solo-saas
Build on Vibe team
Killing a 464-page publish timeout
How the wrong prompt killed 25 consecutive publishes and the one-line fix that shipped the site.
Revenue
N/A (internal)
Time to build
1 weekend of pain
Launched
July 2026
Location
The story
We wired a drip system into our Command Center that auto-generated hundreds of SEO pages. Once the repo crossed 464 static pages, every publish failed with the same 'files missing' page and 25 retries didn't help. The Puppeteer prerender step chained into our `vite build` was silently exceeding the publish timeout. Only when we changed the prompt from 'publish again' to 'watch the streaming logs and tell me what's actually failing' did the AI diagnose the real cause. We moved prerender to `bun run build:full` and reduced publish time from 300+ seconds (killed) to 25 seconds (shipped). The lesson was worth more than the fix: your prompt shapes what the AI can see.
Stack
Lovable
Vite
Puppeteer
Supabase
Reported metrics
Failed publishes
25+
Time to fix once diagnosed
60s
New publish time
25s
Sources
Public contact
The full build breakdown
Lessons
- Never repeat the same command 25 times. The AI is not learning between attempts.
- Add temporal context to your prompts ('I added X at time Y').
- Force diagnosis before action — 'watch the logs' beats 'try again'.
- Keep your critical publish path lean. Move slow steps (crawls, prerenders, submissions) to separate scripts.
Pitfalls
- Chaining Puppeteer to `vite build` in package.json.
- Trusting the retry loop to eventually work.
- Not checking whether recent feature additions touch the build/publish pipeline.
Publications & press mentions
Live searches across the biggest tech publications, podcasts and communities covering Build on Vibe team.
Want to be the next case study?
Ship your product with Build on Vibe and share your story with our community.