How will the way people build AI products change in the future?

What the market sees is that these companies reach incredible revenue milestones with far fewer people, faster than we've ever seen before. Josh's hypothesis is that the difference runs deeper than the product — AI-native companies going from one person to 50 never adopted the same legacy systems and processes everyone else runs on, which is how they attract incredible talent quickly while staying lean and fast.

His own target for the refounded InDebted is the factory metaphor: "all companies are a factory and our job is to produce a widget — and we should have a brand new factory and produce widgets at speeds that seem completely impossible to the business that we are today." Whether the widget is a line of code, a hire, or a performance review, mastering the automation means the whole company's brainpower can concentrate on one problem at a time — "and that's the next unlock for the company."

As a CTO, Satya sees the world going toward the latter of two trends — specific models over one general-purpose model — because only a few big players have the dollars to build general-purpose models. Startups and teams ambitious enough to solve for a particular vertical will fork, fine-tune or build narrower versions that work well in a specific use case. He cites one example: a VC batchmate building a text-to-speech model whose selling point is being more emotive, trained on podcast data and real conversations so the speech has emotions. He's seeing a lot more startups in this space building for narrower use cases, and notes that with AI "the time to product and time to market has dropped by an insane amount," so you can test ideas quickly and get feedback from customers.