为工作申请、登录和数据抓取运行浏览器代理
Run browser agents for job applications, logins, and data fetching
已完成Manus。一个海报测试了六个浏览器代理在日常工作任务如工作申请、登录、数据抓取和发布；Manus功能多样但触发了机器人检测且成本过高。
做了什么
The poster compared six browser agents on everyday multi-step tasks including job applications, logging into sites to fetch data, posting on their behalf, solving assignments, and API/troubleshooting discovery. They tested ChatGPT agent, Manus, Perplexity Computer, Perplexity Comet, and two local Playwright-based setups; the key difference they called out was Comet running inside their own browser, while the local stacks failed even on simple tasks.
效果如何
Comet was their best balance, while Manus was capable but too expensive for daily use and still hit bot detection; ChatGPT agent was the worst performer, and the two local model setups were not production-ready. In comments, they said most of their cases were multi-step tasks inside authenticated sessions.
值得注意
Their main catch was economics and auth friction: tools that start fresh browser sessions kept running into logins and bot detection, and the strongest cloud options were too costly for daily use.
试试这个 Prompt
Please do this routine web task for me: [e.g. find new [job title] openings on [job site], fill out the application using my details in [file or profile], and log results]. Log in using the session I have open and tell me if a site blocks you with a bot check instead of retrying repeatedly. Stop after [number] items and give me a summary table of what you did. Pause and show me each application or post before you submit it.