Superset is an IDE for running up to 10 coding agents in parallel, each in an isolated workspace, built by three ex-YC CTOs: Kiet Ho, Satya Patel, and Avi Peltz. The product generates 1,000 to 1,400 deployments per week, roughly 600 preview deployments per day, with an average build time of 30 seconds. Week-over-week DAU growth sits at 57 to 64%. When a Hacker News launch tripled user counts overnight, no one provisioned infrastructure manually.
The infrastructure case is specific. Superset runs six Next.js projects on Vercel with no dedicated platform engineering team. Every branch becomes a preview deployment automatically. Vercel's AI SDK and AI Elements handle multi-model agent orchestration. AI Gateway manages model routing without custom logic. Vercel Blob stores agent artifacts. Fluid Compute with Active CPU pricing means the team pays for compute, not idle wait time on model responses. Cron Jobs clean up parallel environments before they pile up. BotID filtered bots during the HN spike without custom middleware.
The part worth reading in full is how Superset eats its own cooking. GitHub issues flow into Superset, get distributed across parallel workspaces, and Satya Patel runs up to a dozen instances simultaneously for their own engineering work. Customer-reported bugs can be fixed, previewed, and merged in under 30 minutes, with instant rollback if the fix is wrong. The article details exactly which Vercel primitives map to which agent workflow problems, making it a concrete infrastructure blueprint for any team building on top of parallel AI workloads.
[READ ORIGINAL →]