I totally agree that the optionality of AI use matters. Agentic browsers take away too much control and visibility. They automate away necessary friction in some tasks, making our online activity less transparent and understandable. With a normal browser, you can trace your actions and usually figure out where something went wrong. But an agentic browser could be looking up dozens of pages to complete a task (like writing an email), making it hard to know what it’s doing and how it’s using information.
Exactly. Also, I think the opacity is a big deal. None of these browsers is open source -- so not only do users have no idea what the agent is doing, there's also no way to identify how the agent or the browser provider is collecting or processing data.
It really is just pinky swears and promises, with no means of sense-checking anything.
Great overview of the many agentic browser risks. I like the point of trust. We have built an internet that is decentralised but also not inherently trustworthy and we are now trying to run agents that don't distinguish between trusted and untrusted input on top of this architecture. What could possibly go wrong? I wonder how long it will take for credit card companies to limit or prohibit use of their cards in this context.
I don't know how a card company could though. It's the same trap as the website owners are in.
In the end, I don't think the card providers will care if a bot makes a purchase --the human is the one on the hook. And there are many monied interests who are keen to make sure it stays that way.
I totally agree that the optionality of AI use matters. Agentic browsers take away too much control and visibility. They automate away necessary friction in some tasks, making our online activity less transparent and understandable. With a normal browser, you can trace your actions and usually figure out where something went wrong. But an agentic browser could be looking up dozens of pages to complete a task (like writing an email), making it hard to know what it’s doing and how it’s using information.
Exactly. Also, I think the opacity is a big deal. None of these browsers is open source -- so not only do users have no idea what the agent is doing, there's also no way to identify how the agent or the browser provider is collecting or processing data.
It really is just pinky swears and promises, with no means of sense-checking anything.
Great overview of the many agentic browser risks. I like the point of trust. We have built an internet that is decentralised but also not inherently trustworthy and we are now trying to run agents that don't distinguish between trusted and untrusted input on top of this architecture. What could possibly go wrong? I wonder how long it will take for credit card companies to limit or prohibit use of their cards in this context.
I don't know how a card company could though. It's the same trap as the website owners are in.
In the end, I don't think the card providers will care if a bot makes a purchase --the human is the one on the hook. And there are many monied interests who are keen to make sure it stays that way.