How to Create an Easily Murdered "AI" Company
If you're simply creating wrappers on OpenAI or other LLMs, you're not going to make it.
Before we start…
Today’s post is a paid one. If you’re a free subscriber, you can read a preview. Paid subscribers get access to one additional long-form post a week, “One” Thing Wednesday weekly posts, the full archives of past posts, and you get to request future topics for upcoming posts.
The paid version is just a few dollars a month and I’m confident that even if you learn one thing here, it will bring a big ROI to your life and business. Given that, you can just go ahead and expense it. Here’s a nice discount to get you started:
And if you’re just finding this blog and haven’t subscribed (free or paid), do that now!
One additional limited time offer…if you’re a current subscriber and reading this in your email, click reply and tell me you’re real living person and I’ll send you a code to get the paid version for 75% off. That means it’ll cost you less than 2 measly dollars a month for the paid version. Seems fair.
How to Create an Easily Murdered "AI" Company
As I navigate my way around the technology world, I frequently see and get pitched the next big thing. As you can imagine, today most of these deal with AI at some level. However, I can’t help but see a familiar pattern that’s been repeated many times in the past. One company I worked for was even a victim.
It’s January 2012 and I just joined my first startup as employee 13 (Lucky 13?). The company was Zipscene and its largest source of revenue was building Facebook pages and tabs for restaurants. If you can recall ancient Facebook in 2012, as a brand you could choose what tab people saw when landing on your page. You could even “likegate” it and hide all the content until the person liked you page. This worked great for promotions where you could get a Like in exchange for a discount code.
And then March 2012 came around and Facebook killed the ability to set a default tab for your page. It sounds like a minor thing, but it completely killed the mechanism of our company’s main product and how we made most of our money. As the VP of product marketing, everyone looked at me for what we were going to sell next. Fortunately, we we were able to come up with some much higher order product offerings and eventually the company was acquired by NCR.
The lesson here is to not completely base your product on a platform that doesn’t care about your business and that one tweak to said platform can kill you. There is a massive graveyard of companies that relied on platform features that eventually went away. Changes to features and the APIs for Facebook, Reddit, Instagram, and Twitter/X have murdered a tons of startups over the years including every one that offered a wrapper for viewing content from those platforms. And of course, Google has killed a ton of categories of products over the years.
And now the hottest platform on which to build your new company is OpenAI and ChatGPT. OpenAI has done a great job with its APIs to allow anyone to build on top of the LLM it created at tremendous expense. And lots of companies popped up using these APIs, but many of them were recently killed by OpenAI.
Take ChatPDF. It’s a great idea…you upload a PDF and then you can ask questions in natural language about the PDF. And it was doing well. But then OpenAI added a feature to upload any PDF and do exactly the same thing. That’s the end of the road for ChatPDF and tons of other “wrappers” that did this same sort of thing.
With so much hype around AI, there’s a massive temptation to just build something that leverages the existing LLMs out there, but you need to understand the risk that comes with that and not enough companies are paying attention to what history has taught us about this sort of thing.
There are actually only a few defensible positions around AI and make you “unkillable” when the next feature or new restriction is introduced. What’s interesting is that these defensible positions have been the same forever in tech regardless of what innovation (e.g., AI) they are adjacent to.
I found a great article from Darmesh Shah (co-founder of Hubspot and others) that talked about this specific issue that I’m going to leverage heavily here.
There are really two things that matter in terms of your survivability:
Does it create enough customer value?
Are you doing something that is particularly hard?
If the answer to either of these is no, then you probably don’t have much in terms of defensibility and are probably risking an instant death. The customer value piece is not directly related to what I’ve been talking about thus far where you’re killed by a platform change. You should look at customer value on its own. If there is enough value created, then you can probably survive anything. But that’s not what happens in most new companies. There just isn’t enough value created that it’s very easy to move to something else when it comes along.
The “particularly hard” part is where I’ll focus. The definition of hard that I’m using here is when something is very difficult for others to do in a reasonable amount of time. A wrapper on top of ChatGPT does not fit this criteria in the least. The world is full of products that a focused team could roughly recreate in a matter of day (especially if this team happens to be part of a company like OpenAI that attracts the best of the best).
Darmesh has a nice list of three types things are hard and while these are presented in the context of AI, this list is always going to be applicable in tech. That is, doing these types of things or having these things will always provide some measure of defensibility regardless of the market or what technology the company leverages.
Proprietary data or assets. This will always be a moat that is hard for competition to overcome. If you have the best data set of, say, items ordered in every restaurant in the country (what Zipscene started working on after the Facebook apocalypse), then you have very unique set of data on which to build a company. You have access to insights and information that no one else has and that is hard to build because, in this example, you need to get the data from each restaurant one by one. That’s very hard to do.
Network effects. One of the most vastly underrated challenges in tech applies here as well. It’s why it’s not hard at all to technically build a new social media platform, but it’s nearly impossible to build another Facebook. You need to create the network effects where there is enough “supply” and “demand” to make it useful for many people.
Outstanding customer acquisition. This is the hugely underrated one on the list. Even if you have a slightly sub-par product compared to other competitors, if you’ve figured out a way to access and acquire customers that’s way more efficient (faster and cheaper) then you will win.
As you can see from this list, it doesn’t matter if you’re building an AI company or any other tech company. These things are always going to determine who wins and who loses in tech.
Before you go…