Shipping AI copilots in a weekend
How I build and deploy tiny AI assistants with rapid feedback loops.
Launch velocity matters. I treat every AI sidekick like a Friday hackathon project: scope it to one magical moment, instrument everything, and ship while the excitement is still peaking.
My current weekend stack
- Figma storyboard to map the conversation flow in under 30 minutes.
- TypeScript edge functions on Vercel with a thin Remix-style router for prompts.
- PostHog event stream hooked to every model call and human override.
- Airtable ops dashboard so teammates can nudge instructions without redeploying.
The real unlock is capturing live feedback. I record every improvement suggestion as a structured ticket. By Sunday, the backlog writes the next prompt iteration for me.
Shipping beats polishing. Users remember the velocity more than the pixel-perfect UI.
Next up: migrate the telemetry layer to a shared event bus so each new copilot inherits analytics, guardrails, and release notes without extra wiring.
Keep reading
Building the perfect Black Jack assistant
Using computer vision in python to create a black jack assistance HUD
Read postJun 14, 2024The personal website stack I keep rebuilding
Notes on typography, motion, and polish from version 7 of this site.
Read postApr 30, 2024Ship notes #14
A monthly roundup of experiments, wins, and bugs worth remembering.
Read post