Vercel Enterprise AI Platform
A platform handbook for designing enterprise AI products with AI SDK, AI Gateway, Workflow, Sandbox, and Queues.
Written and reviewed by Eric Han · Co-founder & CTO, reopt
Recently Updated Chapters
Use AI SDK as the application runtime layer for streaming, tools, agents, and telemetry.
Orchestrate repo-aware coding agents from issue intake to tests and pull request drafts.
Manage AI cost, latency, error budgets, rate limits, fallback, and workload classes.
Connect enterprise systems to AI agents through MCP tools, resources, and prompts.
Move from prototype AI features to a governed Vercel enterprise AI platform.
Platform baseline reviewed 2026-10-23
Runtime, observability, costs, and migration guidance now cover AI SDK 7, Workflow 5, and Node.js 24 on Fluid Compute.
npm versions checked: ai=7.0.148, workflow=5.2.3, @ai-sdk/workflow=2.0.80.
Existing application dependencies require a separate reviewed migration.
AI SDK 7 migration · Workflow 5 changes · Functions limits.
Enterprise AI is not just attaching a model to an app. It is a platform design problem: separate control plane, runtime plane, data access, quality, security, and cost governance.
This handbook organizes Vercel's AI platform capabilities into an enterprise operating model for platform engineers, AI infrastructure leads, and Staff+ engineers.
Core View
Durable AI systems need a clear split between application experience, model routing, long-running work, isolated tools, async jobs, and governance.
Platform Model
Maturity Model
| Level | State | Signal | Promotion condition |
|---|---|---|---|
| L1 Prototype | Single model and prompt | Manual operation | Call logs and owners exist |
| L2 Controlled | Gateway and key separation | Policies are visible | Long work moves to Workflow |
| L3 Reliable | Fallback, retry, tracing, isolation | Errors and cost are reviewed | Quality dashboards stabilize |
| L4 Governed | Team-level accountability | Audit and risk reviews operate | Quarterly architecture review |