[RFC] Making browser agents reliable

[shadowfax92]

Hey 👋

One-shot browser agents work maybe 30-60% of the time. You ask it to do something, it works once, fails the next time. Not great when you actually need to get stuff done.

We're trying to fix this with Workflows—you chat with the agent to build a step-by-step graph instead of just giving it a goal and hoping.

Say you want to unsubscribe from Gmail marketing emails. Instead of "unsubscribe from all my emails" and crossing your fingers, you end up with a graph: Navigate to Gmail → Click More menu → Manage Subscriptions → Extract subscriptions → Loop through each → Unsubscribe → Confirm if needed.

You can test it, tweak it, run it again later. It follows the same steps every time.

Examples:

• Fill out a form for each row in a spreadsheet
• Monitor prices across a few sites
• Accept LinkedIn requests matching certain criteria
• Move data from one app to another

---

What we want to know:

1. What browser task would you automate if it actually worked?
2. What would make it useful? (Scheduling? Notifications? Something else?)
3. How often do you do this task?

Reply below—trying to figure out what to build next.