Cloudflare used Agents Week from Monday, August 3, through Friday, August 7, to announce tools, products, and ideas for developers and customers building AI-native apps and agent infrastructure. The company framed agents as both a new application of AI and a new class of software, with technical needs beyond models: identity, communication, orchestration, memory, observability, and security. Cloudflare said the launches were aimed at an “Agentic Internet” where humans and agents cooperate instead of collide.

Monday, August 3, centered on runtime and infrastructure for intelligent, autonomous apps. Cloudflare introduced @cloudflare/computer, a runtime designed for agents that can choose the right environment for a task. It also said Workers RPC now works across Python and JavaScript, enabling direct communication between Python and JavaScript Workers, and described work on serving Kimi and GLM models more efficiently without sacrificing quality, reliability, or safety. Other Monday updates included the Billable Usage API for programmatic cost visibility, plus inbound TCP connections and gRPC support for Cloudflare Workers and Containers, with voice AI backends and real-time voice agents named as use cases.

Tuesday, August 4, introduced the Agent Development Lifecycle, or ADLC, which Cloudflare described as superseding the SDLC (Software Development Lifecycle) for moving agents from prototype to production. The company launched Cloudflare Agents with live run watching, tracing, replay, and human-in-the-loop approvals for production actions. Tuesday also added local tracing for debugging Workers, Cloudflare Wallets for agent transactions, programmable CI/CD for millions of repos, AI-powered enforcement of Cloudflare engineering standards, and an Astro workflow that automated issue analysis, categorization, and routing to drive the project’s GitHub issue count to zero.

Wednesday, August 5, extended Zero Trust from users and devices to agents. Cloudflare described the Agent Access Model as a framework for agents to securely access resources and services on behalf of users. The company also described Cloudflare OS, its internal AI-enabled operating model, and open-sourced Cloudflare OS as a platform for agents, apps, and work. Security and oversight updates included identity-aware analytics for attributing AI activity to real users and systems, catching anomalies and spend spikes, and WriteGuard, a private beta for fine-grained controls over MCP Servers and risky tool calls.

Thursday, August 6, defined Cloudflare’s model for the Agentic Internet around websites that are readable, discoverable, callable, and payable by agents while publishers keep control. Cloudflare previewed WebMCP for giving websites and web apps an interface usable by agents, described AEO as an adaptation of SEO for agent discovery, introduced Kitesurf as an agent-first browser running in V8 isolates on Cloudflare Workers, announced MCPv2 as a rewrite of MCP support for deployment and scalability, and launched Cloudflare AI Search to turn files or websites into an agent-ready search engine with one command.

Friday, August 7, focused on agent behavior, AI control, community programs, and Internet data analysis. Cloudflare described trust-based bot mitigation for “good and bad” agentic behavior, combined Workers AI and AI Gateway into a single AI control plane with one binding, one wallet, and one dashboard, and said model-first routing is next. The company also announced Cloudflare Ambassadors, Community Engineers, and another $1M in open-source funding over the next two years, plus Radar Researcher, an AI tool that turns plain-language Internet-data questions into interactive charts.

Cloudflare used Agents Week from Monday, August 3, through Friday, August 7, to announce tools, products, and ideas for developers and customers building AI-native apps and agent infrastructure. The company framed agents as both a new application of AI and a new class of software, with technical needs beyond models: identity, communication, orchestration, memory, observability, and security. Cloudflare said the launches were aimed at an “Agentic Internet” where humans and agents cooperate instead of collide.

Monday, August 3, centered on runtime and infrastructure for intelligent, autonomous apps. Cloudflare introduced @cloudflare/computer, a runtime designed for agents that can choose the right environment for a task. It also said Workers RPC now works across Python and JavaScript, enabling direct communication between Python and JavaScript Workers, and described work on serving Kimi and GLM models more efficiently without sacrificing quality, reliability, or safety. Other Monday updates included the Billable Usage API for programmatic cost visibility, plus inbound TCP connections and gRPC support for Cloudflare Workers and Containers, with voice AI backends and real-time voice agents named as use cases.

Tuesday, August 4, introduced the Agent Development Lifecycle, or ADLC, which Cloudflare described as superseding the SDLC (Software Development Lifecycle) for moving agents from prototype to production. The company launched Cloudflare Agents with live run watching, tracing, replay, and human-in-the-loop approvals for production actions. Tuesday also added local tracing for debugging Workers, Cloudflare Wallets for agent transactions, programmable CI/CD for millions of repos, AI-powered enforcement of Cloudflare engineering standards, and an Astro workflow that automated issue analysis, categorization, and routing to drive the project’s GitHub issue count to zero.

Wednesday, August 5, extended Zero Trust from users and devices to agents. Cloudflare described the Agent Access Model as a framework for agents to securely access resources and services on behalf of users. The company also described Cloudflare OS, its internal AI-enabled operating model, and open-sourced Cloudflare OS as a platform for agents, apps, and work. Security and oversight updates included identity-aware analytics for attributing AI activity to real users and systems, catching anomalies and spend spikes, and WriteGuard, a private beta for fine-grained controls over MCP Servers and risky tool calls.

Thursday, August 6, defined Cloudflare’s model for the Agentic Internet around websites that are readable, discoverable, callable, and payable by agents while publishers keep control. Cloudflare previewed WebMCP for giving websites and web apps an interface usable by agents, described AEO as an adaptation of SEO for agent discovery, introduced Kitesurf as an agent-first browser running in V8 isolates on Cloudflare Workers, announced MCPv2 as a rewrite of MCP support for deployment and scalability, and launched Cloudflare AI Search to turn files or websites into an agent-ready search engine with one command.

Friday, August 7, focused on agent behavior, AI control, community programs, and Internet data analysis. Cloudflare described trust-based bot mitigation for “good and bad” agentic behavior, combined Workers AI and AI Gateway into a single AI control plane with one binding, one wallet, and one dashboard, and said model-first routing is next. The company also announced Cloudflare Ambassadors, Community Engineers, and another $1M in open-source funding over the next two years, plus Radar Researcher, an AI tool that turns plain-language Internet-data questions into interactive charts.