Cloudflare Targets Agent Runtime

- Cloudflare expanded its Agent Cloud with infrastructure, security, and developer tools for autonomous AI agents. - The update adds runtime, policy enforcement, and edge-scaling capabilities aimed at production agent deployments. - Cloudflare is positioning to own the execution layer for agents, shifting infrastructure demand toward orchestration and latency control (pulse2.com).

Cloudflare said on April 13 it is expanding Agent Cloud to run long-lived artificial intelligence agents on its network, not just chatbots in demos. (cloudflare.com) The company said the package adds infrastructure, security, and developer tools aimed at moving agents from local laptops to “production-grade workloads” across Cloudflare’s global network. Cloudflare tied the push to coding agents and other tools that take multi-step actions over time. (cloudflare.com) An agent is software that does not just answer one prompt and stop. Cloudflare’s developer docs say its Agents software development kit gives each agent memory, scheduling, tool use, real-time connections, and its own Durable Object, which is Cloudflare’s stateful micro-server with a built-in SQL database. (developers.cloudflare.com) Cloudflare is pitching that setup as cheaper than keeping one virtual machine or container running for every user. In its launch materials, the company said the current model of “always-on virtual servers or isolated sandboxes” will not scale if each worker or customer ends up with dozens of personal agents. (cloudflare.com; blog.cloudflare.com) The new stack reaches beyond one runtime. During its April 2026 “Agents Week,” Cloudflare also rolled out Project Think for long-running agents, Browser Run for browser control, Cloudflare Mesh for private network access, and higher-concurrency Workflows for multi-step jobs. (blog.cloudflare.com; blog.cloudflare.com; blog.cloudflare.com) Cloudflare is also trying to sit in the middle of model access, not just execution. On April 16, it said developers can call 70-plus models from 12-plus providers through one application programming interface, with AI Gateway handling routing, retries, logging, and provider switching. (blog.cloudflare.com) That matters because agents tend to make many model calls for one task. Cloudflare said a chatbot may make one inference call per prompt, while an agent may chain about 10 calls together, making latency, outages, and cost controls more important than in a simple chat app. (blog.cloudflare.com) The company has been building toward this for more than a year. In 2025, Cloudflare launched an Agents software development kit on top of Workers and Durable Objects; in April 2026, it added sandboxes, persistent sessions, sub-agents, and durable execution features under Project Think. (blog.cloudflare.com; blog.cloudflare.com; blog.cloudflare.com) Cloudflare’s bet is that the next fight in artificial intelligence infrastructure is not only who trains the best model, but who hosts the agent after the model responds. Its April rollout puts the company squarely in the business of running, securing, and coordinating those agents at internet scale. (cloudflare.com; blog.cloudflare.com)

Get your own daily briefing

Scout delivers personalized news, insights, and conversations tailored to your role and industry.

Download on the App Store

Shared from Scout - Be the smartest in the room.