Watching Muse and this new wave of personal agents, I keep noticing a shift in what we expect from AI.
We used to open an AI app with something of a testing mindset: Can you answer this question? Write this code? Explain this article? Increasingly, the thought is becoming: Can I leave this with you while I get on with something else?
There is a long distance between those two expectations. Answering a question requires understanding the conversation in front of you. Taking on a task means remembering the background, respecting constraints, using tools, and continuing when something goes wrong.
Meta’s Muse brings memory, app connections, and ongoing execution into one product. According to its official introduction, users can hand over a task, leave the interface, and let it keep working until it needs their approval.[1]
A product launch does not mean every problem has been solved. But it does let us ask a more concrete question: what would an agent of our own actually be useful for?
I think the answer begins with the small things we never quite finish.
An email waiting for a reply. A refund we should request. A subscription we forgot to cancel. A trip we still haven’t organized. None of these is particularly difficult on its own. What wears us down is that they stay on our minds, and often send us back and forth between several apps.
The order is here, the policy is somewhere else, and customer support has a separate entry point. Once you have submitted everything, you still have to remember to check back in three days.
We already have plenty of tools. Connecting the work between them is still our job.
That is where the opportunity for personal agents lies.
Until now, asking AI to help with a refund often meant asking it to write an email. An agent we could rely on would need to find the order, check the conditions, prepare the information, and follow up until the matter was resolved.
The moment that makes someone willing to pay might come a few days later, when they suddenly remember the task and discover it has already been taken care of.
That value can also accumulate over time.
Take travel planning. The first time, you might need to explain your budget, hotel preferences, daily routine, and where you are willing to compromise. The next time, it should remember. Eventually, it should also understand that you make different choices on business trips and holidays, and that rejecting a hotel because of its inconvenient location does not mean you dislike that kind of hotel altogether.
The weight of the word “personal” is in those details. The agent should gradually reduce the work of explaining yourself.
In the past, that kind of ongoing attention usually required another person’s time, which made it expensive. If AI can take on part of that work, assistance once affordable to only a few could gradually become part of ordinary life.
There is still a substantial hurdle, though: does it actually give us less to worry about?
If I have to check every step, explain the background before every action, and fix its mistakes afterward, I have simply acquired another assistant to manage.
So the value of a personal agent needs to be calculated in full. From the time it saves, subtract the time spent briefing it, supervising it, and correcting it. What remains is the benefit the user actually receives.
This is also why I think its business model deserves more attention than those of many other AI products. Once we start entrusting it with things, how it makes money will influence what we are willing to entrust to it.
Subscriptions are the easiest model to understand. I pay a monthly fee in exchange for ongoing help. Muse currently offers free access alongside paid plans at $20 and $100 a month.[2]
For users, the test is straightforward: did it take enough off my hands this month?
For the company, the challenge is the cost of providing the service and how often people use it. Some users might need help a few times a month; others might hand over time-consuming tasks every day. The same request—“Can you take care of this?”—could mean a few seconds of work or hours of searching, waiting, and trying again.
The price has to feel worthwhile to the user while covering the actual cost of execution. There are many details between an impressive demonstration and a business that can operate sustainably.
Another route is charging for transactions. Meta has already expressed an expectation of earning small fees from transactions in the future.[3]
The logic is clear enough. An agent finds products, makes reservations, and helps complete purchases, while the platform earns a fee. The closer it gets to a transaction, the easier its commercial value is to see.
But if my agent helps me shop, will it also be willing to tell me that I don’t need to buy anything?
It is a small question, but one that can reveal a great deal about its relationship with the user.
Commissions do not inevitably distort recommendations, and subscriptions do not automatically guarantee loyalty. What needs to be clear is whether commercial partnerships affect rankings, whether users can understand why an option was chosen, and whether the platform allows its agent to make decisions that benefit the user even when they reduce transaction growth.
A product that acts on a user’s behalf must make room for this possibility: the better it does its job, the less the user buys.
This gives Muse a tension worth watching. Large platforms have accounts, distribution, app connections, and infrastructure that make it easier to bring agents to a mass audience. At the same time, users will ask them to demonstrate something: when my interests and the platform’s interests do not fully align, what will you do?
The answer will have to be established through individual, concrete actions.
As for who might be willing to pay consistently first, I would look to people running a business largely on their own: freelancers, creators, small shop owners, and very small teams.
They have recurring work every day, and it is easier for them to judge what a result is worth. One fewer missed customer follow-up, a quote sent on time, or a few hours less spent organizing information each week can all create a clear reason to pay.
Muse’s expansion into small business, with connections to tools such as Shopify, QuickBooks, and Stripe, fits that direction.[4]
To me, these situations may offer a more solid commercial starting point for personal agents. They help one person handle more of the work, giving people who cannot yet afford a full team some of the support a team would provide.
Still, an all-purpose super-agent is unlikely to become part of everyone’s daily life overnight. A more plausible path is that it first takes on one clearly defined kind of work, does it reliably enough, and gradually earns permission to take on the next.
That is how trust accumulates. It remembers your constraints several times in a row, avoids making decisions beyond its authority, and comes back to you promptly when something unusual happens. One day, you notice that you no longer keep opening the interface to check its progress.
That is when a personal agent has truly become part of your life.
The future I am hoping for is not especially grand.
You are getting ready for bed when you suddenly remember a small task you have been putting off. You open your phone and see that it has already been handled.
Then you put the phone down, with one less thing on your mind.
For an ordinary person who always feels there is more to do, that is already something worth looking forward to.
Sources
- [1] Meta: Introducing Muse: The World’s First Personal AI Agent Built for Everyone. https://about.fb.com/news/2026/09/introducing-muse-personal-ai-agent/
- [2] Axios: Meta courts small business with new Muse agent. https://www.axios.com/2026/09/29/meta-muse-ai-small-business
- [3] Axios: Meta needs you to believe it cares about privacy. https://www.axios.com/2026/09/25/meta-ai-muse-privacy
- [4] Meta: The Future Is for Everyone: Muse for Small Business. https://about.fb.com/news/2026/09/introducing-muse-small-business/amp/