Introducing Dynamic Workflows: Durable Execution that Follows the Tenant

Why It Matters

Dynamic Workflows enable multi‑tenant platforms to offer fully customizable, durable automation without provisioning dedicated containers, dramatically lowering costs and unlocking new SaaS use cases.

Key Takeaways

• Per‑tenant workflows run with full Workflows durability
• Dynamic Workers boot tenant code in single‑digit milliseconds
• Zero idle cost; active tenants share the same hardware
• Supports AI‑generated agents, low‑code builders, CI/CD pipelines
• Scales to millions of tenants with minimal overhead

Pulse Analysis

Cloudflare’s Workers ecosystem has evolved from a developer‑centric compute layer to a full‑stack, multi‑tenant platform that can host dynamic code, storage, and source control. The latest addition, Dynamic Workflows, fills the missing piece for durable execution, allowing each customer to ship their own workflow logic at runtime. By leveraging a small ~300‑line TypeScript library, a Worker Loader identifies the tenant, fetches its code, and wraps the standard Workflows binding so that the engine can resume execution later, even after hours or days, while always routing back to the correct tenant sandbox. This architecture preserves the robust features of Cloudflare Workflows—step retries, sleep intervals, event waiting—while eliminating the need for pre‑deployed classes or per‑tenant containers.

From a technical standpoint, Dynamic Workflows operate in three layers: the Workflows engine, the Worker Loader, and the tenant’s dynamic worker. When a request creates a workflow instance, the loader injects metadata that travels with the payload, enabling the engine to invoke the correct tenant code on every subsequent step. The system caches isolates, so repeated steps reuse the same sandbox, and isolates are evicted when idle, keeping resource consumption near zero. Because the code boots in milliseconds and uses only a few megabytes of memory, platforms can support millions of distinct workflows without incurring traditional VM provisioning costs.

The business implications are profound. AI‑driven agents can now write and run their own durable plans, low‑code platforms can expose a true “code‑as‑a‑service” model, and CI/CD systems gain per‑repository pipelines that start in seconds and hibernate for free during approvals. This reduces latency, cuts infrastructure spend, and opens new revenue streams for SaaS providers that need to scale custom automation to tens of millions of users. Dynamic Workflows thus positions Cloudflare as a foundational layer for the next generation of programmable, serverless SaaS applications.