Cloudflare Dynamic Workers Run AI Agent Code 100x Faster Than Containers

Quick Facts

• Dynamic Workers achieve 100x faster startup and 10x-100x more memory efficiency than traditional containers
• Zite processes millions of daily execution requests using the platform for LLM-generated apps
• Pricing at $0.002 per unique Worker loaded per day, with beta fees currently waived

Cloudflare launched Dynamic Workers in open beta, ditching containers for lightweight isolates that execute AI-generated code in milliseconds instead of hundreds of milliseconds. The platform uses V8 JavaScript execution engine instances to deliver what Cloudflare claims is 100 times faster startup performance.

The technology targets AI agents that generate code on-demand. Traditional container-based systems require hundreds of milliseconds and hundreds of megabytes of memory to start. Dynamic Workers need just a few milliseconds and megabytes.

Zite, an app platform where users interact through chat while LLMs write TypeScript for CRUD applications, serves as the primary customer success story. The company now handles millions of execution requests daily using Dynamic Workers to connect with services like Stripe, Airtable, and Google Calendar.

“To enable server-side code for Zite’s LLM-generated apps, we needed an execution layer that was instant, isolated, and secure,” said Antony Toron, CTO and co-founder of Zite. “Cloudflare’s Dynamic Workers hit the mark on all three, and out-performed all of the other platforms we benchmarked for speed and library support.”

Dynamic Workers can scale to millions of requests per second with no limits on concurrent sandboxes. One-off Dynamic Workers typically run on the same machine and thread as the originating Worker, eliminating cross-region communication delays.

The platform charges $0.002 per unique Worker loaded per day, plus standard CPU and invocation charges. Cloudflare positions this cost as negligible compared to inference expenses, arguing execution should be a routine part of the agent loop rather than a major expense.

Cloudflare promotes “Code Mode” – having LLMs write code against APIs instead of using sequential tool calls. Converting an MCP server to TypeScript API can reduce token usage by 81% and save up to 80% in inference costs.

The approach contrasts with competitors using Docker Sandboxes and lightweight microVMs for persistent environments. Cloudflare optimizes for short-lived, high-volume execution rather than deeper, persistent environments.

Security remains paramount despite the speed gains. Cloudflare acknowledges that hardening isolate-based sandboxes presents more challenges than hardware VMs, since V8 security bugs occur more frequently than hypervisor bugs.

The company leverages nearly a decade of experience with automatic V8 security patch deployment, custom second-layer sandboxes, dynamic tenant cordoning, and hardware features like MPK for protection.

Read more: Cloudflare’s new Dynamic Workers ditch containers to run AI agent code 100x faster