AI is the lever, not the product
This time is not different
I’ve seen hundreds of AI companies and have also helped a handful of companies build AI products.
I have found it helpful to separate these companies into two groups: 1) AI-native companies and 2) AI-augmented companies.
AI-native companies are those that sell and provide AI-specific services as their core business: foundation model providers (e.g., OpenAI, Anthropic, Cohere, etc.), AI infrastructure companies, AI developer tools, etc..
AI-augmented companies are companies that, while their core business might not rely on directly selling AI products or services, are using AI to make their products more effective and efficient.
It’s important to understand what type of AI company you are, because it changes the way you should run your business, product strategy, and technical development.
The issue is that too many AI-augmented companies are pretending to be AI-native companies — and it will end poorly for them.
AI is the lever, not the product.
Substance, not just style
Tell me if you’ve heard this story before.
Alfred is a CEO of an e-commerce startup in Silicon Valley. One day while perusing Twitter, he sees a demo of a cool generative AI tool. Everyone in his circles is convinced that this is going to transform the industry and disrupt everything. He’s inspired by this and decides that his company needs to reinvent itself by fully leaning into this revolutionary new technology. After all, his board has been bothering him about his generative AI strategy since the release of ChatGPT. So, Alfred tells his CTO that they need to add more AI into their products — go!
From the CTO’s perspective, it’s a very nebulous request, but Alfred sure seems adamant about this particular mandate. So, the CTO proceeds to hire a few top-tier research scientists from the big tech companies and reallocates a handful of software engineers and product managers to help. Unfortunately, the research scientists are used to an incredibly robust technical stack at their old companies, but the startup just doesn’t have the resources to do this. The product managers frantically try to learn more about the differences between all the different model architectures and technical tradeoffs as they spec out the product roadmap. And of course, the software engineers are overworked, because there’s always infinite work to be done here.
While there might be some features shipped and a few working groups formed, the ambiguity of the mandate “add more AI to the products!” is overwhelming. A year and a half later — and three million dollars poorer — the company’s product has a few new AI features, but nothing much more.
When asked about his company’s AI strategy at the next board meeting, Alfred replies: “AI just doesn’t work!”
It goes without saying, but don’t be Alfred. Throwing AI at your product is unlikely to work, and it takes a much more deliberate approach to be successful. A lack of results here points toward a flawed process, not a flawed technology.
For an AI-augmented company, the best way to think about AI is as a lever.
A lever can be immensely powerful, but only if it is rooted atop a fulcrum — as Archimedes said, “Give me a place to stand, and I shall move the world.” But without the fulcrum, the lever is simply a giant stick.
Your existing product is the fulcrum that the AI lever sits on. AI can augment valuable product features or dramatically improve the product's cost structure. But without the product, AI is unmoored and can’t do anything.
Indeed, the most successful AI use-cases follow this pattern. Klarna blew up AI Twitter when it announced its implementation of an AI customer service chatbot which does the work of hundreds of equivalent people. I’ve stumbled across some clever use of LLMs to summarize and personalize enormous amounts of text seamlessly within the existing user experience. Human-in-the-loop AI systems are also really interesting, because they are a pragmatic approach to augmenting companies’ existing operations with AI.
In each of these AI-augmented examples, the product is still the same but the AI goes towards making the customer’s experience with the product better. The teams have taken opinionated design choices, driven from a crucial understanding of the customer’s experience, figured out where AI can be most effective, and implemented the features.
While these particular examples aren’t exceptionally flashy uses of AI, I’d rather be effective over flashy. And plus, as Klarna shows with its AI chatbot implementation, it’s hard to argue with $40M in additional profit.
Focus on your product, figure out where AI can be effectively deployed to make your product better, and then do that.
AI is the lever, not the product.
Product fundamentals don’t change
Jeff Bezos is constantly asked what he thinks will change in the next ten years. But he believes the more interesting question is:
“What is not going to change in the next ten years?”
In an age of accelerating technological advancement, it’s easy to get caught up with everything that is changing. Like Bezos, why don’t we focus more on what doesn’t change?
What won’t change: people have a desire, and they buy your product to solve that.
By itself, AI won’t solve that. It still takes a thoughtful approach to build great AI products.
AI is the lever, not the product.