Engineering Blog
8 posts under this tag.
As of mid-2026, AI browser agents have diverged into four distinct architectural paradigms. This article delivers a comprehensive comparison across six dimensions, with a decision tree for choosing the right tool.
agent-browser replaces Node.js cold starts with a native Rust daemon, replaces brittle CSS selectors with A11y snapshot + stable element ref IDs, compressing DOM from 5000 tokens to 200-400.
Session isolation, security policies, proxy selection, and production deployment checklist for agent-browser.
network route to intercept requests, network har to record performance, react tree for component inspection, vitals for metrics.
batch for multi-step execution in one call, chat for natural language browser control, concurrent sessions.
--profile to reuse Chrome login state, --session-name for auto persistence, state save/load for explicit management.
snapshot -i for interactive elements only, screenshot --annotate with numbered labels, diff for change detection.
npm install -g agent-browser, one command to open a browser, snapshot to understand page structure, click/fill to interact.