Shipping five MVPs in five days
What the nightly-builder pipeline taught me about scope, demos, and the cost of a good first frame.
The pipeline picks a Hacker News topic at 20:00 CET, I greenlight it, an agent builds the MVP overnight, and a 30-second demo gets stitched together before breakfast. That’s the contract. Five nights, five demos, no retries.
What worked
The constraint forces scope cuts. When the deploy window is six hours and the demo has to render itself, there’s no room for “wouldn’t it be cool if”. Every feature has to earn its place in the first frame.
What broke
The demo pipeline kept producing files where the moov atom landed at the end of the MP4 —
fine for desktop, fatal on the gallery card preview. The fix was a one-liner (-c copy -movflags +faststart) but I lost an evening chasing it as a transcode bug.
What’s next
Locking the faststart pass into mvp-finalize so the next builder agent never has to think about it. The whole point of a pipeline is that lessons get baked in, not relearned.