AI Operations
Cloudflare Built a Browser for AI Agents, Not Humans
Kitesurf removes the human-oriented browser baggage and focuses on context, compute, scale, and the new security problems agents bring.
As agents move from answering to acting, the browser becomes operational infrastructure. Teams must optimize token use, compute, context, isolation, and prompt-injection defense while preserving logs and approval gates for consequential web actions.
Cloudflare launched Kitesurf as a cloud-hosted browser for software agents. It can navigate sites, extract HTML, take screenshots, fill forms, and complete browser tasks through Browser Run. Themes, extensions, and pretty tabs are irrelevant to an agent; reliable rendering, context management, token cost, speed, and large-scale execution are the real product requirements.
Cloudflare says Kitesurf uses less CPU and memory than Chromium for common agent tasks and already passes more than 215,000 web-platform tests. It runs on Workers and was built in 12 weeks. Yet a browser that reads arbitrary pages also reads hostile instructions, so prompt injection, credential scope, session isolation, and data leakage become first-class operational risks.
Separate browsing from transactional actions. Use short-lived credentials, allowlisted domains, isolated sessions, download scanning, complete action logs, and human approval before submitting forms, publishing, purchasing, or changing records. Test cost and success by task, not demo. A browser without tabs may look simpler, but once it can click Buy, simplicity has officially left the meeting.