For two years the question was whether AI assistants would ever do real work, or just talk about it. In September 2026 that question got its answer from the largest companies in the field.

On 8 September Meta launched Muse, a personal agent that sends email, books travel, fills in forms and pays for things. Three weeks later, at its DevDay, OpenAI introduced Dots: always-on agents that each run on their own cloud computer, connect to thousands of apps and keep working when you don't. In June, Town raised $55 million from Andreessen Horowitz and Forerunner for assistants with their own email address. And Microsoft has been rolling out an agent mode in Outlook that triages mail, drafts replies and resolves calendar conflicts, starting with organisations in its Frontier programme.

The category has changed name. We are no longer talking about assistants you ask. We are talking about agents that work.

If you run a company in Europe, here is what that means for you.

What the new agents agree on

Read the launch posts side by side and one thing stands out: every serious agent now puts approval at the centre.

Muse places a separate agent, which Meta calls the Sentinel, between Muse and the internet. Nothing Muse does reaches the outside world unless the Sentinel approves it, and sending an email or making a purchase needs your confirmation.
Dots works with default rules for when a dot needs your approval, plus custom rules you set per action: allow, require approval, or block. Before an email goes out, an automatic review checks the recipient and the message. Changing a password or moving money always hands control back to you.
Town waits for your approval before email goes out, unless you have marked a contact as always allowed.

This is good news. It is also what we have argued since we started Prio: an agent should climb a trust ladder, not start at the top. Our own line for it is on our homepage: delegate the work, keep the signature. The market has converged on the same idea. The differences are now in the details: what runs without asking, what an agent knows about your work, and where it is allowed to run.

Where they differ, seen from Europe

Availability

At launch, Muse is available in the United States only. Dots is rolling out to ChatGPT Pro users in every market except the European Economic Area, Switzerland and the UK. Companies on ChatGPT's Business Premium plan can get Dots in all supported regions, including Europe, but an individual founder on a personal subscription cannot.

That will probably change over time. But unless your company is on Business Premium, the largest launches are not yet an option if you want an agent for your working day in Amsterdam, Berlin or Paris this quarter.

Where your data is processed

For a European company, "available here" and "processed here" are different questions. Even inside Microsoft 365, Microsoft's own documentation notes that the Anthropic models offered in Copilot are currently excluded from its EU Data Boundary. None of this is hidden, but it is easy to miss, and it matters if your clients ask where their correspondence goes.

Breadth or depth

Dots and Muse are built to do almost anything. Dots connects to more than 4,000 apps and works on its own cloud computer. Muse runs errands: travel, shopping, payments, forms. That breadth is impressive, and for some people it is exactly right.

The work of a founder or an executive is narrower and deeper. It is the 60 emails since yesterday, of which four matter. The board meeting in nine days that needs a pack. The investor you promised an update to. The renewal nobody flagged. Doing that well depends less on how many apps an agent can reach and more on what it knows: who matters to you, what you promised, what happened last time.

Outlook, and your language

Many European companies run on Microsoft 365. An agent that treats Outlook as an afterthought is not built for them. And a Dutch or German executive writes to half their contacts in their own language. An agent that drafts everything in English creates work instead of removing it.

Five questions before you hand any agent your inbox

Where is it available, and where is my data processed? Ask for the regions, the subprocessors and whether your data is used for training.
What does it do without asking? Not in general, but per kind of action. Can you read the rules, and can you change them?
Whose name is on what it sends? Does it write from your own address, or from its own? Both can be fine. You should know which.
What does it know about my work? Does it remember your relationships, your commitments and what happened last year, or does every task start from zero?
How do I take it back? Can you see every action in one place, reject it before it goes out, and revoke a permission in one click?

If you are comparing specific products, we keep side-by-side pages for OpenAI Dots, Meta Muse and Town, including where Prio is not the right choice.

Where Prio fits

Prio is a personal agent for one job: the work of a chief of staff. It works through your email, calendar and documents around the clock, in Outlook or Gmail. It answers, schedules, prepares your meetings and chases every follow-up. Every email, invite and reply with your name on it waits for your approval by default, and Prio only asks to do more once you have said yes to the same kind of work often enough. It is hosted in the EU, never trains on your data, and works in English, Dutch, German and French. The iPhone app is in public beta.

We don't think the specialist and the generalist are enemies. Many of our users also use ChatGPT or Claude, and we are opening Prio up so those assistants can call on it, without ever being able to send in your name. The big agents will be everywhere. The question is which one knows your work well enough to be trusted with your signature.

The personal agent has arrived. What to look for if you work in Europe