I think the race to put AI on new hardware, glasses, a tiny companion device, misses the actual problem most businesses have, which is that nobody wants to open another app. The AI tools I build for clients live inside WhatsApp and email, the places people already check all day, because an alert or an answer nobody sees is worthless no matter how good the model behind it is.
Meta's personal AI agent Muse is reportedly outpacing ChatGPT's early numbers, and Meta's next move is to put it on smart glasses and a small Tamagotchi-style device. Read that twice. The plan isn't to make the agent smarter, it's to give it a new body to live in. I understand why a consumer platform does that. I don't think it's the lesson a business should take from this week.
A new device is a bet that people will open one more thing
Every AI agent needs a home, somewhere a person goes to ask it something or somewhere it can interrupt them when something needs attention. Meta is testing whether that home should be glasses on your face or a gadget on your desk. That's a fair bet for a company trying to own an entirely new category of hardware. It is a bad bet for almost every business I've worked with, because the honest answer to "where should this live" is usually the app that's already open.
I learned this the boring way, running ad performance alerts for my own marketing work. I set them up to arrive in two places, on WhatsApp and by email. Email is where I go to check the record, to see what happened and when. WhatsApp is where I actually notice something's gone wrong while there's still time to do anything about it. An alert that only lands in an inbox nobody opens during the day isn't an alert. It's a report you read after the money's already spent.
The same pattern shows up inside a business, not just outside it
I built an internal AI tool for Vimal Electric Corporation, the distribution business I run with my family, and the reading side was the easy part. It reads our documents and price lists, and it reads our past WhatsApp and email conversations, because that's where most of the real history with each retailer actually lives, not in a CRM nobody kept updated. That part I expected to matter.
What I didn't expect was how much the asking side mattered just as much. The team can ask it questions on an internal web page, which I built assuming that's where people would go. But they mostly ask it on WhatsApp instead, because that's where they already spend their day. A tool people have to open is a tool they forget exists. A tool they can message like a colleague gets used without anyone having to be reminded to use it.
Not "can we build this on a new channel" but "where does this person already look forty times a day, and can we put the answer there instead."
Where I think Meta's move actually matters
None of this means Meta is wrong to chase smart glasses. A consumer product trying to become part of someone's daily life for entertainment or convenience benefits from novelty, from being somewhere new and slightly magical. A business trying to stop losing money on a bad ad campaign, or trying to get a warehouse team an answer about stock without calling three people, doesn't need novelty. It needs the answer to show up where the person already is, without friction, without a new habit to form.
I'd go further. I think the interface question for business AI is largely already settled, and it was settled by WhatsApp Business API and email long before Muse existed. What's still open is whether the AI behind that interface actually knows anything useful, whether it can read your price list, your past conversations, your actual history with a customer, rather than just holding a generic chat. That's the harder, less glamorous problem, and it's the one I spend most of my time on.
A different kind of interface problem I've been solving lately
There's a related shift happening that has nothing to do with new hardware: getting a business's own site quoted by ChatGPT, Perplexity or Google's AI answers, rather than just ranked in a list of blue links. When I worked on this, the change that mattered most wasn't technical. I rewrote pages so each section answered one specific question directly, the way a person would actually ask it. Only after that did I add the technical layer, structured data describing what each page is, and an llms.txt file telling AI tools what the site covers.
The order mattered more than I expected. Schema markup on a page that never clearly answers anything does very little. A page that answers one question well, in a way a machine can also parse, is the kind of page these tools actually quote. It's the same principle as the WhatsApp alerts and the internal tool, just pointed outward instead of inward: meet people, or the systems acting on their behalf, exactly where they're already looking, in a form that doesn't need translating.
Say a clinic wants patients to get their AI answers through a website chatbot because that's the modern thing to have. If the patients are already texting the front desk to ask about appointment times, that chatbot will sit there unused while the front desk keeps answering the same question by hand. The fix isn't a smarter chatbot. It's putting the answer in the text thread.
I'll be curious to see whether the glasses and the little companion device actually change how people use Muse day to day, or whether most of that usage keeps happening through whatever app was already open on the phone. My guess, based on nothing but what I've seen building these things for smaller businesses, is that the device is the interesting headline and the existing channel is where the actual habit forms.
