's a pain in the ass for the new agent to compete in all uses? But then if they can use other software... Well there's something different about agents I think.

But what do you all think? Will any of these wrapper startups be around in 5-10 years? What are some characteristics of the ones you think will last and grow?