How to build an AI startup: Go big, be weird, and embrace potential doom

It is said that the earth Home to over 10,000 AI startups. They are more abundant than cheetahs. They outnumber the red trees of dawn. This number is just a guess, of course, as startups come, startups go. But last year, more than 2,000 of them received the first round of funding. As investors pour their billions into artificial intelligence, it’s worth asking: What are all the creatures that have emerged from this boom? a job?

I decided to reach out to as many new AI founders as possible. The goal was not to try to pick winners, but to learn what building AI products looks like in reality, and how AI tools have evolved It changed the nature of their work; How terrifying it is to compete in a crowded field. It was like trying to tap dance on the turbulent surface of the sun. OpenAI He releases an update, and a series of posts on X predict the slaughter of a hundred startups. harsh!

Is this a revolution that ends at the feet of many engineers? Of course, they can’t all survive. A startup is an experiment, and most experiments fail. But if you run thousands of them across the economic landscape, you might learn what the near future holds.

Navvi Anand is Co-founder of a company called Bendwell. When we had a video call, he spoke with a half-smile and a mysterious elegance as he told me how he develops pesticides using custom AI models. The Bindwell website once described these models as “insanely fast,” and claimed that they could predict, in “a few seconds,” the results of experiments that would otherwise take days. Hearing Anand explain how he applies AI drug discovery principles to crops, it was easy to forget that he is 19 years old.

Anand grew up in India reading Hacker News with his father and was building large language models of his own by the middle of high school. Before he graduated, he, his co-founder (now 18), and two other friends from summer camp posted an article paper In bioRxiv, we describe the LLM that they built to predict one aspect of protein behavior. Scientists were abuzz about X. The paper was quote In a respectable magazine. They decided to try a startup, brainstormed, and settled on protein-based insecticides. Then, the fairy tale continues, as some wooden being (sorry, venture capitalist) reaches out to LinkedIn and offers him $750,000 to drop out of high school and college and work at the company full-time. They accepted and started. The teens knew almost nothing about agribusiness. That was last December.

Five months later, Anand and his co-founder opened their first biological testing lab in the San Francisco Bay Area, then moved to another lab, where they personally squeezed droplets of promising molecules into small vials. (The protein-based compound could target locusts or aphids more precisely, the theory goes, and not as well kill humans, earthworms and bees.) I asked him how he acquired the skills needed to work in a wet lab. “I hired a friend,” he said cheerfully. The friend coached him over the summer before returning to college in the fall. “Now I can do some biochemical tests,” says Anand. “Not like a full suite of tests, but validation of our basic models in the wet lab.”

Huh, I thought. That a few teenagers in the space of a few months have built their own MBA, learned the biochemistry of pest control, used their models to identify potential molecules, and are now hauling them away in their own lab doesn’t seem like such a bad thing. In fact, once I counted everything they did, it seemed downright ridiculous to me. I expected to hear that AI tools were accelerating parts of company building, but I only had a vague sense of the scale of their impact. So, my next interview is with the founders of a 14-month-old startup Vertigo techniquesyou get right to it: break down what has changed and to what extent.