Mark Zuckerberg Might Have Finally Made Something That’s Actually Useful

The people have spoken, and they’re curious about A.I. agents. Meta’s new app, Muse, is No. 1 on the App Store. There’s a shift at play. Most of the people using ChatGPT are asking it to do things that remain right in front of them. Answer this question, draft this text, or do something else right here in the app. Muse is meant as something different: a robot that performs tasks for you across the internet. Book a haircut, order DoorDash, get a bunch of supplies for my party.

Muse does it on your behalf. It’s a mass-marketed, mobile-friendly version of the machine-as-person work that power users are already doing with Claude and ChatGPT on computers .

Muse is not the first product to try to go out and just do stuff for you. People have been using text messages and other mobile interfaces to have A.I. models run internet errands for them for a while. But those, again, have been for power users. Muse is hooked into your Instagram, giving Meta the same launch platform it used for Threads three years ago. This is the first mega-launch of a big tech company’s A.I. tool whose value proposition is that it will do your online shopping, your reservation booking, your calendar updating, and the other digital annoyances that you might put yourself through dozens of times a day.

It’s a fun concept—the machine taking on pesky tasks that offer no redeeming mental value—but there’s an important group standing in Big Tech’s way: Big Tech.

The simplest everyday value proposition for an app like this is online shopping. Your kid is going off to college next week, and she needs books. You tell Muse to check the syllabi for her classes and order all the books to her dorm room. This seems useful: the computer taking on work that plodding through yourself takes your time but brings no emotional payoff or personal growth. Some online shopping is fun, and some isn’t.

When I die, I will not feel like the time spent perusing Amazon search results was important to my story. It now doesn’t have to be, as long as the app works properly. You can enter various login credentials in a vault in the app, and you can pay for things via Link, an A.I.-forward payment platform. I have been playing around with Muse over the past few days, and while I’ve encountered problems (more on those shortly), it would be foolish to pretend the app can’t work quite well .

Amazon does not like this setup, because the company doesn’t just want you to buy stuff. Amazon wants you to see ads, to become aware of the virtues of Prime membership, and to know that there’s a Thursday Night Football game that you won’t want to miss. But your robot shopper doesn’t watch TV, and if you’re outsourcing the shopping to it, you are not giving Amazon the same chance to harvest dollars from you as you walk through the digital store.

So, Amazon is blocking Muse from shopping on its platform , similar to how it was already fighting another A.I. business, Perplexity, over its shopping tool. One of the biggest companies in the world has spent hundreds of billions of dollars trying to own a corner of A.I., and now that it has released the fruit of that labor, one of the other biggest companies in the world says it’s a no go. (Meta has had more luck with other e-commerce platforms, like Expedia, which it announced a deal with on Tuesday .)

The stakes are enormous for Meta, which has not made a product that normal people care about since Facebook and has not acquired any since WhatsApp in 2014. There is no more obvious implementation of a tool like Muse than online shopping, and the king of American e-commerce is just saying no . The stakes are less big but still huge for Amazon, which reported nearly $70 billion in ad revenue last year.

This might feel like a strictly ideological battle with two tech giants’ business incentives mapped onto it. If you believe in A.I. agents as some liberating force giving people their time back, then Meta has a financial incentive to pretend to be your champion. If you want to preserve the homely, mom-and-pop feeling of picking out your own merchandise and contemplating the value of a dollar before you enter your credit card info, Amazon is there for you.

But this squabble between Goliaths undersells that ordinary online business and regular customers are going to find themselves at odds with each other plenty as more people start deploying these little robot gophers. Customers will fight each other, too.

Restaurant reservations are an illustrative battleground. In theory, they are a perfect use for this tool. You know where you want to eat, whether it’s one restaurant or one of a few. You don’t enjoy clicking around a bunch looking for a table, so you instruct an agent to do it for you. Cool, but the problem is obvious. If agents are always prowling for reservations, human spontaneity will lose, and fewer people who have the good fortune of checking for a competitive seating at the right moment will get it.

In a mass-adoption scenario, will people just send out their Muse agents like Pokémon to fight over scarce tables at every popular restaurant? Resy recently suspended a user’s account when it sniffed agent activity , but it’s not clear if this behavior is officially allowed or not. A friend of mine uses Instinct, the A.I. tool at issue here, for reservations multiple times per week. He actually uses his reservations, but you can imagine people using agents to hoard tables, or the human user not realizing the agent booked a table and the restaurant being stuck with a no-show.

These restrictions have a direct bearing on the usefulness of the tools. My first experience with Muse was on Monday. I needed to book a flight for myself and my wife. I found it myself on Google Flights, then, for journalistic reasons, asked Muse to take a crack at the booking. I put my airline login credentials in Muse’s vault, but Muse met a terrifying obstacle, a captcha page, and asked me if I should take over or if it should attempt to pass the test asking if it was a human.

Authorized to try, Muse failed; I took over and logged in, then passed the sticks back to the agent. It got kicked out of the browser a few minutes later. The process took 30 minutes and achieved nothing. I then booked the flights myself at a cost of five minutes and just over $500. The corporate fight won’t just be tech versus tech, but tech versus every major corporation that sells to consumers.

My second experience with Muse was at 12:40 p.m. Eastern on the day I drafted this story. I asked it to add an event to my calendar at 10 a.m. tomorrow, and it added the event two hours and 40 minutes in the past. So, at least in my experience, Muse is not yet reliable. On top of that, your level of trust in anything created by Mark Zuckerberg should constantly hover in the 0-to-4 range out of 10.

I wouldn’t say this stuff won’t work soon, though. A.I.s are getting good at clicking around a computer. I recently watched with smug satisfaction as OpenAI’s GPT-6 Astra model hunted through spreadsheets to reconcile an uncooperative credit card statement and then navigated QuickBooks to make sure it was reflected in the right places. That is not a mass-marketable use case, and it doesn’t fit into a little Meta phone app.

But if Muse doesn’t become reliable at doing computer errands, one of its competitors probably will. Not all businesses that rely on actual people moving through their interfaces will stand united against agents, either, especially when they have their own web of entanglements with the biggest A.I. companies. Microsoft, for example, wants to be easy with this kind of tech .

That doesn’t mean we’re all about to be shopping through Muse or any of its competitors. One can imagine a slow drip of consumer horror stories having a chilling effect on people’s interest in Muse-like apps. The fraud possibilities around A.I. agents seem legitimately boundless , and so do the ways it could go wrong for people to entrust these apps with sensitive information. But the simplest thing that would chill adoption of agent apps would be if they didn’t gain the ability to reliably do people’s internet chores.

The machine can only do those chores if big companies let it into their houses. They still have reasons not to, and some of those reasons are even good ones.