In early 2020, Francis Pedraza was staring squarely at failure. For more than four years, the Cornell grad had been trying to combine AI and platoons of remote workers to help businesses scale fiddly projects like screening résumés, manning chatbots or rewriting product descriptions—repetitive tasks that remained just slightly too complicated to fully automate. But uptake had been slow, and venture capitalists had been reluctant to invest. Services businesses like his Invisible Technologies were horrible investments, per Silicon Valley lore. Hard to scale, hard to run, hard to defend. Hard pass. Invisible had lost four of its cofounders and had to go back to its few loyal angel investors for more money. Then, in March 2020, DoorDash called.

The food delivery company told Pedraza it needed help, fast. The pandemic's global lockdown measures would more than double demand for takeout orders to $51 billion that year. DoorDash was in a race with Uber Eats and Grubhub to find and sign new restaurants. Specifically, it needed assistance with the messy business of importing menus and pricing. The outsourcing shops that had typically done this work were now shuttered.

It was the deal Pedraza had been looking for. "I hate operations because it's friction all the way down, but that's why people buy it," he says.

Two years later, he got another call from a business struggling with an even bigger data problem. OpenAI wanted Invisible's help hammering the hallucinations out of what would become the model underlying ChatGPT. Contracts with Amazon, Microsoft and AI unicorn Cohere followed, helping skyrocket Invisible's revenue from $3 million in 2020 to $134 million last year, on which it made a profit of $15 million (Ebitda).