Four ways to make money with artificial intelligence.
Way number one: the con-artist method
The first one we can call the classic con-artist method. You get your money from suckers. You go around conferences talking rubbish, you play the shiny decoy, and you round up a whole crowd of gullible people who want easy solutions to hard problems. You sell them snake oil and then you run off to South America on the next plane.
Don’t underestimate this one, because there are plenty of people who can’t be bothered to use their brains, so they fall for it. Just look at all the traffic going to a certain class of nonsense-peddling influencers right now.
Way number two: AI as a feature
Method number two we can call AI as a feature. Inside services that already exist, I find a way to also integrate some artificial intelligence functionality. For example, I’ve got an e-commerce shop for shoes. I add a little chat for customers, to search, to compare shoes, to kick off orders, to check the status of a shipment, to figure out which stores exist and where they are. So a bit of memory, a knowledge base, and a bit of operational capability.
These bots don’t just know things, they also do things, and that’s what people mean by agentic AI. Which is itself a bit of a buzzword for using language models like ChatGPT or Claude, with personalized memory and a set of operations they can carry out, and you can happily bolt it onto the e-commerce shop.
Way number three: AI first
Strategy three is the AI first one, as Google puts it: “we’re an AI first company.” It means the service is designed to be AI as the entry point, not as an add-on to a pre-existing service. The experience itself is largely centered on the artificial intelligence and automation aspect.
In this case I’m building an agent that slots into various places and, whether or not there’s an e-commerce shop behind it, that agent is able to expose the catalog, take orders, and give information. But the agent is the point of entry. You still see few of these services around, but they’re fascinating.
Notice that for both AI as a feature and AI first, either way there’s a database with the products, there are rules for engaging customers, there’s the cart, and there’s the information. There can be a back office for whoever runs the shop, with a set of extra private features. The data and the processes are more or less the same.
Integration is the real market
A lot of people see artificial intelligence as something that replaces everything. In both these cases we’re talking about integration. That’s why I keep repeating it: the bulk of the concrete AI market will be built by system integrators, not by scientists and astrophysicists. This is a matter of integrating language technologies, language models, with traditional systems. It’s through that marriage that you get chatbots doing things that are actually useful to people, things people can understand and act on.
Way number four: join the platforms
The fourth strategy for making money is to participate in the platforms. OpenAI in particular recently proposed a platform for apps. They basically want to do what Android or Apple did with the marketplaces for mobile apps. The experience lives inside ChatGPT: you type “Canva, do this,” “GitHub, do that,” “shoe shop, do this other thing.” Your service is encapsulated inside the ChatGPT experience.
They love this because it turns them into a platform, which is the wet dream of every turbo-capitalist in Silicon Valley. And every transaction we manage to land through these apps, we have to run past them first and get fleeced, obviously.
So here I’ll leave you with a question, especially for those of you who follow this stuff a bit more closely. In this race for agentic AI, will the agents behave like wild web services, the way we already do with websites? I put my site under an address, then I use WordPress, Laravel, whatever, I build the shoe e-commerce, I build the showcase site for the archaeological park. You can do the same thing with AI agents: long tail, maximum freedom, open protocols, everyone goes their own way and then they network with each other.
The other vision is the closed marketplace, which with hardware is clearly much easier to pull off. This closed approach only works up to a point on the web, but the big tech companies really are trying to build these marketplaces.
So I ask you: in your opinion, does the free long tail of AI win, agents on the web doing the most disparate things and eventually talking to each other? Or will the Americans manage to close off these circuits and turn them into marketplaces, the way they did for mobile apps? I leave you with that question. Let me know what you think.
