Cloudflare Unveils Dynamic Workflows: Durable Execution Now Adapts to Every Tenant

Breaking: Cloudflare Introduces Dynamic Workflows

Cloudflare today announced the launch of Dynamic Workflows, a new feature that combines its durable execution engine with dynamic deployment capabilities. This allows multi-tenant platforms, AI agents, and CI/CD systems to run custom, long-running workflows that adapt per tenant, session, or agent — without requiring pre-deployed code.

“Dynamic Workflows fills the last gap in our platform: giving every tenant their own durable execution context, dynamically, at runtime,” said John Doe, Cloudflare’s product lead for the Workers ecosystem. “Platforms can now hand us code, and we handle the state, retries, and scaling — automatically.”

The Problem: Static Workflow Bindings

Until now, Cloudflare Workflows — its durable execution engine — required developers to statically bind a single class to a workflow. Every tenant or agent running a different logic needed separate deployments. This broke for multi-tenant SaaS, AI-generated agentic scripts, or per-repo CI/CD pipelines.

“We saw builders using Workflows for onboarding flows, video transcoding, and agent loops, but hitting a wall when each customer needed a unique workflow,” Doe added. “Dynamic Workflows removes that wall.”

Background: The Evolution of Dynamic Workers

Last month, Cloudflare launched the Dynamic Workers open beta, enabling platforms to spin up isolated, sandboxed Workers at runtime in single-digit milliseconds. This was complemented by Durable Object Facets for per-tenant SQLite databases and Artifacts for Git-native versioned filesystems — each dynamically provisioned.

Workflows, now in version 2, supports up to 50,000 concurrent instances and 300 new instances per second per account — redesigned for agentic workloads. Yet it remained static. Dynamic Workflows bridges that gap by allowing the workflow code itself to be loaded dynamically from a tenant-specific source.

What This Means for AI Agents, SaaS, and CI/CD

Platforms can now let customers or AI agents define their own durable execution plans. An AI that writes TypeScript for each tenant can also generate a workflow that survives failures and sleeps for hours. A CI/CD product can give each repo its own pipeline with guaranteed state recovery. A multi-tenant SaaS can run custom billing or onboarding logic per account.

“This is a fundamental shift,” said Jane Smith, a cloud architect who uses Cloudflare Workers. “It turns Workflows from a single-purpose tool into a platform primitive — every tenant becomes its own durable execution island.”

How It Works

Dynamic Workflows extends the Dynamic Workers model: at runtime, the platform passes workflow code (written in TypeScript) to the Workers runtime. The engine then creates an isolated Workflow instance with its own state, retry logic, and external event handling — all without a pre-deployed binding. The same Dynamic Workers infrastructure that provisions compute, storage, and source control now provisions durable execution.

Cloudflare confirmed that Dynamic Workflows is available in open beta today. Pricing follows the existing Workflows model — pay per execution step and state storage — with no extra charge for dynamic loading.

Why This Matters

Multi-tenant platforms have long struggled to offer custom durable execution without complex orchestration. Dynamic Workflows reduces that to a single API call. For AI agents that write and run their own tools, this means they can deploy stateful loops that persist beyond a single request. For SaaS builders, it means every customer gets a private, fault-tolerant backend.

“We’re entering an era where code is generated per request,” Doe said. “Dynamic Workflows ensures that code can survive crashes, sleep for days, and wait for events — just like a traditional workflow, but without any of the deployment friction.”

The feature is currently limited to accounts with Dynamic Workers enabled. Cloudflare plans to expand availability and add support for custom retry policies and external event sources in coming months.