Sign InOpen Brain
VercelEngineering PostOfficial Source

How Factory scaled its cloud backend to tens of millions of daily requests on Vercel

Factory runs its full cloud backend in one Next.js app while agents deploy internal tools through Vercel APIs. The case study is useful evidence for keeping agent-operated infrastructure programmable and guarded.

Vercel · Aug 3, 2026
Open Source Open MarkdownOpen JSON
Source Summary

Factory says one Next.js backend handles **tens of millions of daily requests** at **350ms-or-better p95** across API routes, middleware, webhooks, and log drains. Its Droids also produce **dozens of deployments a day** through Vercel APIs.

Practical Implication

Builders should expose deployment and configuration through APIs that agents can operate, while reserving high-stakes decisions for human review. A unified backend can remain viable longer than expected if memory, duration, security, and observability controls scale with it.

Agent-Ready Context
Factory says one Next.js backend handles **tens of millions of daily requests** at **350ms-or-better p95** across API routes, middleware, webhooks, and log drains. Its Droids also produce **dozens of deployments a day** through Vercel APIs.

Builders should expose deployment and configuration through APIs that agents can operate, while reserving high-stakes decisions for human review. A unified backend can remain viable longer than expected if memory, duration, security, and observability controls scale with it.

This is a vendor case study, not an independent architecture comparison. It gives no cost profile, failure-rate data, workload distribution, or detail on the guardrails used for autonomous deployments.
Connected Context · Feed7 Judgment

This supplies scale evidence that agent-operated deployment can coexist with a unified web backend, shifting the question from whether agents can deploy to which controls make frequent production mutation governable. It reinforces observability, scoped authority, validators, and human review as operating requirements, but does not establish architecture superiority because cost, failures, workload mix, and deployment guardrails are omitted.

Vercel MCP can now deploy codeThe MCP capability provides the deployment mechanism that Factory’s Droids exercise at high frequency; Factory’s case raises the practical need for approvals, rollback, and failure handling left unspecified by the capability announcement.Project-scoped TokensProject-scoped credentials limit the blast radius of agents making deployments and configuration changes through Vercel APIs, directly supporting the case study’s security requirement.Agent Runs now available in the Vercel MCP and CLIAgent Runs traces supply the production observability needed to diagnose and improve the frequent automated delivery activity described in the case study.How Forward Deployed Engineering is done at Factory — Eno ReyesFactory’s workflow account explains the organizational guardrails behind deployment autonomy—instrumented paths, validators, explicit completion criteria, and human approval—that the infrastructure case study does not detail.
Context Map
infracodingsecurity#cloud-agents#tool-use#observability
Uncertainty
This is a vendor case study, not an independent architecture comparison. It gives no cost profile, failure-rate data, workload distribution, or detail on the guardrails used for autonomous deployments.