Cloudflare has introduced Dynamic Workers, a new isolate-based execution environment built to run AI-generated code with significantly lower latency and resource overhead than traditional containers. The move signals a broader industry shift toward lightweight, on-demand runtimes tailored for AI agent workloads.

Unlike containerized solutions that can take seconds to spin up, Dynamic Workers launch in milliseconds and consume only a few megabytes of memory. According to Cloudflare, this makes them roughly 100× faster to start and 10–100× more memory-efficient than typical containers. That efficiency allows enterprises to create a fresh, disposable runtime for every user request, execute a small code snippet, and discard it immediately afterward.

The service builds on Cloudflare’s existing Workers platform, leveraging V8 isolates—lightweight JavaScript sandboxes—to execute code generated at runtime, often by large language models (LLMs). This eliminates the need for full virtual machines or container orchestration, simplifying the infrastructure behind agent-driven applications.

Cloudflare is also promoting its “Code Mode” philosophy alongside Dynamic Workers. Instead of chaining multiple tool calls, developers are encouraged to have models write short, self-contained TypeScript functions that interact directly with defined APIs. The company says this approach can cut both token consumption and response latency.

Enterprise-grade controls are baked into the platform to address security and operational concerns. Features include outbound request interception for credential handling, automated code scanning, and rapid deployment of V8 security patches. Cloudflare notes, however, that isolate-based sandbags offer different security guarantees compared to hardware-backed confidential computing environments.

Dynamic Workers are now in open beta under the Workers paid plan. Standard pricing is set at $0.002 per unique Worker loaded per day, plus regular CPU and invocation fees—though the per-Worker charge is waived for the duration of the beta.

For organizations moving from static toolchains to dynamic, model-driven execution, Dynamic Workers represent a step toward more agile, cost-effective, and scalable AI infrastructure.