The personal AI assistant race
A year ago, an AI assistant told you what to do next. This autumn, it does it for you.
Between May and September, Google, xAI, Meta, OpenAI and a startup most people had never heard of all shipped the same idea: an assistant that takes a goal, works on it in the background, and comes back when it's done. It books the restaurant, cancels the subscription you forgot about, chases the email you never answered.
This is how that happened in five months, how the five assistants compare, and what it all looks like from Portugal, where most of them haven't arrived yet.
May: Google goes first
The race started at Google I/O on 19 May, when Google announced Gemini Spark, a personal agent that runs in Google's cloud and keeps working after you close your laptop.
Spark's pitch was integration. It lives across Gmail, Calendar, Docs, Sheets and Drive, so it can clean up your inbox, turn meeting notes into a report or untangle two clashing meetings without you opening any of them. When something sensitive comes up, like a payment, it hands control back to you.
Google made another move on the same stage. It cut its top AI plan from $250 to $100 a month, and that is where Spark launched. An agent this ambitious needed a price people would actually pay. By July it had reached the $19.99 Pro plan in the US.
August: a startup makes it personal
Then a 23-year-old founder showed the giants what people actually wanted.
Instinct didn't ship an app. You text it, or you call it. It plugs into your email, messages and calendar, and early users said it planned road trips, ordered groceries and cancelled subscriptions they had forgotten they were paying for. What set it apart was initiative. It picked up dropped threads and messaged you first, the way a good assistant would.
It launched invite-only and became the most talked-about AI product of the summer. Weeks later, Instinct raised $1 billion at a $10 billion valuation, about four times what it was worth a month before.
The same month, xAI put Grok Bot into early beta. Grok had been a chatbot wired into X. Grok Bot is pitched as a teammate that signs into your tools and hands back finished work.
September: Meta and OpenAI arrive
On 8 September, Meta launched Muse. Mark Zuckerberg called it the beginning of "personal superintelligence", a phrase that, much like AGI, means whatever the seller needs it to mean.
Muse is the most consumer of the group. It runs on its own private virtual computer, works inside WhatsApp as well as its own app, and pays for things with single-use card numbers so it never sees your real one. It's free, with $20 and $100 tiers for heavier use. In six days it passed 900,000 downloads and topped the US app charts.
Three weeks later, at its DevDay conference, OpenAI answered with Dots. Where Muse goes after life admin, Dots goes after work. Each Dot gets its own cloud computer, connects to thousands of apps and can be messaged from Slack or Microsoft Teams. OpenAI talks about companies running "specialist Dots" with their own credentials and roles. To keep things safe, a Dot only uses read-only tools unless you allow more. The first one comes with ChatGPT plans starting at $100 a month.
By the end of September, every major lab had an assistant that does the work instead of describing it. Whether all of them deserve the word "agent" is a debate of its own.
Why everyone moved at once
Five launches in five months is not a coincidence. Chatbots got good enough, and similar enough, that answering questions stopped being a way to win. The next prize is the layer between people and every app they use. Whoever owns the assistant you delegate to owns your day.
That explains the money. A billion dollars into a startup barely out of beta. A price cut of more than half on Google's top plan. Meta putting its agent inside WhatsApp, where its users already are.
It is still a bet, though. None of these companies has shared revenue or retention numbers for its assistant, and handing an AI your inbox, your cards and your calendar is a big ask. Instinct learned that fast when its terms, which allow it to use what you share for model training, became a story in its first week. With assistants this capable, trust is the real product. The direction looks right. The timing is unproven.
Side by side
Same job description, five very different products. Here is how they compare as of October 2026.
| Assistant | Maker | Where you talk to it | Best at | Price (USD/month) | Available in Portugal |
|---|---|---|---|---|---|
| Gemini Spark | Gemini app, Gmail, Workspace, Chrome | Inbox, calendar and docs inside Google | $19.99 (US) or $99.99 (elsewhere) | No | |
| Instinct | Instinct | Text messages and phone calls | Proactive follow-ups, calling businesses | Free during beta | Yes, invite-only |
| Grok Bot | xAI | Inside the tools it signs into | Tasks across your existing software | $30 SuperGrok or $100 SuperGrok Plus | Yes, early beta |
| Muse | Meta | Muse app, web, WhatsApp | Life admin: bookings, shopping, bills | Free, $20 or $100 | No |
| Dots | OpenAI | ChatGPT, Slack, Teams | Long-running work projects | From $100 | No, not on Pro plans |
Read across the rows and two bets appear. Instinct and Muse go where you already chat, and win on feeling personal. Spark, Dots and Grok Bot go where you already work, and win on reaching into your files and tools.
One column, though, decides more than the rest for anyone reading this from Portugal.
Meanwhile, in Europe
Go back through the story and look at where it happened. Almost all of it took place on the other side of the Atlantic.
In July, Google opened Spark to more than 160 countries and left out the European Economic Area, Switzerland and the UK. It gave no reason, though most reporting points to the extra requirements the AI Act and GDPR place on software that handles personal data on its own. OpenAI did the same with Dots on its Pro plans. Muse is still limited to the US and Canada.
So here is what a person in Porto or Lisbon can actually use today: Instinct, if they get an invite, and Grok Bot, in early beta. The three biggest launches of the year are not on the list.
That has a cost. European users and companies are missing the first months of learning what to delegate, what breaks and what people actually trust. But it has an upside too. Europe gets to watch the problems surface first. Instinct added a data deletion tool after the criticism of its launch week. When these assistants do arrive here, they will arrive with more of those answers already built in.
What it means for companies here
The wait won't last forever. And the day these assistants land in Portugal, some of your team will connect one to their work email the same afternoon.
That is the part every launch post skips. All five are built for one person, not for a company. They run on the vendor's terms, so retention, training rights and where your data lives are decided on someone else's legal page. They belong to an individual, and Spark even requires a personal Google account, so when that person leaves, the context leaves with them. And they don't know your business: your pricing rules, your support policies, your ERP.
None of this is a reason to ignore them. It's a reason to decide now what you would delegate, to whom, and under which rules your data gets touched, before the decision gets made for you.
An assistant your company owns
While the big labs race to own each person's assistant, a quieter category is growing next to it: assistants built around one company instead of one person.
That's why we built Ardaven. It does a lot of what the assistants in this story do, but on your terms. It connects to the tools your business already runs on, like WhatsApp, Google and Moloni, answers from your own documents and policies, and is built for GDPR and the EU AI Act from day one.
You can run it internally for your team or put your own brand on it for your customers. And unlike most of this list, it's available in Portugal today.
The assistant that does the work is here
A year ago, assistants told you what to do. Now they do it, and the biggest companies in tech are betting that this is how we'll all use software. For Europe, the question isn't whether these assistants arrive. It's whether your company knows what it wants them to do when they do.
Use the head start. Pick the three tasks your team would hand off first, check what data each one touches, and decide whose rules it should run under. If you want help mapping that out, talk to us.
