Cloudflare Unveils Dynamic Workflows: Durable Execution for Multi-Tenant, AI-Driven Platforms
Breaking: Cloudflare Bridges Dynamic and Durable Execution
Cloudflare today announced Dynamic Workflows, a major update to its durable execution engine that finally lets platforms run customer-defined workflows without pre-deploying code. The new feature merges the company's dynamic deployment capabilities — which already power AI-generated code and multi-tenant apps — with its Workflows engine, enabling workflows that are unique per tenant, agent, or request.
“The moment you want to let your customer ship their own workflow, the old model breaks,” said a Cloudflare spokesperson. “Dynamic Workflows removes that bottleneck. Platforms can now hand us a workflow definition at runtime, and we’ll run it durably — with survival of failures, long sleeps, and stateful progress — all within an isolated, sandboxed Worker.”
Background: The Gap Between Durable and Dynamic
Three years ago, Cloudflare launched Workers as a direct-to-developer platform. Since then, the ecosystem has grown to support multi-tenant SaaS, AI agents that write their own code, and CI/CD products with per-repo pipelines. Last month, Cloudflare shipped the Dynamic Workers open beta, which lets platforms hand code to the Workers runtime at runtime, returning an isolated Worker on the same machine in milliseconds. Dynamic Workers solved the compute side. Durable Object Facets extended the idea to storage — per-tenant SQLite databases spun up on demand. Artifacts handled source control with a Git-native, versioned filesystem.
But durable execution remained stuck in the old paradigm. Workflows, Cloudflare's engine for long-running processes, required workflow code to be part of the deployment — a single class bound in wrangler.jsonc. That worked for traditional applications but failed for platforms where every customer, agent, or session has a unique workflow.
What This Means: A New Primitive for Platform Builders
Dynamic Workflows bridges that gap. Platforms can now define workflows dynamically — one per tenant, one per AI agent, one per CI/CD pipeline — without pre-binding classes. The engine runs each instance with full durability: steps survive failures, sleep for hours, wait for external events, and resume exactly where they left off. Workflows V2 already supports up to 50,000 concurrent instances and 300 new instances per second per account.
“This is the missing piece for platform builders,” said the spokesperson. “AI platforms where the AI writes TypeScript for every tenant, CI/CD systems where each repo has its own pipeline, agent SDKs where each agent writes its own durable plan — all of these can now use a single, scalable primitive without custom infrastructure.”
Impact on Multi-Tenant and AI Workloads
The announcement is particularly significant for the growing ecosystem of AI-driven development platforms. Companies building tools where users describe what they want, and AI generates and runs the implementation, can now automatically make those generated processes durable. Similarly, multi-tenant SaaS products can now let each customer’s business logic survive crashes and scaling events.
“Dynamic Workflows completes the dynamic deployment trifecta — compute, storage, and now durable execution,” the spokesperson added. “Platforms get full isolation, per-tenant state, and the guarantees needed for production workloads, all without managing servers.”
Cloudflare expects the feature to be used for onboarding flows, video transcoding pipelines, multi-stage billing, and long-running agent loops. The company plans to roll out Dynamic Workflows to all Workflows users in the coming weeks, building on last month's betas.
Related Articles
- Centralized AI Safety Across Accounts: Amazon Bedrock Guardrails Cross-Account Safeguards Q&A
- Ask the AWS Expert: Key AI and Compute Updates – April 2026
- Navigating Ingress-NGINX Quirks: What to Know Before Migration
- Building AI at Scale: Why Kubernetes Is Your New Foundation for Inference and Production Workloads
- From Digital Clutter to Daily Ritual: Mastering Photo Decluttering with This Day
- How to Safeguard Your SaaS Against Rogue AI Agents: A Comprehensive Data Recovery Guide
- Mastering Amazon S3 Files: Transforming S3 Buckets into High-Performance File Systems
- 10 Ways Runpod Flash Revolutionizes AI Development by Cutting Out Containers