What changed
Vercel said on July 24 that Workflow steps now support extended function durations. On Pro and Enterprise plans, a workflow step can now run for up to 30 minutes, up from 800 seconds, if you opt in with VERCEL_ENABLE_WORKFLOW_EXTENDED_MAX_DURATION=1. The feature requires Fluid Compute and a supported Node.js or Python runtime.
This builds on Vercel's June 15 rollout where Vercel Functions can now run up to 30 minutes. That earlier change raised the base execution ceiling for supported functions; the July 24 update is the product step that makes the same ceiling usable inside workflow steps.
Why it matters
This is more important than a simple timeout bump. Vercel's AI SDK 7 launch framed WorkflowAgent as the durable runtime for long-running agents, with execution state persisted between steps so runs can survive deploys, restarts, interruptions, and delayed approvals.
Inference from those Vercel sources: longer per-step limits reduce the amount of orchestration builders need to do around heavy tool calls. Research loops, browser automation, long reasoning passes, OCR pipelines, and human-in-the-loop flows can stay inside the workflow model longer before you need to split work across extra queues or custom retry systems.
Practical takeaway
The interesting product move is that Vercel is narrowing the gap between agent demos and production agent jobs. A lot of "agent infrastructure" work is really timeout management, retries, and deciding which tool calls must move into a separate backend. By extending workflow steps to 30 minutes, Vercel is making more of that problem look like deployment configuration instead of platform glue.
For teams already using Vercel for AI apps, this makes Workflow-based agents more credible for tasks that were previously awkward at 800 seconds. It does not remove the need for good step design, but it makes long-running agent behavior feel more like a first-class platform feature.