Cloudflare's Agents Week: The Agentic Internet Takes Shape
Cloudflare wrapped up Agents Week with a barrage of launches aimed at making the internet agent-ready: a new runtime, an agent development lifecycle, identity and security controls, and an open protocol for agent-web interaction. Here's what matters.

Cloudflare's Agents Week (August 3–7) delivered a dense stack of announcements aimed at one thesis: the internet is about to be populated by autonomous agents, and the infrastructure needs to be rebuilt accordingly. Over five days, the company rolled out everything from a new runtime designed for agents to a rewrite of MCP, plus a framework for how agents should access resources securely.
The runtime and primitives
Monday set the foundation. The headline was @cloudflare/computer, a runtime that lets agents choose the right execution environment—not just a container, but something more flexible. Workers RPC now spans Python and JavaScript, so mixed-language projects can talk directly. They also added inbound TCP and gRPC support to Workers, which is a big deal for hosting real-time voice agents. And they showed how they're running Kimi and GLM models at scale, focusing on efficiency without sacrificing safety.
The Agent Development Lifecycle
Tuesday introduced the ADLC—the Agent Development Lifecycle—which they position as superseding the traditional SDLC. The centerpiece is Cloudflare Agents, a platform for building and running agents with live tracing, replay, and human-in-the-loop approvals. Local tracing is now available for debugging Workers. They also launched Cloudflare Wallets, a programmable wallet for agents to transact in the emerging agentic economy, and a programmable CI/CD system that can repair failures and stage fixes automatically.
Security and identity for agents
Wednesday extended Zero Trust to agents. The Agent Access Model is a framework for how agents can securely access resources on behalf of users. They open-sourced Cloudflare OS, the platform they use internally to build apps and automate work. Identity-aware AI Gateway analytics can attribute AI activity to real users, and WriteGuard gives fine-grained control over MCP server tool calls to prevent rogue actions.
The Agentic Internet
Thursday was about making the web agent-ready. They proposed a model for an open Agentic Internet that is readable, discoverable, callable, and payable. WebMCP is a preview of a simple way to give any website an interface for agents. They're pushing AEO (Answer Engine Optimization) as the evolution of SEO. Kitesurf is an agent-first browser that runs in V8 isolates on Workers, trading pixel-perfect rendering for lower resource usage. MCPv2 is a rewrite of MCP aimed at simplifying deployment and scalability. And AI Search turns your files or website into an agent-ready search engine with one command.
Observability and community
Friday wrapped up with insights into real agent behavior, a unified AI control plane combining Workers AI and AI Gateway, and community programs including $1M more in open-source funding. Radar Researcher lets you query internet data in plain language.
Cloudflare is betting big on agents being the next major computing paradigm, and they're building the infrastructure to support it. Whether the ADLC catches on or MCPv2 becomes the standard, the direction is clear: the internet is getting agent-native.
Your agent needs a computer, not a container.
Aug 3
Runtime & infrastructure
@cloudflare/computer, Workers RPC for Python/JS, inbound TCP/gRPC, Kimi/GLM at scale, Billable Usage API
Aug 4
Agent Development Lifecycle
Cloudflare Agents, local tracing, Wallets, programmable CI/CD, engineering standards enforcement
Aug 5
Zero Trust for agents
Agent Access Model, Cloudflare OS open-sourced, identity-aware AI Gateway, WriteGuard for MCP
Aug 6
Agentic Internet
WebMCP, AEO, Kitesurf browser, MCPv2, AI Search
Aug 7
Observability & community
Agent behavior insights, Workers AI + AI Gateway unification, Radar Researcher, $1M open-source funding
Discussion
0 Comments
Be the first to start the discussion.