The AI Browsing Agent
The extension's agent is an in-browser assistant that completes web tasks for you — it looks at the page, decides the next step, and performs it, repeating until the task is done.
How it works
Each step, the agent:
- Looks at the current page (the interactive elements — buttons, links, fields).
- Decides the next action using Claude.
- Acts — clicks, types, presses a key, scrolls, waits, or switches tabs.
- Repeats until the goal is met or you stop it.
You see every step in the side panel as it happens.
Writing a good task
The agent does best with specific, grounded instructions:
- Start it on the right page. Open the LinkedIn search (or the page you want it to act on) first, then give the task.
- Be concrete. "On this search, send connection requests to the first 5 results with the note 'Hi {name}, loved your post on X'" beats "network on LinkedIn."
- One outcome at a time. Chain bigger goals into a few smaller runs.
What it can do today
- Navigate, click, type, press keys, scroll, and wait for content
- Work across multiple tabs (in the "wrrk agent" tab group)
- Common LinkedIn / X / Reddit flows (search, connect, message, comment)
- Launch tasks straight from Discover
Current limits
- Logins & CAPTCHAs — it won't solve a CAPTCHA or log into a site for you; sign in first.
- Some custom widgets — content inside certain iframes, canvas UIs, or hover-only menus can be hard to see; nudge it ("open the menu, then look again").
- File uploads & native dialogs — choosing a file or confirming a browser dialog isn't supported yet.
Reliability built in
The agent detects when it's stuck repeating an action, recovers from transient hiccups, caps runaway loops, and snaps back if it wanders off-task. If a step genuinely can't be done, it stops and tells you why rather than spinning.
Stopping & re-running
Hit Stop any time — the run ends and nothing further happens. To retry, restart from the right page with a sharper instruction.