CascadiaJS 2026 · Notes
← All talks

It's Time to Rethink Everything

Theo · T3 Chat · Day 2

Building software is WAYYYY different because…

AI the cloud!

Life before the cloud

  1. Building was more capital-intense
  2. You had to predict the future
  3. Experimentation was expensive

Life after the cloud

  1. You can start small and scale up
  2. Experimentation is cheap
  3. You don't need to hire God-tier infra devs anymore

Easy scaling made a lot of things possible.

Amazon-tier hardware without Amazon-tier budget.

And AI… we've been through this before…

Life before the Claude

  1. Building was capital-intense
  2. You had to predict the future
  3. Experimentation was expensive

Amazon-scale software, with Amazon-scale SWE teams.

Everything we believe is no longer true.

Expertise required is a lot less now; things that were rigid aren't true anymore.

Things we should rethink

We may be past where this stuff makes sense:

  • Why can't we commit .env files?
  • Why can't a public repo have private parts?
  • Why can't a file be in two folders at once?
  • Why do we need a file system to compile my app?
  • Why do we assume our codebases still matter?
  • Why do we assume our packages still matter?
  • Why can't we rebuild everything?
  • Why can't we boil the ocean?

Salesforce

Salesforce is the opposite of everything mentioned…

Salesforce has a lot of features.

Everyone needs a small number, Fortune 500 needs some, and then there's the long tail of things 1% of users use.

Features [company] needs.

You have to build for every snowflake.

Or do you?????

What if you could cover what a company needs trivially?

  • breadth of features — range
  • depth of features — # of features in a given area

Too broad? Hard. Too deep? Not good.

But a broad range of things you cover — it's viable now.

The depth isn't a problem; build a platform and let the client deal with that with their agents.

What does this look like?

Before

  • build — 40 hours
  • deploy — 3 hours

After

  • build — 30 minutes
  • deploy — 3 hours

So obnoxious! Why!! The 3 hours sucks worse now.

shoo.dev

  • shoo.dev
  • auth in 2 lines of code
  • easy way to add auth, not secure or good necessarily
  • didn't ship? built too deep in the vertical

BOIL THE OCEAN!!! Make all the problems go away.

Theo went too far.

Lakebed [alpha]

LIVE DEMO!!

npx lakebed new sup-seattle
  • nothing special on his machine
npx lakebed deploy
  • deployed just like that, super quick
  • changed something, deployed in seconds

Built language, compiler, CLI, deployer.

Holy crap — so simple.

We kid ourselves that we need all the perfect things, just prompt something into existence.

lakebed.dev

Why haven't we been thinking right

WE ARE NOT BUILDING BIG ENOUGH.

Find the place where we feel STUPID for going so big.

Where will things be in 5 years? The last 5, 10, 50 years don't matter.

YOU AREN'T BUILDING BIG ENOUGH.

Referenced