Today I was reading John-Anthony Disotto’s article from TechRadar about Perplexity’s new LLM-enabled Comet browser.
Seems like the guy loved it.
Even with Perpexity’s own PR team warning him that the a-ha moment may “take a week,” John-Anthony says that happened in just days. .
But it didn’t happen immediately.
Actually, early on John-Anthony found his flow disrupted by the browser. He was less productive.
His work slowed. Tasking the agent, playing with it and monitoring it. It didn’t feel natural to just let it go an do its thing…. Book real reservations. Buy real things.
Slowly, John actually did let the AI take over. And that’s where his a-ha started.
I guess, based on that warning they gave, Perplexity’s press people knew there was this learning curve. As he got comfy and really did ‘delegate and forget’, he stopped having to context switch so often and worry about so many things.
Here’s a quick fact for you: Harvard estimates that we burn 5 weeks of work, per year, just trying to reorient our brains after changing focus. That doesn’t count the time actually BEING distracted (or snacking, in my case) either.
I guess Comet (and there are other LLM browsers, including one coming soon from OpenAI) might give us some of that back.
But none of that is the most interesting part of the story.
I think the most interesting tid-bit here is that John-Anthony didn’t actually ever LET the browser fully book or buy for him.
It’s worth noting that Comet is still in beta, so I didn’t feel truly comfortable giving the AI my personal details to complete a booking for a London steakhouse next week, as I had requested.
This is the real revelation, as far as I’m concerned.
This — trust — is something we’ll really have to grapple with in the AI Agent adoption curve, coming soon.
How much do we trust AI?
What recourse is there for AI making a mistake with your wallet?
Can AI accept terms of service?
And maybe the most realistic right now — i.e., something I feel already — do I really trust it really did all the best research?
Did it go as exhaustively on finding the best white t-shirt in the world (an impossible challenge, as far as I’m concerned), as I would?
How will we learn to trust AI? And how will it be held accountable?
