Investors are suddenly fixated on “personal AI agents.” They are wondering: Are personal AI agents going to replace all apps (or at least replace travel apps)? Are they going to collapse all software into a single user interface (or at least eliminate many single-purpose apps)?
On social media, posts abound about using personal AI agents to accomplish varying tasks: cancel subscriptions, negotiate bills, book trips, reserve restaurant tables, purchase fancy items, arrange an entire wedding… The list goes on.
Having used various personal AI agents like Muse, Instinct, and Dot, I have become increasingly convinced that what we think we are seeing—those use cases on social media, those expectations we have, those grand visions that feel so real to us—many of these are probably not enduring, but a mirage.
Enduring Use Cases, It Cannot Deliver
Personal AI agents, assistants, helpers. What are the enduring demands that we humans have for them?
I lived in Hong Kong for almost a decade. In Hong Kong, it is extremely common for families to hire “foreign domestic helpers,” a phenomenon that started in the 1970s. Today, Hong Kong has nearly 400,000 “foreign domestic helpers,” so many that this group constitutes more than 5% of Hong Kong’s total population.
In Hong Kong, those foreign domestic helpers live at their employers’ homes. What are they hired to do? Cleaning, cooking, taking children to and from school, elderly care, pet care.
In the U.S., families with financial means often hire housekeepers or nannies, or both. The former does cleaning work; the latter does childcare.
Decades have gone by. Those are the time-tested, enduring use cases for personal agents, assistants, helpers.
If these use cases are made free to you (like most of today’s personal AI agents, “free to use!”), I bet you will say “Yes, give it to me now!”
Unfortunately, for real enduring use cases, today’s AI agents cannot deliver. Not even close.
Experiences Consumers Enjoy, It Tries to Take Over
Consumer clients are not corporate clients. Consumers are not consistently looking for maximum productivity gains. What consumers want is personal fulfillment.
That is why some of the most popular consumer apps are social media apps (humans are social animals), entertainment apps (a fun way to kill time), and shopping apps (a pleasant experience to figure out what we actually want to buy). Consumers love those experiences, because they are fun.
Here is another angle to see this. Our species (Homo sapiens) has been around on this planet for about 300,000 years. For well over 95% of this time, we were hunter-gathers—I guess, we are hardwired to like discovery, exploration, seeing things, trying a variety of experiences. We did not start writing until some 5,000 years ago, well under 2% of our evolutionary history.
Visually browsing, finger clicking, those are primitive instincts. We probably all know people who are addicted to watching short videos on their phones. What does that experience feel like? “I don’t feel like using my brain!”
Writing is a learned behavior. Writing (and reading) is mental work. Do you know a lot of people who are addicted to writing? After a busy day at school or at the office, what do you say? “My brain is toasted.”
Back to today. Today’s personal AI agents appear to be taking these fun life experiences away from consumers (“One text message to my AI agent… Boom, my new shoes are ordered”). And, if you really believe that browsing shopping websites and viewing travel blogs are “work” and therefore they should be streamlined away, how about you have an AI agent shop the cheapest hotel and the cheapest flight for you, and have the AI agent take the entire vacation on your behalf. Efficient, indeed. That’s better for you, isn’t it? You don’t even have to leave your sofa for the entire thing. By the end, you will receive a text message from your AI agent—a 20-page AI deep research report summarizing for you that travel experience. Now, it is your turn to read it and write a response. Are you deeply satisfied now?
Are you going to have a personal AI agent live your life for you, exist for you, so you may rot?!
Therefore, I argue that the text interface that personal AI agents rely on has its inherent limitations. To a degree, the text interface is against human nature.
And more importantly, consumers do not want some AI tools to live their lives for them.
The China Case
China’s Alibaba, in a sense, occupies the most advantageous position that anyone can think of when it comes to building a consumer AI experience.
As one company, Alibaba controls the following:
- AI models
- AI compute
- Consumer awareness
- E-commerce ecosystem (e-commerce, food delivery, grocery, OTA, maps, health, etc.)
- Payments
- Logistics and last-mile delivery networks
So, on the supply side, Alibaba has it all. And of course, instead of being blocked (like Amazon blocks Muse), Alibaba’s own e-commerce systems are connected natively with Alibaba’s own AI agents.
On the demand side, Chinese consumers are probably among the most ideal users of personal AI agents. They are tech-savvy, always on their phones. They are price sensitive and they love comparing prices.
In early 2026, Alibaba ran a giveaway campaign, giving away free items and shopping vouchers to consumers who were willing to try Alibaba’s consumer AI agents. In total, about 200 million orders were placed by consumers using Alibaba’s consumer AI agents (so as to get free stuff!).
Guess what? After massive subsidies, Chinese consumers went back to using incumbent apps. Today, very few Chinese consumers are habitualized into using Alibaba’s consumer AI agents to shop or compare prices.
If Alibaba, with full stack (software to hardware, e-commerce ecosystem to logistics networks), with ideal consumers (tech-savvy, price-sensitive), and with massive subsidies (free items, free vouchers), still cannot find a way to make people use consumer AI agents, I think it sends one loud message to us.
Perhaps, it is not that the product is not good enough. It is that consumers do not want it!
For Personal AI Agents to Make Money, That Is a Tall Task
Mark Zuckerberg has said in recent days that Muse will make money by taking “a small fee” from merchants (source: Meta Connect Keynote 2026). Instinct’s Noah Shinn has said in his first long-form interview that “I see a blanket transaction take rate being enforced across the platform, which is just us exchanging distribution for being able to serve products on behalf of merchants.” Both men have expressed their desire to keep their AI agents free for consumers to use.
Thinking laterally, I can see that those ideas make a lot of sense. That is how many of today’s businesses (Visa, Mastercard, Apple, Amazon, DoorDash, Booking, etc.) make money.
But wait… Think about it again…
How to make merchants pay?
How can a personal AI agent “enforce” a take rate on merchants?
What if some merchants say “No, I do not want to pay you”?
Would a personal AI agent block these “non-compliant” merchants? What does that mean? So, your personal AI agent would then not mention these merchants to you? Your AI agent only tells you half of the story, not the full story?
Does my agent represent me? Or represent some merchants?
A personal agent, by definition, should represent me and my interests.
Then, your agent only tells you half of the story… Is that a form of lying?
Then, would you still use that AI agent?
Let’s stretch it a bit more. Translate that dynamic into real-life terms. Imagine you have a real human personal agent (a butler, an assistant). He or she goes onto the main street, talking with a bunch of merchants, asking for “kickback” money. If a merchant gives an X% kickback to your personal agent, your personal agent will then spend your money at that merchant. If a merchant refuses to cooperate with your agent, your agent will then skip that merchant, not mentioning to you there is a great store here and another great shop there.
What the f^.^k is that?
Will you still employ your personal agent then?
I reason that the world in which AI agents can comfortably charge merchants is the following: One or two personal AI agents become absolutely world dominant, like today’s Google and Facebook. In that world, merchants will then be willing to pay those middlemen in exchange for consumer demand.
The key to open that world is called “distribution power,” and you need a really big one!
Today, any sensible observer would agree that consumer time and attention are already being dominated by giants like Google, Meta, TikTok, and Amazon. There is little spare time and attention left. Low-hanging fruit is mostly gone.
To imagine a world where AI agents can dominate consumer time and attention is to imagine an alternative world where the following realities take place: most of the world’s consumers shift their time and attention away from Google, Meta, TikTok, and Amazon to one or two dominant personal AI agents.
In that world, those one or two dominant personal AI agents would then possess enough distribution power to “enforce” a take rate on millions of merchants.
In order for that scenario to happen, today’s giants Google, Meta, TikTok, and Amazon would have to lose their connections with billions of consumers, and be reduced to a shadow of their former selves. (Meta owns Muse, but for the sake of making my case, allow me to describe it that way.)
What is the likelihood of that happening?
If today’s giants continue to own consumer time and attention, and personal AI agents continue to be free to use, how can personal AI agents make money then? I don’t know an easy answer!
Making it even more interesting, in prior cycles, apps scaled in an asset-light way. Today, personal AI agents scale in an asset-heavy way. AI inference—the compute is not cheap! If personal AI agents cannot find a way to monetize quickly, and they do not have “deep pockets,” they will quickly run out of cash and they will probably have to pivot or shut down. Time is not their friend.
Conclusion
To be clear, I think it is totally possible that consumers might use personal AI agents more, in particular for things consumers consider “work”—those low-fulfillment tasks, such as tracking flight prices, tracking refunds.
What I am skeptical of is today’s inflated expectations, as reflected in the public discourse, and in the valuation of some start-up companies in this space.
The things I really need help with, you cannot do. The things I love, you try to take away. The world where you can make money is a world that is highly unlikely to come to pass. And you burn cash fast. So, I do not know how you can last.
A mirage does not last.
(END)
