When Cloudflare launched Workers eight years ago, it was a direct-to-developer platform. Over time, the ecosystem evolved to support multi-tenant applications, allowing platforms to let their own customers ship code through Workers. This shift introduced dynamic compute, storage, and source control—via Dynamic Workers, Durable Object Facets, and Artifacts. Now, Cloudflare bridges two worlds: durable execution and dynamic deployment with Dynamic Workflows. This Q&A explores what that means for developers building scalable, tenant-aware applications.
What is Dynamic Workflows and how does it differ from traditional Workflows?
Dynamic Workflows is a new capability that combines Cloudflare's durable execution engine (Workflows) with the ability to deploy workflow code dynamically at runtime. Traditional Workflows require the workflow code to be part of your deployment—a single class defined in wrangler.jsonc that the engine executes. This works for monolithic apps but fails when each tenant needs a different workflow. Dynamic Workflows solves this by letting you hand the engine arbitrary TypeScript code at runtime, just like Dynamic Workers does for compute. The engine then runs that code as an isolated, durable workflow where every step survives failures, can sleep for hours, and resumes exactly where it left off. This means platforms can offer each customer their own custom pipeline without pre-deploying it.
How does Cloudflare's platform evolution from direct-to-developer to multi-tenant relate to Dynamic Workflows?
Originally, Workers was a single-tenant platform: developers wrote code and deployed it directly. As the ecosystem grew, Cloudflare introduced primitives for multi-tenancy—Dynamic Workers for compute, Durable Object Facets for storage, and Artifacts for versioned file systems. Each allows a platform to spin up isolated resources per tenant on demand. Dynamic Workflows extends this same philosophy to durable execution. Now, a platform can not only give each tenant an isolated Worker and database but also a custom, long-running workflow that the tenant defines. This completes the vision: a fully dynamic, multi-tenant stack where every aspect—compute, storage, source control, and execution—can be created and destroyed per-tenant without manual deployment.
What are the key components that paved the way for Dynamic Workflows?
Three components set the stage. First, Dynamic Workers allowed platforms to hand arbitrary code to the Workers runtime and receive an isolated Worker in milliseconds. Second, Durable Object Facets gave each dynamically-loaded app its own SQLite database, spun up on demand with the platform acting as a supervisor. Third, Artifacts provided a Git-native, versioned filesystem that can be created in tens of millions—one per agent, session, or tenant. These three solved dynamic deployment for compute, storage, and source control. The missing piece was durable execution: the ability to run long-lived, fault-tolerant workflows that are also dynamic per tenant. Dynamic Workflows fills that gap, completing the dynamic deployment suite.
How did Cloudflare Workflows handle durable execution before Dynamic Workflows?
Cloudflare Workflows is a durable execution engine that lets developers write a run(event, step) function where every step survives failures, can sleep for hours or days, wait for external events, and resume exactly where it left off. It's designed for tasks like onboarding flows, video transcoding, multi-stage billing, and long-running agent loops. With Workflows V2, it supports up to 50,000 concurrent instances and 300 new instances per second per account, optimized for the agentic era. However, Workflows had one strict assumption: the workflow code must be part of your deployment—one class bound in wrangler.jsonc. This meant you couldn't vary workflows per tenant or request. Dynamic Workflows removes that restriction by allowing workflow code to be provided at runtime.
What problem does Dynamic Workflows solve for multi-tenant platforms like AI code generation or CI/CD?
Consider a platform where an AI writes TypeScript tailored to each tenant's business logic. Or a CI/CD product where every repository defines its own pipeline. Or an agent SDK where each agent creates its own durable plan. In all these cases, the workflow is different for every tenant, agent, or request. With traditional Workflows, you'd need to pre-deploy a single class—impossible when the workflow code is unknown until runtime. Dynamic Workflows solves this by letting platforms hand the engine the workflow code on the fly. The code is isolated, sandboxed, and durable. This empowers platforms to offer custom, long-running workflows to each of their customers without manual intervention, scaling to thousands of unique workflows concurrently.
What are the capabilities and limits of Workflows V2?
Workflows V2 is designed for the agentic era, supporting up to 50,000 concurrent instances and 300 new instances per second per account. It's built to handle massive parallelism for tasks like agent loops, which spawn many sub-workflows. However, it still assumes a static deployment—the workflow class is part of your project. Dynamic Workflows extends V2 by allowing dynamic deployment of that class. Limits now include the ability to create hundreds of unique workflow definitions per second, each sandboxed. The underlying durability remains: steps survive failures, can sleep indefinitely, and handle external events. For platforms, this means they can offer each tenant a custom, durable workflow without worrying about deployment bottlenecks or resource contention.