Watch First¶
The full theory session.
Just under two hours covering everything that happens before you open the agent: how a language model actually works, why vibe coding falls apart in production, and the spec-driven workflow that replaces it.
Every hands-on chapter that follows assumes you already have this.
Chapters¶
Click any timestamp to jump straight there.
00:00 |
Welcome — what you're going to build |
06:53 |
The finished extension, demoed end to end |
16:48 |
How a language model actually works |
23:50 |
Context is everything — and it has a limit |
33:28 |
From model to agent: tools and the loop |
46:42 |
Giving the agent your context: rules and base code |
1:00:58 |
MCP servers — extending the agent's reach |
1:07:01 |
Which model should you use? |
1:14:01 |
Vibe coding, and why it breaks in production |
1:25:21 |
The constitution: tech design and roadmap |
1:31:11 |
Feature specs and the build loop |
1:50:25 |
How to use this portal |
1:54:36 |
Who owns the code |
Prefer reading?
The Foundations chapters cover the same ground in writing. Watch it, read it, or both — the practice afterwards is identical.
Set up before you build, not during
The recording shows the workflow running against a working environment. Get the prerequisites in place first — a sandbox and a frontier model — or you'll stall on the first feature.
Next¶
Theory done. Now you build the extension yourself, one feature at a time.