While there were undoubtedly people who came in for the food, worked 3 hours a day, and enjoyed their early retirements, all the people I met were earnest, hard-working, and wanted to do great work. What beat them down were the gauntlet of reviews, the frequent re-orgs, the institutional scar tissue from past failures, and the complexity of doing even simple things on the world stage. Startups can afford to ignore many concerns, Googlers rarely can. What also got in the way were the people themselves - all the smart people who could argue against anything but not for something, all the leaders who lacked the courage to speak the uncomfortable truth, and all the people that were hired without a clear project to work on, but must still be retained through promotion-worthy made-up work.

Another blocker to progress that I saw up close was the imbalance of a top heavy team. A team with multiple successful co-founders and 10-20 year Google veterans might sound like a recipe for great things, but it’s also a recipe for gridlock. This structure might work if there are multiple areas to explore, clear goals, and strong autonomy to pursue those paths.

Good teams regularly pay down debt by cleaning things up on quieter days. Just as real is process debt. A review added because of a launch gone wrong. A new legal check to guard against possible litigation. A section added to a document template. Layers accumulate over the years until you end up unable to release a new feature for months after it's ready because it's stuck between reviews, with an unclear path out.

We still believe that there will be a separation between the “application layer” companies and foundation model providers, with model companies specializing in scale and research and application layer companies specializing in product and UI. In reality, that separation hasn’t cleanly happened yet. In fact, the most successful user-facing applications out of the gate have been vertically integrated.

We predicted that the best generative AI companies could generate a sustainable competitive advantage through a data flywheel: more usage → more data → better model → more usage. While this is still somewhat true, especially in domains with very specialized and hard-to-get data, the “data moats” are on shaky ground: the data that application companies generate does not create an insurmountable moat, and the next generations of foundation models may very well obliterate any data moats that startups generate. Rather, workflows and user networks seem to be creating more durable sources of competitive advantage.

Some of the best consumer companies have 60-65% DAU/MAU; WhatsApp’s is 85%. By contrast, generative AI apps have a median of 14% (with the notable exception of Character and the “AI companionship” category). This means that users are not finding enough value in Generative AI products to use them every day yet.

generative AI’s biggest problem is not finding use cases or demand or distribution, it is proving value. As our colleague David Cahn writes, “the $200B question is: What are you going to use all this infrastructure to do? How is it going to change people’s lives?”

What is the testable hypothesis? What would falsify the hypothesis? How do we know when we are getting into a danger zone? These questions go mainly unanswered apart from “You can’t prove it won’t happen!” In fact, these Baptists’ position is so non-scientific and so extreme – a conspiracy theory about math and code – and is already calling for physical violence, that I will do something I would normally not do and question their motives as well.