How Superset Built an IDE for AI Agents on Vercel

Superset fully leveraged Vercel to build an IDE that runs up to 10 coding agents in parallel. Each agent runs in an isolated workspace, and every branch is automatically rolled out as a preview deployment, instantly getting a live URL.

The core of this architecture is eliminating bottlenecks that block parallelism. As the number of agents grows, if CI configuration, preview environment management, and deployment queues get serialized, tasks stretch from minutes into hours. To avoid this, Superset put 6 Next.js projects on Vercel from the start and focused on the product instead of platform engineering.

As the product grew, the team attached individual Vercel features as needed.

• AI SDK and AI Elements handle agent orchestration
• AI Gateway simplifies model routing
• Vercel Blob stores agent outputs and user files.
• Fluid compute absorbed a large number of parallel jobs, and Active CPU pricing meant costs were incurred only for actual computation rather than waiting time.
• Cron Jobs prevent environment buildup, and BotID filters out bots during traffic spikes.

Real-world usage results also show up in numbers. Internally, about 600 preview deployments are created per day, build times average around 30 seconds, and weekly deployments reach 1,000-1,400. Right after the Hacker News launch, user count tripled overnight, but no separate infrastructure work was needed, and incident response has settled into a flow where an agent generates a fix, spins up a preview, and reflects it within 30 minutes.