Thematic Pulse
All I want for Christmas…is a personal AI assistant
Imagine Christmas without the endless to-do list. Your personal AI assistant finds the perfect gifts, books the family getaway, coordinates everyone’s calendars and makes dinner reservations — all while you focus on the festivities. Sounds like a Christmas wish? AI agents are bringing us closer to that reality, moving beyond answering questions to actually getting things done.
What a personal AI agent is, and who is building them
A chatbot answers questions. An agent carries out tasks. It plans, uses websites and connected services, works through several steps and keeps going after you close the app, returning when something changes or when it needs your approval. The idea is to replace a dozen apps and browser tabs with one conversation in a place you already use, such as a messaging app.
The agent is always on. It runs in a single, persistent chat in which the context from all previous conversations stays loaded, so you don’t need to repeat yourself. It also runs on its own dedicated computer, so it keeps working while you’re offline. This may sound like a minor feature, but in our testing it made a substantial difference.
The category is only weeks old, and several providers launched within a short window:
Meta’s Muse launched on 8 September. It runs on a private cloud computer that holds the agent and the user’s data, and is rolling out first in the US on iOS, Android, the web and WhatsApp. Third-party estimates put it above 2 million daily users within five weeks of launch.
- OpenAI’s Dots launched on 29 September for ChatGPT Pro and Business Premium users in eligible markets. Personal Pro access initially excludes the EEA, Switzerland and the UK.
- Instinct, a start-up, launched its agent in August. You text or call it, and access is invite-only.
- Google’s Gemini Spark has also rolled out. It is a personal agent that works in the background.
For readers in Europe and Asia, most of these products are US-first or invite-only today, so you can’t try them yet. We haven’t seen confirmed timelines for a wider rollout, but the direction of travel is clear. To give you a first-hand view, we gained access to one of these agents and tested it for you. What follows is what early users can already do.
Field notes: we tried it so you don’t have to (unless you want to)
Muse hasn’t reached the EU yet, so our team tested Instinct instead. It lives in WhatsApp and connects to whichever apps and accounts you allow. What you do with it is limited mainly by your creativity and your risk tolerance. Here’s what we tried.
“146 mailing lists. One message.”
|
VLADA YATSKOVSKAYA, Associate Portfolio Manager
|
“I started small. I asked it to send me flat listings that matched my criteria (I hope my landlord isn’t reading this) and tracked my calories and nutrients by sending it photos of my meals throughout the day. Then I got braver and, at my own risk, connected my email. It listed every mailing list I was on: 146 of them, mostly e-commerce sites. I named the ones to keep, and it unsubscribed me from the rest. It also booked a restaurant table online from nothing more than a date and a time.”
“Less life admin. More living.”
|
NEZHLA MEHMED, Portfolio Manager
|
“I turned 20 open tabs into one message: ‘Plan a weekend in Rome within my budget.’ Instinct compared flights and hotels and built an itinerary around that budget. Then I asked, ‘Where could I save each month?’ It spotted forgotten subscriptions, compared recurring bills and found better savings rates. I’m building trust with the app step by step. Next, I want to give it access to my email and calendar so it can do more in the background, while I keep the final say on big decisions. The ambition? To have it help me organise a private event for more than 50 people, from finding venues and writing to them to keeping track of quotes, guest logistics and deadlines.”
“One new chat. A few new habits.”
|
NEDIALKO NEDIALKOV, Portfolio Manager
|
“The biggest change for me was that the agent is always on and busy… and it knows me. That doesn’t sound like a lot, but the difference is substantial. I no longer had to remind it of my daughter’s shoe size and her favorite color, my preferred online shops and logins, or how much of the predefined budget was already spent. It navigated websites and filled out contact forms to find accommodation for our ski holiday, while finding time in the family calendar for doctors’ appointments or the kids’ Halloween events became mostly a one-liner. Soon I stopped typing and just talked to it on the go. It isn’t perfect yet: I ran into blocked access, missing tools, some misunderstandings and a model less considerate than other chatbots I use. Still, the tool’s potential clearly shone through.”
But trust is the product
A personal agent is only as useful as what it knows about you. To book the right villa, it needs your dates, your budget and your family’s dietary needs. To buy a gift, it needs your payment details and a sense of your relatives’ tastes. To coordinate calendars, it needs access to yours. The more it can see, the more it can do for you.
That is also why security and trust are critical to the success of these tools. An agent that reads your inbox, remembers your preferences, signs in to your accounts and spends money on your behalf holds the keys to a large part of your life. Coverage of the launches notes that giving an agent more access to personal data widens what it can do but also creates safety and reliability risks. Productivity and security pull in opposite directions, and trust sits where they meet: users will share more only if they believe the agent is safe, and the agent becomes more useful only if they share more.
Providers say they have built protections around each part of the interaction.
Below are their own descriptions; we found no independent audit of any of them.
- Memory and personal data: Muse lets users inspect and edit stored memories; Dots and Instinct offer controls over shared or indexed data.
Open issue: granular memory management remains inconsistent across providers. - Secure environments: Muse and Dots run in isolated cloud environments that separate user data and limit agent access.
Open issue: independent security audits remain limited, and Meta says it is developing additional privacy protections. - Permissions and autonomy: Muse uses a dedicated permission agent to authorise actions; Dots lets users choose when approval is required.
Open issue: the balance between seamless automation and sufficient human oversight is still unsettled. - Credentials and payments: Muse shields passwords and uses single-use payment cards; Dots protects logins through encrypted credential services; Instinct offers a secure Vault.
Open issue: industry-wide standards for agent-led authentication and payments are still emerging. - Privacy and data protection: Providers describe encryption, training-data restrictions and user privacy controls.
Open issue: data usage policies, transparency and regulatory compliance still differ across providers. - Prompt injection: Muse and Dots use safeguards to stop malicious instructions embedded in websites or emails from manipulating the agent.
Open issue: prompt injection remains an unresolved, industry-wide security challenge.
The usual advice for users is to start small and review each authorization individually.
In agentic AI, model intelligence and a mature ecosystem of connected services drive capability, but trust determines how much of that capability users will actually adopt. We are in the early innings of the agentic transformation, and much remains unknown: how agents will be regulated, which use cases will prove most valuable and whether consumer behaviour will fundamentally change. Like every technological shift, this one will create winners and losers. Who they will be remains to be seen.
