Agents Week: Building the Infrastructure for an Agentic Internet
Cloudflare's Agents Week introduced the foundational tools and concepts for a new class of software: autonomous agents. The week-long series explored what developers need to build AI-native applications that operate on an Agentic Internet—an Internet designed for both humans and the software acting on their behalf.
The key message: agents are not just another AI application, but a new category of software that fundamentally changes how people interact with technology and how software interacts with the Internet. The challenges go beyond the models to identity, communication, orchestration, memory, observability, and security.
The Execution Layer and Development Pipeline
The opening day focused on the runtime and infrastructure agents rely on—the execution layer that forms the foundation for building and running intelligent, autonomous applications.
Day two introduced the Agent Development Lifecycle (ADLC), a framework defining the primitives necessary to move agentic software from prototype to production. This lifecycle represents a shift in how agent applications are systematically developed, tested, and deployed.
Security from Users to Agents
Midweek, the focus shifted to security. Cloudflare extended its Zero Trust framework from users and devices to the agents themselves. The company showed how it applies this internally, treating agents as first-class entities within its security architecture rather than as extensions of their users.
The Protocol for a Cooperative Internet
The Agentic Internet was the subject of a day dedicated to defining a cooperative web. Cloudflare outlined how website owners, publishers, and agents can all participate in an Internet that serves both people and agents—a key requirement as agent traffic grows and existing protocols may need adaptation.
Observing Real-World Agent Traffic
The final day focused on data: observing what agents are actually doing on the web, where AI is running in applications, who is contributing to the ecosystem, and what new analytics tools are available for understanding these patterns.
What an Agent Cloud Must Provide
The week's conclusion points to a clear set of requirements for an Agent Cloud: an execution layer and primitives to run on, a development lifecycle that increasingly automates itself, secure access controls for both people and agents performing work, an Internet protocol that accommodates both, and the human communities keeping everything grounded.
The work continues—Cloudflare will keep iterating on these foundations, with updates available in its changelog.
