Why AI is not a Moat

Stirring up some controversy on a Tuesday

For the loyal readers out there, sometimes you know I like to go on a good rant. Well strap on your seat belt as here’s one fresh in my mind for all of you.

Shomik Ghosh 🇺🇦 @shomikghosh21

I may tape a sign to my forehead reading "AI is not a moat, utilizing AI is not product differentiation" 😡

Lately there has been a lot of hype about AI and rightfully so. If you’ve played around with Github CoPilot, DALL-E or heck even the Honda assisted cruise control, it’s hard not to be blown away by how incredible algorithms can be in guessing what humans are looking for.

The magic of these products have captured founders’ minds all over leading to many different companies being built with AI as the central foundation of the core product.

For me personally, that is always the part where my heart sinks when I meet a great founder and that’s the essence of the pitch. I’ll explain why below.

When is/not AI a Moat?

We could have a whole separate discussion around what AI is really, etc. But now is not the time for that.

Here I want to address why AI is not a moat for startups. But also, when it can be.

At its core to do AI/ML/any predictive modeling, you first need data. And this is where the plot thickens. For any startup, you are starting with 0 users, 0 data, 0 revenue. Your goal is to get all of those to grow quickly by delivering a product that users want and solve clear pain points. When AI is pitched in your product as the core differentiation, herein lies the problem.

You are pitching the main reason that your product should exist as a byproduct of having lots of data. That data in turn needs to be ingested, cleaned and then put into a model which then needs to be trained, tuned, and deployed. Meanwhile, while building all of that, you also need to get the fundamental input to your product that drives all of this….data.

That’s the dilemma. The core product differentiation is AI which rests on having enough data to deliver outcomes to end users, however there are no end users currently. Even if you train models on publicly available data, that data likely does not mimic the end user pain points which leads to false positives and a poor end user experience.

A startup must first and always focus on what pain point is being solved. So if you are trying to build a better code generation product, that’s great. But it needs to be fairly tightly scoped so that the cost and availability of the data needed to deliver a great experience is as small as possible. This could mean focusing on one language, one area of software engineering, or even one specific type of engineer. But if you are going to pitch AI as a core product in and of itself, focus on scoping it down as narrowly as possible otherwise it’s going to be hard to create a sustainable business.

When can AI be a moat?

The answer is in larger companies! They already have mounds of data flowing in through the system because they solved a clear pain point for the user, have years of data on how the product is being used, and then can train models to better implement the product solutions for end users. Security is a great example of this. Cloudflare can utilize AI/ML for DDOS protection because they see so many attacks across different regions and methods. Crowdstrike and SentinelOne can do the same as they aggregate huge amounts of endpoint data showing different attacks on all sorts of laptops, phones, etc. However, and this is key, Crowdstrike can not offer better DDOS protection than Cloudflare because they don’t see the same volume of data in that area and vice versa for Cloudflare with endpoint protection. So AI can indeed be a moat but is restricted to more “vertical type” product areas.

AI Enabling New Types of Products

The other thing a lot of founders mention is that the wave of AI tooling is enabling new types of products to emerge. For example, with OpenAI or Stable Diffusion’s easily accessible APIs, lots of companies can be built. I’ve seen customer feedback tooling, dev testing products, and video parsing solutions recently all with AI built in from day one.

This is indeed a very exciting thing! As an end user I can not wait to use these products.

But does that make it a good product and company idea from day one?

The answer to that is perhaps but be aware of the challenges. Just like better APIs at Salesforce have led to a massive amount of sales tooling being built, the same thing goes for these AI products. Chances are that as 3 engineers in a room, the AI tooling you’re building is leveraging the same building blocks that 3 other engineers are using next door. Maybe one team can build it better than the next. But how much better? How important is that difference to the user? And also how long before someone else with equal or better ability comes along and builds the same product leveraging that same AI tooling as they see the opportunity as well?

By having AI as the core product differentiation, the company is essentially choosing to play in the rat race. A pure execution game where the company has to outcompete, outlast, and outsmart the competition to get in the hands of users. This is different than solving a pain point of an end user uniquely through earned secrets after deeply understanding the end user’s workflow. In those types of startups, there is a knowledge barrier that is harder to replicate. A technical barrier, which sometimes can be enough, will likely diminish quickly with respect to AI tooling as they become easier and easier for any developer to use (especially through APIs).

Conclusion

Don’t fall for the hype! Build for the end user and solve a pain point that either you have experienced yourself or you have researched deeply enough to be able to put yourself in the users’ shoes. Incorporating AI earlier into the roadmap can certainly work, but first the focus should always be on attracting users, customer feedback, and data to see if the product is solving anything valuable in the first place.

I hope you enjoyed this edition of Shomik’s Rants :)

---

Some things I enjoyed reading/listening to this week:

• Tradeoffs between scaling and staying lean

  Yevgeny Pats @yevgenypats

  small team + constraints = improving product big team + no constraints = deteriorating product

• How to be a continuous learner

  🎙 The Learning Mindset for Investors with Alix Pasquet

• “Why you should pay attention to ebpf”

• Finding reader/content fit and growing an audience by CJ Gustafson, CFO at PartsTech and prev Director of Financial Planning & Analysis at Snyk

  CJ Gustafson @cjgustafson222

  Yesterday my newsletter hit 5,000 subscribers. I went from hopelessly screaming into the abyss to riding rocket ship. This graphic sums up the journey. Some humble tips for those who want to build an audience: