In the space of three weeks this September, Meta launched Muse, Apple put Siri AI into beta, and OpenAI introduced dots. All three are personal agents: AI that does not just answer questions but reads your email, connects to your apps and takes actions for you.
Someone on your team is probably trying one already. The useful question for a business is not which agent is smartest, but which one you would trust with your company inbox. I compared them on four things that decide that, using each company's own published answers.
Where each agent runs
Muse runs in what Meta calls a Muse Secure VM, a dedicated computer in Meta's cloud for each person. A separate agent, Sentinel, sits beside it and has to approve anything Muse sends to the internet.
Each dot gets its own cloud computer and browser at OpenAI. Your own laptop stays separate unless you choose to connect it.
Siri AI works on your device where it can and sends the rest to Apple's Private Cloud Compute, Apple's own servers for AI. Apple says the models were built in collaboration with Google's Gemini.
Two of the three work in their company's cloud by design; Siri AI keeps as much as it can on the phone in your hand.
Who learns from your conversations
This is where they differ most. According to CNBC, Muse users have to opt out if they don't want their conversations used to train Meta's models. If they stay in, Meta says it scrubs "critical personally identifying information" first.
With dots, personal ChatGPT plans follow your "Improve the model for everyone" setting. Business, Enterprise and Edu workspaces are not used for training by default. OpenAI's privacy FAQ adds that people may still review activity in limited cases, such as safety, even with that setting off.
Apple says that when Private Cloud Compute handles a request, personal data is not stored or made accessible to Apple or anyone else.
For company data, the default setting matters more than the privacy page, because most people never change it.
What you can take back
Muse lets you tell it to forget specific things it has learned.
With dots, you currently cannot view or delete individual memories. You can delete the whole dot, which removes its context, but files and chats it created stay where they are.
Apple's launch page does not say how Siri AI's memory of your personal context can be reviewed.
If you cannot see what an agent remembers, assume it remembers everything you connected.
Who approves before it acts
Muse checks with you before sensitive actions like sending an email or making a purchase, and keeps an audit trail of what it did and plans to do.
Dots run each risky action past an automatic review first. The most sensitive ones, like changing a password or transferring money, you have to finish yourself.
Apple's page does not say when Siri AI asks before acting in other apps.
An agent that can act should tell you, in writing, which actions it will never take alone.
Which one to trust with what
My answer is less about the brand and more about the data.
- Company email, customers and files: only on a business plan where training is off by default and someone in your team controls the account. Of the three, only OpenAI states that today, for dots on Business and Enterprise plans.
- Personal context on a phone: Siri AI's on-device design is the most cautious of the three, though it is still in beta.
- Anything else: read the default, turn off training if you don't want it, and connect one app at a time.
Check where you operate, too. Muse launched in the US only, dots in "eligible markets", and Siri AI is not on iPhone, iPad or Apple Watch in the EU, or in China, at launch.
The Personal Agent Protocol: your side of it
Soon these agents will not just work for your team. They will turn up at your business on behalf of your customers, to check an order, book an appointment or ask for a refund.
The Personal Agent Protocol, announced on 6 October by Sierra and Meta with Shopify, Stripe, Walmart and others, is a draft standard for that moment. It lets a business decide what a customer's agent may do, from blocking it entirely to letting it act with the customer's approval. The customer signs in on the business's own page and chooses whether their agent can view their account, make changes, or both.
It is still version 0.1, and OpenAI and Apple are not among the founding members. The direction is clear, though: a business will need something on its side for these agents to talk to. That is what we keep building in our custom AI agent and helpdesk work: an agent connected to your own systems, following your rules, that knows when to hand a decision to a person.
In my experience, trust in these tools rarely comes from the launch video. It comes from reading the default settings.
Getting ready for customers who send their AI agent instead of calling? BrillMark builds custom AI agents and helpdesks around your own knowledge, systems, rules and channels. Get a free working prototype built on your website and one of your customer journeys.
Top comments (0)