A new programming model for durable execution

Vercel Workflows is now generally available after processing 100M+ runs and 500M+ steps across 1,500+ customers in beta. It introduces a programming model where durable, long-running functions are written in TypeScript or Python using 'use workflow' and 'use step' directives — no separate orchestrator, Kubernetes, or queue infrastructure required. The system is built on an event log, Fluid compute, and Vercel Queues, with end-to-end encryption by default. Key features include durable streams (resumable after disconnection), hooks for human-in-the-loop flows, sleep primitives, and deep AI SDK integration for building resilient agents. The SDK is open source and portable via a 'Worlds' adapter system supporting Postgres, MongoDB, Redis, and Cloudflare. Upcoming Workflows 5 adds native concurrency controls, globally deployed infrastructure, and snapshot-based runtime to reduce replay overhead.