by Tan Aik Keong (AK)
If the last three years of AI hype were an arms race over whose model is smarter, Meta's newly launched Muse may mark the start of the next phase: AI that no longer just answers questions, but starts doing things for you.
This September, Meta officially released its personal AI agent, Muse. Compared with the ChatGPT-style chatbots we've grown used to, Muse is positioned closer to a "digital butler." According to Meta, Muse can browse the web, fill in forms, connect to authorised apps, and handle multi-step tasks — helping with shopping, booking trips, dealing with customer service, and sending emails.
That might sound like just another incremental upgrade. The change underneath it is actually much bigger.
From "person operates software" to "AI acts for you"
For more than two decades, the way we've used the internet has basically been "person operates software." We open Google to search, then click through to different websites; we open an e-commerce app and compare products ourselves; we open an airline, telco or banking app and work through a task step by step.
The agent era may run on a completely different logic. In future, we may not need to face dozens of separate apps at all — we might simply tell an AI what we want done, and leave it to decide which page to open, which service to use, and which details to fill in.
Take a familiar example: a consumer notices their credit card is still being charged every month for a subscription they forgot about. The old way is to work out which company it is, log into the account, hunt down the subscription-management page, and cancel it yourself. US media have reported that Muse can already help users handle exactly this kind of task.
That's the real difference between an AI agent and a traditional chatbot. In the past, we asked AI: "How do I cancel this subscription?" In future, we may simply say: "Just take care of it." Going from "tell me how" to "do it for me" looks like a small change in wording, but it could end up redefining our whole relationship with the internet.
Beyond the screen: into smart glasses
Meta clearly doesn't want Muse confined to a phone or computer screen. At this year's Meta Connect, Mark Zuckerberg announced that Muse will also make its way into Meta's AI smart glasses. Smart glasses matter here because they give AI a direct window onto the real world — what you're seeing and hearing.
Consumers have long asked a very practical question about smart glasses: why would I actually wear one? If all a pair of glasses does is move phone notifications in front of your eyes, the value is limited. But once glasses can understand what's in front of you, and combine that with an AI agent's ability to act, the equation changes.
In future, when a user notices a product, a restaurant, or an event poster, they could, in theory, ask AI about it on the spot and then ask it to take action. To be clear, this describes a possible direction once Muse is combined with smart glasses — not a claim that every such scenario is a mature feature today. Meta also showed off a small Tamagotchi-like companion device, reportedly named Muse Charm, which tech media describe as still in development, with Meta saying at the time that it planned to launch it properly later.
The real prize: becoming the internet's next front door
Why is Meta pushing this so hard? Because the AI race may ultimately not just be about whose model is smartest — it may be about who becomes people's main gateway into the digital world every day.
Google has long owned the search gateway; Apple controls the smartphone gateway; Microsoft has long held the PC and Office productivity gateway; and Meta itself commands enormous platforms in WhatsApp, Instagram, Facebook and Messenger. If users get used to telling an AI what they want done, rather than opening a specific app first, AI agents get a real shot at becoming the internet's new front door.
That could also give rise to what's being called the "Agent Economy." E-commerce sites today are built for people — how the homepage is laid out, how ads appear, how to keep a shopper browsing a few extra products — all of it is designed around the assumption that a human is personally visiting the site. But if more and more shopping starts being done by an AI agent searching and comparing on a consumer's behalf, then presenting a handful of results directly to the user, a lot of business models will need rethinking.
Some of this is already playing out. Shopify has started supporting Muse-related AI shopping experiences, letting qualifying merchants reach consumers through agents. Amazon, by contrast, has moved to block Muse from directly making purchases on a user's behalf on its platform, citing concerns including privacy, security, transparency and verifying third-party agents' identity.
That foreshadows a big argument the internet is about to have: if AI starts acting online "as a person," does a website have the right to block it? And how should a platform even tell whether the thing operating on it is the actual person, or an AI acting with their authorisation?
The bigger question is trust
A chatbot that only answers questions is low-stakes even when it gets things wrong — worst case, you simply ignore the answer. But an agent that can buy things, send emails, manage subscriptions and interact with other websites on your behalf is a different matter entirely once it makes a wrong call.
So an AI agent's real competitive edge may not be how "smart" it is, but how trustworthy it is. Meta says Muse will ask for confirmation before sensitive actions like making a purchase or sending an email, that users can choose which services to connect, and can review a log of what the AI has done; Meta also says some sensitive data is handled through a separate, independent security mechanism.
All of that, though, is still just a company describing its own system design — the real test comes from actual use. Since launch, US tech media have already reported privacy and permissions controversies around Muse: some users questioned why it was able to reference information from Messages, and Meta responded that the relevant permission had previously been switched on, while also acknowledging that Muse's own explanation of where its information came from wasn't accurate.
That episode is worth sitting with. The question we'll increasingly face isn't just "how much does this AI know" — it's "exactly how much does it know about me." And that may be the question the agent era most needs society to seriously debate.
Judging AI by time saved, not demos
I've always believed that whether a piece of technology has truly matured shouldn't be judged by how many jaw-dropping demos it can produce, but by how much time it actually saves an ordinary person, every single day.
An AI generating a beautiful image is impressive. An AI writing an article for you is convenient. But if AI can save someone even 30 minutes a day — sorting email, managing subscriptions, comparing products, arranging schedules, tracking orders, and clearing away all sorts of small digital admin — that adds up to more than 180 hours a year. That may be where AI agents are genuinely disruptive.
For the past few years, when we talked about AI, the question was usually: "Will AI take my job?" With products like Muse now here, a different question may start mattering just as much: will we first let AI take over the most tedious, repetitive tasks in our lives — the ones we can't avoid doing, but would rather not?
As AI moves from being something we chat with to something with vision, memory, tools and the ability to act, the computing platform that follows the smartphone may not simply be an even more powerful phone. It might be a pair of glasses. It might be a small wearable device. But what actually matters isn't what the hardware looks like — it's the possibility that, one day, we stop operating the internet ourselves. We just tell AI: "This is the result I want — go get it done."
When that day comes, the question we face will shift from "how smart is AI, really" to something far more concrete: how many of the decisions and actions in our lives are we actually willing to hand over to it?
Part of the AK AI Corner column. Originally published in Oriental Daily (东方日报) on Sep 25, 2026.