Browserbase turns browser agents into a managed web runtime

Browser agents are becoming a managed runtime layer rather than a single Playwright script running on a developer laptop. Browserbase's `llms.txt` describes a Browser Agent Platform that bundles browsers, fetch, search, agent identity, functions, a model gateway, Stagehand, MCP integration, framework integrations, concurrency controls, observability, and session management. Its Agents documentation is more specific: each run follows an observe-reason-act loop, gets a dedicated browser session, can use built-in browser control, web search, fetch, file-system, and shell tools, and exposes asynchronous run states, structured results, run messages, live view, session replay, logs, and the underlying `sessionId`. The identity documentation adds the operational boundary that browser agents need for real sites: contexts for cookies and app data, proxies, Verified sessions, CAPTCHA handling, Cloudflare Web Bot Auth signed agents, credential vault integration, and proof-of-humanity options. For agent builders, the practical signal is to evaluate browser-agent platforms by the whole runtime contract: whether agents can browse, authenticate, observe, replay, constrain navigation, persist useful context, stream run state, and prove legitimate delegated identity without turning every web workflow into brittle selectors and unmanaged credentials.