A new programming model for durable execution
Not a pilot. Mux, Durable, and Flora are already running 50+ AI models through Vercel Workflows—no queues, no orchestration layer, zero config.
Why it matters
Vercel Workflows GA shifts how teams build production AI agents and long-running systems. By baking durability, retries, and encryption into application code instead of separate orchestration services, it removes infrastructure complexity that typically delays AI product launches by months.
The key facts
12 to knowVercel Workflows GA launched April 16, 2026
100M+ runs and 500M+ steps processed since beta (Oct 2025)
1,500+ customers, 200K weekly npm downloads
Python SDK now in beta alongside TypeScript
50 MB per step payload, 2 GB per run (supports multimodal agents)
Mux using Workflows for video AI inference pipelines
Durable rewrote entire site-creation pipeline on Workflows (160+ directives, 75 files, ripped out self-hosted infra)
Flora orchestrates 50+ image models with recursive workflows and durable streams
Built-in E2E encryption (data encrypted before leaving deployment)
Durable streams allow clients to reconnect and resume from last event
WorkflowAgent native implementation in AI SDK v7
Open-source Workflow SDK with community-built adapters (MongoDB, Redis, Turso, Cloudflare)
Go to the source
Vercel Blogvercel.com
Publisher excerpt: The gap between prototypes and production-ready systems is huge. Code that's trivial to run locally falls apart the moment it needs to handle failures, restarts, and real traffic. Framework defined infrastructure solved this for web applications. When you deploy, Vercel infers the right…