How Superset built the IDE for AI agents on Vercel
By Steven Van ·
Six Next.js projects, 600 daily preview deployments and 30-second builds underpin Superset's parallel-agent IDE, built entirely on Vercel.
Superset, an IDE that runs up to 10 coding agents in parallel, each in its own isolated workspace, built its platform on Vercel from day one. Founders Kiet Ho, Satya Patel, and Avi Peltz, all former CTOs at YC-backed companies, had used Vercel at previous companies and ran six Next.js projects on it without a platform team: the web app, marketing site, docs, and three supporting services.
Every branch a developer or agent creates becomes a preview deployment automatically, reaching around 600 preview deployments a day at peak. Superset's AI stack runs on several Vercel pieces:
- AI SDK and AI Elements handle orchestration across multiple models and agents
- AI Gateway routes models without custom routing logic
- Vercel Blob stores artifacts from agents and users
- Fluid compute absorbs parallel tasks as agents fan out, and Active CPU pricing charges only for compute used, not time waiting on model responses
- Cron Jobs stop parallel environments from piling up, and BotID filters bots during traffic spikes
Superset also uses its own product internally, splitting incoming GitHub issues across parallel workspaces and running up to a dozen agent instances at once. During a Hacker News "Show HN" launch, user counts tripled overnight without manual provisioning, and the team says a reported bug can go from fix to merged preview in under 30 minutes, with instant rollbacks if a deploy goes wrong. Average build time is around 30 seconds, deployment volume runs 1,000 to 1,400 a week, and weekly active users have grown 57 to 64% week over week.
