How Navi and his small team actually build — and his "easier to build, harder to build value" thesis.

Stack and AI adoption

  • Stack: Angular on the front end, Firebase, Google Cloud, TypeScript — self-described "pretty outdated tech" 53:03
  • Over the last ~6 months most code has been written by AI; before that Navi was "a really big AI hater" — his shift is pragmatism: the tools finally got good enough for his mostly-CRUD workload (the code execution service being the main exception) 52:16, 52:43
  • Early code quality was deliberately bad — any everywhere, inline CSS — a trade-off to move fast while he knew the whole codebase. AI later cleaned it up cheaply, retroactively validating the shortcut: "that trade-off was so worth it" 53:03, 53:28, 53:54

"Easier to build, 10x harder to build value" (2026 hot take)

  • Because features are now trivial to ship, people ship features nobody cares about — cluttering products, slowing sites, making things worse. "Speed matters in business, but decisions matter as well" 54:14, 54:30, 54:58
  • Example: Anthropic moved so fast they didn't notice quality regressions for a month+ until users forced a public acknowledgment blog post 54:58, 55:37
  • The pre-AI "move fast and break things" slider still holds — AI did not repeal the speed/stability trade-off 56:01
  • OpenAI shutting down Sora as a social network to refocus on coding shows even the fastest-growing companies are choosing focus over breadth 56:25, 56:51
  • The irony: the lesson from observing AI companies is that focus beats executing quickly on everything — fittingly, the Transformers paper was titled "Attention is all you need" 57:15, 57:40