Machine-generated traffic on the Fastly network hit a critical inflection point in July and August 2026, crossing the 50 percent threshold. This shift marks a transition from human-centric web architecture to an agent-dominated landscape, underscored by AI traffic growth that outpaced human interaction by 6.5 times between January and May 2026. As autonomous model calls become the primary driver of network load, the infrastructure layer is being forced to adapt. Fastly is responding to this reality by integrating three agent-specific capabilities directly into its global edge cloud, which maintains 622 Tbps of capacity and processes over 5 trillion requests daily.
The Edge Governance Suite
Fastly’s new capabilities aim to provide the visibility and control that enterprises currently lack, a gap highlighted by McKinsey data showing that 93 percent of organizations are exceeding their AI budgets. The suite includes an AI Firewall, which performs prompt injection mitigation natively within Fastly Compute. By running an ML classifier as an inline firewall, the system avoids the latency and cost penalties associated with GPU-dependent security checks. Alongside this, AI Runtime Control acts as a centralized routing mechanism for model calls. It utilizes virtual keys to protect provider credentials while offering real-time visibility into token spend, rate limiting, budget controls, and automated failover. Finally, the platform introduces expanded API Security, which enforces API contracts against agentic traffic, allowing organizations to observe or block non-conforming requests at the service level.
As Kelly Shortridge, Chief Product Officer at Fastly, noted, enterprises require control in production and at runtime, without introducing delay, friction, or disruption. These tools are designed to provide that oversight at the edge, intercepting traffic before it reaches the model provider.
The Architectural Tug-of-War
The industry is currently fracturing into three distinct approaches for agent governance: edge-based control, cloud-native runtimes, and protocol-level enforcement. Fastly is betting on the edge, but it faces competition from other infrastructure providers. AWS has moved into this space with AgentCore Runtime V2, a managed cloud runtime that utilizes snapshot-based microVM execution environments to secure agent operations. Meanwhile, the CNCF is evaluating the Model Context Protocol (MCP) as a cloud-native wire specification under Initiative #1746. Cloudflare has also entered the fray by shipping MCP Detection, which provides protocol-level visibility by identifying signals within TLS-inspected HTTPS requests. The fact that three major enterprise vendors have shipped policy enforcement through the MCP governance surface within a single week underscores how quickly the infrastructure layer is standardizing around this interface.
Enterprise Implications
The core challenge for enterprises is determining where governance should reside. Each approach-edge, cloud runtime, or protocol-carries different trade-offs regarding latency, visibility, and vendor lock-in. Fastly’s strategy assumes that the edge is the most efficient place to enforce policy, as it allows for unified control across disparate model providers and internal services. However, the choice of where to anchor governance will likely dictate an organization’s ability to scale AI operations without sacrificing security or budget predictability.
Market signals suggest investors are closely watching these developments. Fastly’s stock performance, which rose approximately 32 percent from $20.77 on September 1 to $27.50 on September 21, reflects the high stakes of this infrastructure shift. As machine traffic continues to outpace human interaction, the question of whether edge providers become the default governors for enterprise agents remains the central tension in the infrastructure stack.