Vercel 上で AI エージェント向け IDE を構築した Superset の取り組み

Superset on Vercel

1,000–1,400 deployments per week

~600 preview deployments per day

~30 second average build time

57–64% week-over-week DAU growth

Software development with AI started as a single engineer chatting with a single agent about a local repo. Today, developers direct fleets of agents in the cloud, but traditional tools were built for the old shape of the job: IDEs, terminals, and review systems designed for one developer moving one ticket at a time.

Co-founders Kiet Ho, Satya Patel, and Avi Peltz, all former CTOs at YC-backed companies, built Superset as the IDE for multi-agent development. It runs up to 10 coding agents in parallel, each in its own isolated workspace. Developers use it to direct teams of agents generating code across multiple branches simultaneously.

Running a team of agents in parallel changes what the platform underneath has to do. The product Superset offers its users only feels parallel because nothing on the platform forces the work to wait. If any layer slows down, even briefly, the parallelism on top collapses with it.

Parallel agents need parallel infrastructure

This workflow has a dependency that's invisible from the product surface. Every agent thread needs its own isolated environment, every branch needs a live URL, and every change needs a safe place to run.

Without instant provisioning, parallel agents stop being parallel. CI pipelines have to be configured per branch, preview environments have to be managed by hand, and deploys back up behind one another. For a team running a dozen agents at once, that serialization is what breaks the product. Twelve workflows collapse into one queue, and a task that should take minutes takes hours. The developer is back to waiting, which is the exact problem Superset exists to solve.

Six Next.js projects, no platform team

Vercel was the default choice from the start, as all three founders had built on it at previous companies. From day one, Superset ran six Next.js projects on Vercel: the web app, marketing site, docs, and three supporting services. The team skipped platform engineering entirely and stayed focused on the product.

Every branch a Superset developer or agent creates becomes a preview deployment automatically, often spinning up multiple services. At its peak, Superset generates roughly 600 preview deployments a day internally. Every branch gets a live URL, and the team never waits on a deploy queue.

One platform for every workload

The stack grew with the product, and each piece was pulled in to solve a specific problem as functionality was added:

Vercel Blob stores artifacts from agents and users, no object storage to manage.

Cron Jobs prevent parallel environments from piling up.

BotID filters bots during high traffic periods, no custom middleware needed.

The agent orchestration itself runs on the AI SDK and AI Elements, with AI Gateway handling model routing. Fluid compute carries the scaling load underneath, absorbing parallel tasks as agents fan out without forcing the team to rearchitect.

As Superset has expanded into new product areas, the entire stack has stayed on Vercel. There's no second cloud to glue in, no orchestration layer to maintain, and no platform engineering team to keep it glued together. New surface areas gets built on the same primitives that handled the old surface area, which is what frees the team to keep moving on product instead of plumbing.

Superset is its own super user

The most credible proof how the Superset team uses Superset themselves. GitHub issues flow into Superset and get split across parallel workspaces, and Satya has tuned the team's setup to run up to a dozen instances at once. Multiple efforts move forward without anyone waiting on serial decisions. Compared to their previous dev workflows, Superset's commit graph looks exponential.

Scaling through a Hacker News spike

During a Hacker News "Show HN" launch, user counts tripled overnight. Superset absorbed the spike without anyone provisioning infrastructure mid-flight.

That extends to incidents. If a customer reports an issue to Superset, their agents can spin up, write the fix, generate a preview, and merge the code in under thirty minutes. If the fix makes things worse, rollbacks are instant, so the cost of a bad deploy drops to near zero.

"Almost no time to deploy" as the bar

For Superset, immediate deployment matters because it keeps the loop between writing code, previewing it, and shipping it short enough that velocity never stalls, even across dozens of parallel workstreams. Build time averages around 30 seconds, and deployment volume runs between 1,000 and 1,400 a week.

What's next

The pattern for success is already clear: a product built for parallelism, by a team that works in parallel, on agentic infrastructure that doesn't force them back into a queue. Every new agent capability they ship to customers gets stress-tested first by their own engineers running a dozen at once. The dozen will become two dozen, and the infrastructure underneath was built to expect it.

About Superset: Superset is built by a team of three ex-YC CTOs and its the IDE for the AI agents era, letting developers run multiple coding agents in parallel.

Read more