To Build a Successful AI Company Look for Things that Don’t Change

Ever thought of starting an AI company?

Everywhere, LLM API wrappers and GenAI DevTools are popping out of the ground like a weed.

However, nation-states and the biggest companies in history are literally pouring billions into the space.

A large number of today’s GenAI startups are picking up pennies in front of a steam roller. The waves of progress will inevitably crush them. If your goal is to build a company that reliably generates wealth over a long time (>5 years), AI can be the worst field to be in.

However, there is a surprising method to the madness!

In the following, we will see why you should stay away from the technological frontier and innovation if you want to build a sustainable AI startup.

By the end, this will give us a tool to reliably come up with startup ideas that will make a company last for generations.

Innovation and The Frontier Tax

All technology is deflationary.

What does that mean you ask?

Well, it means that technology generally makes things cheaper by allowing us to do more with less. This process - sometimes termed technological progress - is generally a good thing. It makes things cheaper (hence deflation) and societies more prosperous.

However, if you are a startup founder, building in an ultra-fast-paced field, this can be pretty bad. Whatever was cutting-edge last year is an old hat today. And the software you are selling is just a few months away from becoming an open-source library - completely worthless that is.

This is true in many areas and it is especially true in AI today.

This means whatever fat margin, you were able to charge for your product, is getting thinner. Fast.

Soon, technological progress will have asymptotically put you out of business.

Technological Progress Shortens Company Lifespans

The rapid progress is pulling every business model down with it, like a giant whirlpool. Companies face tremendous pressure to innovate themselves.

If they can’t keep the pace, they die like flies.

According to Credit Suisse, the average lifespan of a Fortune 500 company is three times shorter than it was in 1950. Back then the average lifespan was 60 years.

Today it is less than 20.

Aggregating all Fortune 500 companies is obviously not very informative. It is a bit like thinking you had sex when in actuality you are just a voyeur who saw someone’s naked elbow through a keyhole.

You’re not getting the complete picture is what I’m trying to say.

Nobody would say the temperature in Europe is 10°C. That might be the average. But temperature is a field whose values vary substantially from one region to the next.

A country’s rate of technological progress is the same in that regard. It varies heavily across industries.

Software’s rate of progress has completely outstripped almost any other industry. It is not uncommon for software to improve 100x over a couple of months. Think about the improvements from GPT3 to GPT3.5. The latter is faster, cheaper, and better than its predecessor.

This rapid rate of change makes it hard for pure software companies to last even a decade.

Companies that play at the technological frontier need huge budgets for innovation and a healthy risk appetite. You can think of this cost as a sort of frontier tax. It is the price you need to pay to play this type of game.

To minimize uncertainty, stack the odds in our favor, and build a sustainable company, we should go exactly the opposite way!

We should anchor our business to things that are not changing.

In the next part, I will show you how and why that’s the case. Spoiler: The answer is not ‘proprietary data’.

Innovate as Little as Possible

If you are reading this, you probably heard the saying: ‘The future is here but it is not evenly distributed’?

In many ways, people share this planet physically but live in totally different decades or even centuries when it comes to the technology they use. This divide can happen along multiple dimensions, such as:

• Geographically - e. g. Rural India vs. Tokyo or San Francisco

• Across generations - Grandparents can’t use smartphones vs. teenagers on TikTok

• Field of work - Social media marketing (invented 15 years ago) vs. massage therapist (work is almost the same for centuries)

Each of these gradients in technology adoption can be a tremendous opportunity to build companies.

The last dimension is particularly interesting when looking for a niche in which to build!

A neat economic theory from the 1960s will help us to shine some light on why that’s the case.

The theory that I am talking about is the Baumol Effect also known as Baumol’s Cost Disease.

It explains the rise of wages in industries that have seen little to no increases in productivity. It argues that this happens in response to rising wages in other industries that did see rising levels of productivity.

If your head’s spinning, stay with me. It will get clear in a second.

Let’s take massage therapists as an example.

Their work of sending customers to cloud nine by kneading oily backs has looked the same forever. Yet, the wages of massage therapists today are a lot higher than they were one or two centuries ago.

Their wages have risen in tandem with other industries.

The world has a certain demand for massages and since technological progress has made people richer as a whole, they are willing to pay more for such an oily pleasure. If they didn’t pay more no one would do the job. In other words, if one industry gets more productive the competition for labor drives up wages across all industries.

That’s interesting, no?

It means that inherently some industries see rising wages, are subject to very little innovation, and their workforce is not at risk of losing their job to automation.

Some examples of these industries include:

• Industrial equipment/manufacturing - e.g. paper mills

• Packaging

• Real estate management

If we anchor our (AI) technology to one of these businesses, we can build something that lasts for generations.

Paper mills for example use a lot of energy. You try to figure out how high-end manufacturing uses AI to save energy. Then export the solution to paper mills.

In real estate management, a lot of the work is scheduling appointments, answering inquiries, and emailing contractors. You could look into building agent-based workflows to streamline their work.

I am not saying, don’t build a fancy new orchestration framework, such as LangChain, or a cool new fine-tuning tool for LLMs.

I would love to see you succeed with it. Just be aware of the game you are getting into and the speed at which you will need to move.

However, especially for your first shot at a startup, it can be much easier - and frankly more lucrative - to build far away from the frontier.

I sincerely hope you find this useful!

If you have feedback or questions, send me a reply or hit me up on Twitter or LinkedIn.

Lots of love and see you next time!

P.s. I send out an email like this every week. If you are not subscribed yet, go ahead and subscribe here ⭕️