145 / 168 · 01 Agent Skills for Browser Automation · Tool Comparison Matrix: Browser Automation for AI Agents (July 2026)← prev⊞ allnext →☰ Read as one page
20.1The Contenders
| Tool | Creator | Approach | Protocol | Language |
|---|---|---|---|---|
| Playwright CLI + Test Agents | Microsoft | CLI skill + disk-first snapshots; planner/generator/healer agents | CDP (Chromium), custom (FF/WebKit) | Any (shell) / TypeScript |
| Playwright MCP | Microsoft | MCP server + accessibility trees | CDP (Chromium), custom (FF/WebKit) | TypeScript |
| Stagehand | Browserbase | TypeScript SDK, intent API: act("click the submit button") |
CDP | TypeScript |
| browser-use | Community | Python library + vision/LLM reasoning | CDP | Python |
| Vibium | Jason Huggins (creator of Selenium, 2004) | CLI + BiDi proxy, skills-first | WebDriver BiDi | Go binary (JS/Python clients) |
| Selenium 4.x | Community + browser vendors | WebDriver + low-level BiDi | WebDriver + BiDi | Java/Python/JS/C#/Ruby |
| Commercial platforms (testRigor, mabl, KaneAI, Applitools, Testim, Functionize) | Various vendors | "Agentic execution" SaaS | Proprietary | SaaS |
Two notable changes from the January 2026 landscape: Octomind is winding down (farewell letter published — remove it from any evaluation shortlist), and OpenAI's Operator no longer exists as a product (absorbed into ChatGPT Agent / the Agents SDK; the Atlas browser is also being shut down as of August 2026).