Peji
The "real browser you control" bit is what gets me. Plenty of AI tools will happily pretend to browse for you, but they're usually working off a scraped snapshot or a simulated session. Peji is saying it will actually operate a live browser, which is a whole different level of trust and utility. That's the difference between an assistant that summarizes a page and one that can book the flight, fill the form, and click through the checkout for you.
I'd want to know how much control is real and how much is theater. Does it pause for approval at every step, or does it just go and let you watch? The best version of this is probably somewhere in between, where the AI handles the drudgery but flags anything that costs money or looks risky. The worst version is an agent that barrels ahead and you're left cleaning up the mess.
There's a bigger wave here too. Every AI startup is racing to prove their agent can actually get things done in the messy, broken, real world. Peji's bet is that giving the user a visible hand on the wheel is the trust bridge that gets people to let go. I think that's a smart instinct. The browsers we use are full of logins, cookies, and two-factor prompts, and the companies that crack how to navigate that safely are going to be the ones people actually stick with.