Cloudflare used the start of its Agents Week to describe a broader push toward infrastructure designed for AI agents: software that can plan tasks, call tools, and act across online services with less direct human control.
The company says its original framing was too human-centered. Rather than asking what developers want from an “Agent Cloud,” Cloudflare argues the harder question is what agents need from the network, security, storage, and compute systems beneath them.
That distinction matters because agents can generate many more requests, tool calls, and identity checks than a typical user session. They also need guardrails so they can act without leaking data, abusing credentials, or getting blocked by services that cannot tell good automated activity from bad traffic.
The post is an agenda-setter rather than a single product launch, so the practical details will depend on what Cloudflare announces during the week. Still, it signals that major infrastructure providers are preparing for agents as a new class of internet client, not just another feature inside chatbots.