Cloudflare Unveils Dynamic Workflows: Durable Execution Now Follows the Tenant
Breaking: Cloudflare bridges durability and multi-tenant flexibility with new Dynamic Workflows engine
Cloudflare today announced the launch of Dynamic Workflows, a new feature that lets platforms run durable, fault-tolerant workflows that are unique to each tenant—without requiring developers to pre-bind workflow code at deployment time. The move closes a critical gap in the company's serverless ecosystem.
“With Dynamic Workflows, we’re giving multi-tenant platforms the ability to define workflows dynamically, at runtime, just like we already did for compute and storage,” said Sarah Chen, Cloudflare’s VP of Product for Developer Platforms. “Now an AI-written pipeline, a CI/CD rule, or an agent’s plan can be durable and survive failures—without the platform having to know the code in advance.”
Background: The evolution of dynamic primitives
When Cloudflare first launched Workers eight years ago, it was a direct-to-developer platform. Over time, the company expanded so that platforms could build multi-tenant applications where customers ship custom code. Last month, the company shipped the Dynamic Workers open beta, allowing platforms to hand code to the Workers runtime and get back an isolated, sandboxed Worker in milliseconds.
That was followed by Durable Object Facets for dynamic storage—each app gets its own SQLite database on demand—and Artifacts for dynamic, Git-native source control. “We had dynamic deployment for compute, storage, and versioning,” explained Chen. “But the one piece missing was durable execution that follows the tenant.”
What Dynamic Workflows solves
Cloudflare Workflows is the company’s durable execution engine. It turns a simple run(event, step) function into a program where every step survives failures, can sleep for hours, wait for external events, and resume exactly where it left off. Until now, the workflow code had to be part of a deployment—a single binding per class per deploy.
That assumption breaks when platforms need to run custom workflows for every tenant. “If you’re building an AI coding platform where the LLM writes TypeScript per tenant, or a CI/CD product where each repo has its own pipeline, you can’t pre-declare a single Workflow class,” said Chen. “Dynamic Workflows removes that restriction.”
The new service allows platforms to hand a workflow definition to the runtime at the same time they hand compute code. The engine then creates a durable, stateful instance that is uniquely associated with that tenant, agent, or session—just like Dynamic Workers do for compute.
What this means for platform builders
With Dynamic Workflows, any multi-tenant platform can now offer each customer a fully isolated, durable workflow that survives server restarts and network blips. Developers building AI agents, CI/CD products, app marketplaces, or SaaS platforms with per-customer logic gain a new level of flexibility.
“This is a massive unlock for the agentic era,” said Chen. “An agent that writes its own plan can now make that plan durable, so if the isolate is recycled mid-task, the agent simply resumes from the last completed step.” The feature also integrates with the existing Workflows V2 engine, which already handles up to 50,000 concurrent instances and 300 new instances per second per account.
Key takeaway: Dynamic Workflows extends the “code follows tenant” pattern to stateful, long-running execution. Platforms no longer have to force-fit every customer into a single static workflow.
What’s next for dynamic deployment
Cloudflare says it plans to continue expanding the dynamic deployment suite, with future releases likely focusing on better developer tooling and deeper integration with the company’s AI platform. “We started with compute, added storage and versioning, now execution,” concluded Chen. “The goal is to give platforms every primitive they need to run code that is truly per-tenant—without sacrifice.”
Related Articles
- Kubernetes v1.36: Enhanced Staleness Detection and Controller Observability
- Serverless Spam Detection API: Deploying a Scikit-Learn Model with AWS Lambda and API Gateway
- Inside the Pentagon's $17.9 Billion Golden Dome Laser Defense Program
- How to Implement Managed Daemons for Amazon ECS Managed Instances
- Clean Up Your Photo Library One Day at a Time: A Step-by-Step Guide to Using 'This Day'
- Mastering Kubernetes Controller Health: New Staleness Solutions in v1.36
- Microsoft Launches Smart Tier for Azure Storage: Automated Cost Optimization Now Generally Available
- 10 Critical Actions to Secure Your Software Supply Chain Today