Station 08 · pass at 75%
Ship from the chat
The Luma coordinator reads 187 RSVPs, finds 32 lightning-talk hopefuls, builds a dashboard, then — when eight slots are not enough — spins a QR app on Next.js, Convex, and Clerk via a Cursor cloud agent. Ray never opened the Cursor site.
2:20 – 6:30
Luma night: guests to dashboard to app
2:20 – 6:30 · The first telling of the event coordinator loop.
53:22 – 56:38
The same story with the PR on the phone
53:22 – 56:38 · He retraces it: cloud agent, merge, QR at the door, phone thread.
Thesis
Computer use gathers the facts. Plugins ship the artifact. P-Stack is how the artifact is not slop. That chain is the product.
- Cloud computer is already in Luma: it writes event copy and guest questions.
- Responses arrive as a giant spreadsheet — perfect bot work.
- Dashboard first. App only when the dashboard proves you have a new problem (32 speakers, 8 slots, no pitch decks).
- Handoff: 'open in Cursor' → cloud agent (he mentions Opus) → merge to main with P-Stack.
- He still has to paste Convex/Clerk keys. The human remaining work is credentials, not scaffolding.
Field work
Spec a ship-from-chat app
Do not open an editor. Write the smallest app your bot is allowed to emit after it has seen real data. Scope is a weapon.
Station quiz
Pass at 3/4. You can retake.
1.About how many people had signed up for the Convex HQ event?
2.Why did a dashboard turn into an app?
3.What stack did the bot reach for?
4.What did Ray still have to do himself?