Cloudflare AI Sandboxing: 100x Faster Than Containers

March 25, 2026 Rachel Kim – Technology Editor Technology

Cloudflare launched Dynamic Workers in open beta on March 24, 2026, a new system designed to sandbox AI agent code with startup times measured in milliseconds and memory usage significantly lower than traditional container-based solutions.

The new system, detailed in a Cloudflare blog post and reported by VentureBeat, aims to address the performance and cost limitations of using containers – the current standard for sandboxing AI-generated code – as the demand for AI agents scales to potentially millions of users. Containers typically require hundreds of milliseconds to start and consume hundreds of megabytes of memory, often necessitating “warming” them to avoid delays and potentially compromising security through reuse.

Cloudflare asserts Dynamic Workers are approximately 100 times faster to start than containers and between 10 to 100 times more memory efficient. This efficiency is achieved through an “isolate-based” system, allowing multiple instances to run on the same machine, even the same thread, according to VentureBeat’s reporting.

The development builds on Cloudflare’s “Code Mode” initiative, introduced last September, which advocates for agents to generate code that directly calls APIs rather than relying on tool calls. Cloudflare previously demonstrated that converting a Multi-Cloud Platform (MCP) server into a TypeScript API could reduce token usage by 81%. Dynamic Workers are positioned as the secure execution layer to facilitate this approach at scale.

According to Cloudflare documentation, Dynamic Workers are priced at $0.002 per unique Worker loaded per day, in addition to standard CPU and invocation costs. The system allows developers to run code written in any programming language and built for any runtime as part of applications built on Cloudflare Workers.

Cloudflare’s documentation highlights the ability to run resource-intensive applications, applications requiring a full filesystem, and existing containerized applications. The documentation provides example code using JavaScript and a wrangler.jsonc configuration file demonstrating how to define and deploy a container instance.

The launch of Dynamic Workers comes as Cloudflare continues to expand its offerings for agentic AI, with a dedicated platform, Cloudflare Agents, available for building AI agents entirely on the Cloudflare network.