When Browser Automation Will Work
Many users want to use OpenClaw for tasks that require a web browser: sending connections on LinkedIn, buying stuff on Amazon, filling out forms and tax paperwork.
Bailey and I spent the first month of YC working on browser agents, so I’m excited to write down once and for all what works (in March 2026).
Sites that don’t want you there
Avoid websites that could easily make an API for you to use their service, and purposely do not do this. LinkedIn is the classic example, but most social media applies. They don’t want you spamming connection requests and farming likes. There are countless services that help you automate LinkedIn, and they work, but you risk getting your account banned. I think Amazon falls in this bucket too, but I don’t fully get it — if I were them I’d just make an API and target ads at agents just like I target ads at humans.
If you do really want to use a site like this, you’ll have very little luck doing so from a remote machine. Most bot detection works by analyzing your browser fingerprint and IP address, not through CAPTCHAs. You should use Klaus’s browser extension to allow remote control of your Chrome tabs.
Long forms with reactive UIs
Be cautious trying to fill out long forms with reactive UIs. AI is still surprisingly bad at this. I spent too many hours of my life getting Gemini to fill out expense reports in Navan. Angular (a popular frontend development framework) was doing some very fancy validation on the exact keystrokes used to enter dollar amounts. The agents also struggle with those fields that give you suggestions as you type, e.g. suggesting the rest of your address.
Where browser agents shine
Go nuts with letting browser agents navigate sprawling UIs. Bailey and I used to love letting Bittie (our browser agent) change settings in Google Cloud Console, or update company info in Rippling.
Most of the time you don’t need a browser
Most of the time you don’t need to use a browser agent. If you just need to read what’s on a site, your agent should use its web fetch tool. If you’re doing research, clicking through links is OK, but you’ll get better results faster and more cheaply from using a service like Perplexity or Exa. If you’re taking actions and an API can do the job, always prefer it.
Want to try Klaus? Sign up at klausai.com and get your own AI agent running in under a minute.