Cloudflare released Kitesurf, a headless web browser built from scratch to run inside V8 isolates on Cloudflare Workers rather than wrapping Chromium. That architectural choice is the whole point: Chromium-based automation tools like Puppeteer or Playwright give an agent a full desktop browser — tabs, a compositor, a GPU rendering pipeline, extension support — none of which an AI agent actually needs when its job is to read a page's content and click through a workflow. Kitesurf strips that down to what an agent does need: fast page loads, machine-readable DOM and text output, low memory per instance, and isolation between sessions so that content on one page can't leak into or manipulate another agent's session (a real concern given how often prompt-injection attacks hide instructions in scraped web content). The efficiency gains are the headline number: benchmarking Cloudflare published alongside the release shows 3 to 7 times lower CPU and memory usage than a Chromium-based browser doing the same navigation and extraction tasks, at the cost of roughly 1.7 times slower wall-clock time per page. For a team running one browser session, that trade looks bad. For a team trying to run thousands or millions of concurrent agent sessions — the actual use case Cloudflare is targeting — the memory reduction is what makes horizontal scaling on commodity infrastructure economically viable at all, since memory per isolate, not raw speed, is usually the binding constraint. Kitesurf already passes over 215,000 Web Platform Tests and speaks the Chrome DevTools Protocol, so existing Puppeteer, Playwright, and MCP-based browser-automation code can point at it with a single connection parameter rather than being rewritten. It's free in beta through Cloudflare's Browser Rendering product now, with open-sourcing planned. For anyone building web-browsing agents at scale, it's a direct alternative to running a fleet of headless Chrome instances.