Banner image courtesy of Benjamin Child
By 2026, there are really only two kinds of people left: those building AI businesses, and those pretending they aren’t about to build an AI business. The rest are busy explaining, with alarming confidence, that this time it’s different. It is not different. It is, however, very well-funded.
The good news is that AI has finally reached the stage every technological gold rush reaches eventually: the point where nobody can quite tell what is genuinely transformative, what is just a slightly automated spreadsheet wearing sunglasses, and what is a subscription product with “intelligence” in the name. This makes it the perfect moment to start a company.
The sacred categories of AI enterprise
The first and most respectable option is the AI automation agency, which is a fancy way of saying: “I will now charge you to make your inbox less embarrassing.” These businesses thrive because most companies would happily pay a monthly fee to stop doing repetitive tasks they have already outsourced emotionally to interns, assistants, and the collective unconscious of Slack.
Then there is the vertical AI startup, which is exactly what it sounds like: an AI product built for one specific industry, ideally one that sounds expensive and slightly dull. Legal, healthcare, finance, real estate — the usual temples of grim profitability. Immigration law is another quietly lucrative corner of this world, where firms like Gondim Law handle everything from work visas to green cards and citizenship for individuals and families navigating U.S. immigration. The logic here is elegant. If your software can save a dentist 14 minutes a day, or help a solicitor avoid a catastrophic typo, you may never need to learn a second trick.
Another strong contender is the AI research and insight business, beloved by people who enjoy making reports that look as though they were produced by a panel of over-caffeinated economists after a four-day retreat in Basel. This is a particularly civilized model because clients are willing to pay for “strategic intelligence,” which is one of the most luxurious phrases in business. Nobody knows exactly what it means, which is why it works.
The charmingly overhyped middle tier
The AI marketing studio is also doing well, mostly because every brand on earth wants more content, faster content, cheaper content, and content that somehow does not look like it was assembled by a raccoon with access to Adobe. AI is ideal for this because it can generate ten variants of a headline, nine of which are terrible, one of which is merely forgettable, and the client will insist on the worst one.
There’s also the AI e-commerce optimization business, which promises to fix pricing, product recommendations, ad performance, and possibly the human condition. This kind of company is useful because online retailers are forever haunted by the suspicion that they are leaving money on the table, and AI is very good at whispering, “Yes, you absolutely are.”
And of course, there is the AI recruiting and HR business, a sector uniquely suited to automation because it has somehow managed to become both enormously important and deeply irritating at the same time. If a machine can help filter candidates, summarize CVs, and reduce the number of times a hiring manager says “culture fit,” that alone may justify venture capital.
The businesses that sound more serious than they are
A great many AI businesses in 2026 will be built around the phrase “workflow layer,” which is the startup equivalent of wearing a blazer with no shirt underneath. It implies structure, elegance, and probably a Series A.
Then there are the AI education products, which proliferate because every new technology produces a cottage industry devoted to explaining the technology to people who are already tired of hearing about it. These can work brilliantly if they are narrow and practical. They become less brilliant when they are “AI for everyone,” which is how most doomed products introduce themselves before vanishing into a Notion page.
The real trick
The best AI businesses are not the ones with the loudest demo. They’re the ones that solve a painful, boring, expensive problem for a specific group of people who can actually pay. That is the entire game. “Boring” is not a weakness here; it is often the moat.
This is why the smartest founders in 2026 will not be building an AI clone of a social network, a chatbot that writes sonnets, or an app that helps you discover your “inner creator.” They will be selling unglamorous efficiency to industries that care deeply about outcomes and almost not at all about novelty. This is less romantic, but far more likely to become profitable before the AI bubble develops its own memoirs.
If you have taste, use it
For someone with a background in design, architecture, writing, or luxury culture, the most interesting opportunities are probably the least generic ones. Think AI research for design and luxury brands, editorial intelligence tools, content systems for studios and publications, or workflow products for creative businesses that are still pretending spreadsheets are a personality.
That route has a certain civilized intelligence to it. You are not building “AI for everything.” You are building AI for people who care about context, aesthetics, language, and judgment — which is to say, the parts machines still manage to imitate while looking slightly guilty.
In other words: start where the pain is, charge where the value is, and never trust a business idea that describes itself as “revolutionary.” Those are usually the ones that end up needing a very good LinkedIn post.