None of these concerns seem very specific to AI startups (*).
Like the defensibility and competitive advantage: These concerns exist with most (all?) non-AI startups as well, and they are equally valid. See countless examples of incumbents cloning startups, and using distribution advantage and what's effectively dumping to crush the competition.
(*) Interestingly, prompting ChatGPT with the article title & sub-titles produces almost the same kind of content.
You're right - the same advice applies to non-AI startups too. I wrote the piece to help the founders who come to me for advice, and they are AI founders in the vast majority of cases.
Here is the problem. Every investor is going to ask what your moat is. What differentiates your Whisper -> Llama -> Midjourney Pipeline.ai from the next one? And the answer is, if you’re just making API calls, nothing. Sorry. There’s nothing stopping Jian Yang from creating newpipeline.ai in a weekend.
Here’s a couple of things which could set you apart off the top of my head.
Customers. Having customers is an advantage over the next guy who doesn’t, because now you can start customizing your product for unique needs rather than having a generic crud app.
Custom models. A custom model means some kid can’t just replicate your app easily.
Unique data. Data which is infeasible for another company to acquire or replicate.
Special people. People who will give your startup an edge in creating all of the above.
Or you know, just build a valuable company and VCs will line up to your door and you can tell them "I won't speak your language". This is how much I need for this much as a normal investor would.
Midjourney is a great example. An actual valuable product which people pay to use without VC funding.
And companies like context.ai and the thousands of others which are exactly the same have to do this whole monkey dance for VCs.
YC was unique because it was helping nerds who would never start a company on their own, start one while protecting them from the VC leeches. YC is maybe the one unique case although these days I'm not sure of it's value as much.
> Some tough questions to nail answers to are: What is your defensibility and competitive advantage? ...
Curious to hear some of the answers you used here. User analytics platforms are going to be huge as LLM backed tools start entering production. No doubt that the concept is valuable. But what is to stop DataDog or some other analytics platform from adding LLM analytics?
Perhaps the market will just be large enough that it doesn't matter. And people typically don't switch analytics providers once they're locked in.
The only classes of investors in these new LLM-specific analytics companies are the ones that don't know better and the ones that expect a quick exit or some pivot the rest of us aren't privy to.
The only thing LLM specific in analytics right now is tracking the right REST calls. Most ultra-LLM specific analytics are an anti-pattern if anything: if your agent setup is so convoluted you can't hook up a normal tracing library to it, it's a mess.
That's why 99% of them target Langchain specifically, it's so horribly written that adding basic tracing really is a challenge worth solving. But how long can you bank on the tool of the day just being that terrible?
I've been getting daily "Join us in disrupting the estimated future $2.8T AI market." fundraising emails to my WORK address from something called GenesisAI. The prospectus shows some young-looking cryptobro (or is it ai-bro?) with "We would like to lay a foundation for the creation of Artificial General Intelligence and smash the current oligopolistic system of a few large companies basically owning all the AI." blahfest. Yeah, right.
A note to AI companies: Do not use AI generated images in your posts if they can't generate hands. Your whole thing is that AI is "here", but images like this undercut that message.
Like the defensibility and competitive advantage: These concerns exist with most (all?) non-AI startups as well, and they are equally valid. See countless examples of incumbents cloning startups, and using distribution advantage and what's effectively dumping to crush the competition.
(*) Interestingly, prompting ChatGPT with the article title & sub-titles produces almost the same kind of content.