A new programming model for durable execution
By Steven Van ·
Vercel Workflows exits beta after 100M+ runs, adding a beta Python SDK, an AI SDK v7 WorkflowAgent, and real-time streaming to clients.
Vercel Workflows is now generally available, out of the beta it entered in October 2025. Since then it has processed over 100 million runs and 500 million steps across more than 1,500 customers, with over 200K npm downloads a week.
- Agents: deep AI SDK integration supports durable agents that keep state and tools running indefinitely and can handle external events or interruptions; AI SDK v7 adds a WorkflowAgent built on this.
- Backends: the Workflow SDK, previously TypeScript-only, is now also available for a new language, with a Python SDK in beta.
- Long-running workloads: workflows can handle multi-step onboarding flows, payment processing, ETL pipelines, or other backend work that would otherwise need custom queues and retry logic.
Workflows can also stream data to clients in real time as a run executes, rather than waiting for the whole workflow to finish. Steps write to the run's stream with getWritable(), and a client reads it via the run's readable stream, which is useful for progress updates or incrementally delivered AI-generated content. On Vercel, these streams are backed by persistent, Redis-based storage, so they keep their state across workflow suspensions and reconnects.
