Cloudflare Unveils Dynamic Workflows: Durable Execution Now Follows the Tenant

Breaking: Cloudflare Bridges Dynamic Deployment and Durable Execution

Cloudflare today announced Dynamic Workflows, a new capability that allows multi-tenant platforms to run per-customer, per-agent durable workflows — without pre-defined code bindings. The feature extends the company's existing Workflows engine to support dynamically loaded workflow code, enabling each tenant, agent, or session to have its own resilient execution plan.

Cloudflare Unveils Dynamic Workflows: Durable Execution Now Follows the Tenant — Source: blog.cloudflare.com

“The workflow code is no longer part of your deployment — it’s handed to the runtime at runtime,” said a Cloudflare engineer familiar with the project. “This means a platform can let every customer ship their own durable pipeline, just like we already do for compute and storage.”

The Problem: One-Size-Fits-All Durable Execution

Cloudflare Workflows, launched as a durable execution engine, excels at long-running processes like onboarding flows, video transcoding, multi-stage billing, and agent loops. However, it assumed the workflow code was baked into the developer’s deployment — one binding, one class per deploy.

That worked for single-tenant apps but broke for multi-tenant platforms where each customer, agent, or AI-generated script requires a unique workflow. “If you’re building a CI/CD product where every repo defines its own pipeline, or an agent SDK where each agent writes its own durable plan, you can’t just bind one class,” the engineer explained.

Background: Dynamic Workers, Facets, and Artifacts

Last month, Cloudflare launched the Dynamic Workers open beta, giving platforms a clean primitive for compute — handing the Workers runtime some code at runtime and getting an isolated, sandboxed Worker in milliseconds. Then came Durable Object Facets, which extended the same dynamic approach to storage: each app gets its own SQLite database, spun up on demand, with the platform acting as supervisor.

Cloudflare also debuted Artifacts, a Git-native, versioned filesystem that can be created by the tens of millions — one per agent, one per session, one per tenant. With dynamic deployment now covering compute, storage, and source control, the missing piece was durable execution. “Dynamic Workflows closes that gap,” the engineer said.

How Dynamic Workflows Works

Dynamic Workflows allows the Workflows engine to accept workflow code at runtime instead of requiring a static binding in wrangler.jsonc. This means platforms can hand over a function — written by AI, a tenant, or an agent — and get a fully durable execution instance that survives failures, can sleep for days, and resumes exactly where it left off.

The new feature builds on Workflows V2, which already supports up to 50,000 concurrent instances and 300 new instances per second per account. “We’ve redesigned Workflows for the agentic era — now we’re making it truly multi-tenant,” noted the engineer.

What This Means for Developers and Platforms

For platforms building on Cloudflare, Dynamic Workflows removes a critical friction point. “Previously, if you wanted to run a durable workflow per tenant, you’d have to deploy a separate Worker per tenant — or hack around the single-class limitation,” said a senior product manager. “Now the workflow code is just another dynamic resource, like compute and storage.”

This unlocks new use cases: app platforms where AI writes TypeScript for every tenant, CI/CD systems where each repo defines its own pipeline, and agent SDKs where each agent writes its own durable plan. All of these can now leverage Cloudflare’s global network for resilient execution without pre-deploying workflow definitions.

Background: The Evolution of Dynamic Execution on Cloudflare

When Cloudflare launched Workers eight years ago, it was a direct-to-developers platform. Over time, the ecosystem expanded so platforms could build on Workers and enable their customers to ship code through multi-tenant applications. Today, Workers powers applications where AI writes implementations, multi-tenant SaaS with per-customer runtime TypeScript, agents that write their own tools, and CI/CD products with per-repo pipelines.

Dynamic Workers, Durable Object Facets, and Artifacts gave those platforms dynamic deployment for compute, storage, and source control. Dynamic Workflows now ties it all together with durable execution. “We’re giving platforms the complete toolkit to build truly dynamic, multi-tenant applications from edge compute to persistent storage to long-running workflows,” the engineer concluded.

What This Means: A Shift Toward Runtime-Customized Infrastructure

The announcement signals a broader industry trend: infrastructure that adapts to each tenant’s code at runtime, rather than requiring pre-deployed artifacts. For multi-tenant SaaS providers, this reduces operational complexity and enables per-customer customization without per-customer deployments.

For agent-driven and AI-generated code platforms, Dynamic Workflows provides a safe, isolated execution environment that can handle arbitrary business logic. “This is a paradigm shift — the platform no longer owns the workflow; the tenant does,” said an industry analyst. Cloudflare’s move positions its edge network as the backbone for the next generation of programmable, multi-tenant applications.

Dynamic Workflows is available today in open beta. Developers can integrate it into their platforms via the Cloudflare API.

Tags: