How General Intelligence used agents to build an agent platform on Vercel
By Steven Van ·
A case study on how the 8-person team behind Cofounder ships 10 PRs per engineer daily using agents on Vercel's infrastructure.
In a case study published by Vercel, General Intelligence, an 8-person startup with 5 engineers, described moving its infrastructure to Vercel to build Cofounder, an AI agent platform meant to let a single founder run a company through agents that cover engineering, marketing, SEO, finance, sales, support and operations. The company had been hosting on Render, but says Render's Python support couldn't keep up once it started building Cofounder's CTO agent, and it needed a cloud where every action a human could take, killing deployments, changing DNS, managing billing, editing configs, was also reachable by a coding agent through a CLI or API.
Since migrating, General Intelligence's engineers no longer develop locally: every change from the CTO agent goes straight to a Git branch, gets its own preview environment, and is tested on the live URL by a browser agent. The company reports:
- 10 PRs shipped per engineer per day, with 70+ commits each, on a $5,000-per-engineer monthly token budget
- Over 4,000 branches in flight, with around 100 app versions running in parallel at any point in a workday
- Fluid compute usage growing 6.5x month-over-month earlier this year, mostly from internal engineering work
- 90% of SRE work automated through Vercel and Cofounder itself
General Intelligence also migrated Cofounder's Python backend to Vercel so the agent could drive deployments, configs and compute across the whole stack rather than just the frontend. Cofounder now gives each customer a GitHub repository and a managed Vercel deployment, provisioned automatically through Vercel for Platforms, with its own domain and automatic SSL and DNS.
