Hey, Vercel just launched something that could significantly change how we work with backends. It's called Workflows and basically solves a problem every developer faces: spending weeks building orchestration infrastructure when they should be focusing on the product.

The concept is actually very simple. Instead of scattering your logic across queues, workers, state tables, and retry mechanisms everywhere, you tag your TypeScript function with "use workflow" at the top and each step with "use step" in the child functions. Done. Vercel takes care of the rest: queue scheduling, failure retries, state persistence—all automatically without needing to deploy separate orchestration services or state databases.

What caught my attention most is that you only pay for the actual runtime. No fixed costs for services running 24/7 just waiting to do something. This completely changes the game for AI agents and backend tasks in production.

The numbers speak for themselves: since entering public testing last October, Workflows has processed over 100 million executions and 500 million steps. More than 1,500 customers using it, 200,000 weekly downloads on npm. That’s real adoption, not hype.

For those working specifically with AI agents, Vercel included some well-thought-out features. Flows are durable, meaning the agent’s output is stored persistently and continues even after you close the browser. When reconnecting, it resumes from where it left off. All encrypted by default before leaving your environment. You can suspend and resume waiting for human approval or leave it paused for days, months, and pay nothing during suspension.

It also supports heavy loads: up to 50 MB per step, 2 GB per execution. It frees up space to work with images and videos in multimodal agents.

The Workflows SDK is open-source and supports self-hosting through an adapter system called "Worlds." The community is already developing adapters for MongoDB, Redis, Cloudflare, and other tools. They also released AI SDK v7 with the integrated WorkflowAgent, and the Python SDK is in public testing.

The next version, Workflows 5, will bring native concurrency control with locking mechanisms between executions, global deployment infrastructure, and a snapshot-based runtime to reduce reprocessing costs. Vercel is clearly thinking big on this.