Cloudflare has released a "code mode" platform using Dynamic Workers to run AI-agent code more securely and quickly for enterprises [1].
This development represents a shift in how companies deploy AI agents by attempting to remove the overhead of traditional containers. If successful, it could lower the barrier for businesses to integrate autonomous agents into their live web infrastructure.
According to VentureBeat, Dynamic Workers start in milliseconds and use only a few megabytes of memory [3]. This architecture allows AI-agent code to run up to 100 times faster than traditional containers [3].
However, the launch has met with technical scrutiny. YouTube creator Yannic Kilcher released a video response stating that the platform overlooks critical aspects of Multi-Component Programming (MCP) and tool-calling capabilities [2]. Kilcher said the offering fails to address these specific requirements necessary for complex AI agent interactions.
Cloudflare designed the platform to let enterprises deploy agents faster, but the gap in tool-calling support may limit the utility of the agents for advanced developers [1]. The tension highlights a broader struggle in the industry to balance raw execution speed with the flexible standards required for AI agents to interact with external data and tools [2].
Dynamic Workers are intended to provide a lightweight environment that avoids the latency typically associated with booting up full virtual machines or containers [3]. While the speed gains are significant, the utility of an AI agent depends heavily on its ability to execute specific functions, a process known as tool calling, which Kilcher said is missing from the current implementation [2].
“Dynamic Workers start in milliseconds and use only a few megabytes of memory”
The conflict between Cloudflare's focus on infrastructure speed and Kilcher's focus on protocol standards illustrates the current divide in AI development. While high-speed execution reduces latency, the lack of robust tool-calling and MCP support could render the platform insufficient for agents that need to perform complex, multi-step tasks across different software environments.
