Theo Browne's CascadiaJS 2026 talk arguing that AI is a "new cloud moment" — just as the cloud removed the cost of provisioning servers, agents remove the cost of building, so the sacred rules of software (file systems, codebases, packages, git, deployment) are worth tearing down and rebuilding from first principles.
Recorded live at CascadiaJS 2026 in Seattle.
The cloud changed software by removing the need to pre-acquire servers and predict scale — experimentation got cheap, and products like Slack or Salesforce became possible. Theo's argument: AI is the same kind of shift. You now get Amazon-scale software without Amazon-scale teams, which means the assumptions baked into how we build are worth re-examining.
Now is the time to try the thing that feels stupid because every preconceived notion we have doesn't matter anymore.
Software used to win by going deep on one vertical slice, because covering the full breadth meant missing features everyone needed. With agents on both sides, you can cover the breadth and let users prompt the depth — the special-snowflake features become their problem, not yours.
To prove the point, Theo demos Lakebed — npx lakebed new, npx lakebed deploy, and you get a live app with real sync and working auth across platforms, no dashboards or env vars. Building it meant rebuilding the CLI, a frontend framework, a backend framework, a bundler, a language runtime, and a compiler to run in isolates. It's "a shitty cloud for shitty apps" — not how you'd build the next Facebook, but proof we don't have to build the way we did before.