The Feed Has Left the Screen

We used to eat from the earth. Then we ate from the feed. What happens when the feed no longer needs a screen? At Meta Connect this week, Mark Zuckerberg unveiled the Muse Charm, a small AI device designed to live in a pocket or hang from a keychain — kind of Cali-rad, like a skateboard charm. The Financial Times described it simply enough: an AI “charm” that can fit on a keychain. Meta’s recently launched personal agent, Muse, lives inside it, represented by a small digital character called Jolly. Zuckerberg said the company expects the hardware to be ready to ship in December. Berry saw the screen coming long before the feed learned to act.

At first glance it is easy to dismiss. Another AI object. Another piece of technology looking for a reason to exist outside the smartphone. TechCrunch reached immediately for the obvious comparison and called it “Tamagotchi-like”. I remember those from my days in the games business, and I still have one — a living, breathing remnant of that Sega Dreamcast era. There is something of that in it: a little digital companion that travels around with you, waiting to be spoken to.

But the interesting part isn’t the object. It is what the object represents. For almost twenty years, digital culture has trained us to reach for a rectangle. We searched Google, opened Instagram, watched TikTok, checked Maps, read reviews, compared menus and scrolled through photographs of pasta in Rome, bakeries in Copenhagen and sandwiches in New York until, somewhere between desire and repetition, we decided what we wanted.

The screen became the place where appetite was manufactured. Now the technology industry is beginning to imagine something different: not another app competing for space on that screen, but an intermediary between us and the world.

Muse launched earlier this month as what Meta calls a “personal AI agent”. Its description is revealing. Meta says it “doesn’t just answer questions, it actually does the work”. It can use its own browser, interact with services, fill in forms, book travel, make purchases and work across connected applications. It remembers information a person has previously shared and can use that context later. Meta even uses food to explain the idea. Save a recipe reel on Instagram and Muse can turn it into a shopping list. Tell it about friends’ dietary restrictions and it can remember them when helping to organise a dinner.

Within a fortnight that stopped being hypothetical. On the same day as the Charm, the American grocery service Instacart announced it had connected to Muse. A shopper can now name an occasion rather than a list — the company’s own demonstration phrase is “Taco Tuesday”, which tells you where it was written — and the agent interprets it, assembles the basket from a preferred supermarket, checks out and books the delivery slot. There is no British equivalent yet; Tesco, Sainsbury’s and Ocado will have watched it closely.

What is worth noticing is not the speed. Somewhere inside that one sentence, something decided how many people were eating, which cut of meat, whose brand of tortilla and which shop would supply it. None of those were questions anybody was asked.

That may sound like a clever extension of the chatbot. It is more significant than that.

Think about the progression. We searched. Then we scrolled. Then we started asking. The next stage is that we may not need to do very much at all. Muse is also being moved into Meta’s glasses. The company says its AI eyewear can already provide things such as local restaurant recommendations and translation, and that connecting the glasses to Muse will allow the agent to manage more of a person’s daily life without them reaching for their phone. So perhaps, a few issues from now, Eat Your Screen won’t arrive as something you scroll through at all. You’ll be connected to it through the device.

That changes something fundamental about discovery. Imagine walking through Soho at lunchtime. Today you might open Google Maps, search “restaurants near me”, flick through photographs, check a rating, look at Instagram and perhaps discover that somewhere has appeared three times on TikTok this week. That is appetite built on TikTok before the door opens — demand arriving ahead of the room. By the time you choose, a complicated mixture of reviews, images, creators, PR, search rankings, advertising and algorithms has shaped the decision.

Now imagine simply saying: “I’ve got an hour. Find me somewhere nearby that I’ll actually like. Nothing touristy. Somewhere new. Book it.” And carrying on walking. No search results page. No twenty photographs. No tabs. Perhaps no menu until you arrive. The agent knows where you are. Increasingly it knows what you like. It may know what you have booked before, what you have saved, the restaurants you have rejected, the trips you are planning and the conversations that preceded the question.

The difference looks small — speaking instead of tapping — but commercially it is enormous. WhatsApp on acid, if you like: not another feed to beat, but a way of bypassing the scroll altogether. Restaurants, hotels and food brands have spent the past decade asking how to look good on Instagram, how to rank on Google and, more recently, how to appear on TikTok. Part of my day job, running a food and hospitality consultancy, is helping operators think about exactly that. But the next time I am in the room with one, I suspect we will be asking something else.

How does an AI decide that we are worth recommending?

Taste has always involved delegation. We trusted the restaurant critic. The Michelin Guide. The newspaper. The chef. The friend who always knew the new place before everybody else did. Later, we trusted TripAdvisor, Google ratings, Instagram accounts and creators with enormous followings. None of those systems simply recorded taste. They shaped it. AI takes that delegation one stage further. The recommendation layer can potentially know far more about us than any restaurant critic ever could. Meta describes Muse as an agent that learns from conversations, remembers what matters to a person and becomes more useful as it accumulates context. Users decide what they connect to it and how much access it receives; Meta is simultaneously making privacy and security central to its pitch because an agent of this kind only becomes genuinely useful when it knows a great deal.

Once an agent really does understand somebody, the temptation is obvious. Don’t show me 400 restaurants. Choose three. Actually, choose one. Book it. There is something seductive about that. The internet gave us an almost infinite amount of choice and then left us to spend huge amounts of our lives navigating it. AI offers to remove some of the navigation. But it raises another question: what happens to taste when browsing disappears? Because browsing is inefficient, but inefficiency is sometimes the point. It is the wrong turn. The strange place somebody mentions at a bar. The cookbook you pick up because of the cover. The restaurant with dreadful SEO and no meaningful social-media strategy that happens to be brilliant. The cake in a window you had no intention of buying.

Human recommendation is messy because humans are messy. Machines require signals. And increasingly, being discoverable may mean making sure that those signals can be understood not only by people, but by the agents operating on their behalf. That is why restaurants that behave more like software will matter more than restaurants that only look good on a phone.

This is where the story becomes much more interesting for food and hospitality. For the last twenty years, businesses have fought to appear on our screens. Search-engine optimisation emerged because Google became a gatekeeper. Social-media strategies followed because Facebook and Instagram controlled attention. TikTok changed the game again by making recommendation itself the product. AI agents potentially introduce another gatekeeper altogether. A screen gives us options. An agent can give us decisions. If I search for “best pizza in Naples”, dozens of businesses can still compete for my attention. I can disagree with the ranking, look at the photographs, read the reviews and disappear down a side street because I like the look of something else.

If my agent chooses the restaurant and books it, most of that contest has already happened somewhere I cannot see. The fight moves upstream. What information did the agent consult? Which sources did it trust? How much did my previous behaviour influence the answer? Did popularity matter? Availability? Price? Reviews? The preferences of people similar to me? A paid commercial relationship? Something the restaurant published itself? This isn’t simply a technology question. It becomes a cultural one. Food is particularly interesting because appetite is never just hunger. It is memory, fashion, nostalgia, geography, status, curiosity and the powerful human desire to have what somebody else appears to be enjoying.

The feed industrialised those signals. AI may begin to compress them. One consequence could be the fragmentation of the mass feed itself. Ten million people do not necessarily need to see the same viral restaurant. Ten million personal agents could construct ten million slightly different cities. One person’s London becomes sushi counters and natural wine. Another gets pubs, pies and old Italian restaurants. Someone else gets bakeries and coffee shops. A fourth barely encounters restaurants at all because their agent has learned that they would rather eat at home and has already ordered what they need. The feed becomes private. Perhaps we stop calling it a feed altogether. We call it an assistant.

There is an irony in Meta helping to build this world. Facebook and Instagram were among the businesses that helped turn the social feed into one of the defining interfaces of modern life. They built extraordinary systems for keeping people looking at screens. Now the next generation of computing is partly being sold around the idea that we should need to look at them less. Meta says its glasses are intended to help people remain present in the world rather than constantly reaching for their phones. The Muse Charm pushes the same idea into another physical form: AI that accompanies you rather than waits behind an app icon. The proposition is seductive: less looking and more asking; less searching and more delegating. Less screen, certainly, but more of the machine operating just out of sight.

Whether the Muse Charm itself succeeds is almost secondary. AI hardware has already produced its share of strange objects, abandoned experiments and grand promises. A keychain companion may turn out to be transformative, mildly useful or something that disappears into the kitchen drawer beside the old fitness tracker. What matters more is the direction of travel. Meta wants the same agent available through the phone, computer, WhatsApp, glasses and now a dedicated physical device. As the agent moves between them, the interface itself starts to dissolve.

For years we have worried about what algorithms put in front of us: the video, the restaurant, the recipe, the outrage, the product, the destination. We learned to recognise the machinery, however imperfectly, and understood that the feed was deciding what deserved our attention. The next stage is subtler. The machine does not simply decide what deserves attention; it begins to act on that judgement. Recommendation starts to become action. The story of the internet has largely been one of abundance: more restaurants, more recipes, more reviews, more photographs, more recommendations and more choice than any person could reasonably process. The agentic internet offers a strange answer to the problem it helped create. Rather than giving us better ways to navigate infinite choice, perhaps the machine will simply make more of those choices for us.

That shifts the next battle for influence away from the screen, and perhaps even away from the feed, into the answer itself: the sources an agent consults, the signals it trusts and the version of our taste it has learned to construct. Which brings Eat Your Screen back to its original question. We used to eat from the earth. Then we ate from the feed. If this next interface takes hold, we may increasingly eat from the agent — perhaps without ever seeing the machinery that chose for us.

This essay draws on Meta’s own announcements of Muse and of the Muse Charm, on reporting from the Financial Times and TechCrunch around the Connect keynote on 23 September 2026, and on Instacart’s announcement of its integration with the agent. Where a company has published its own account of what it is building, that is what is linked here rather than a write-up of it.

Illustration is original artwork made for Eat Your Screen.