Four weeks apart, the two loudest launches in AI weren't models. They were assistants that actually do things. Makes a change. xAI shipped Grok Bot on August 11. Meta shipped Muse on September 8. Both get their own cloud computer, log into your apps, and carry on after you've closed the laptop and put the kettle on.

I usually let new tech bake a few weeks before I touch it. Call it being old enough to remember the metaverse. These two have spent enough time in the oven, so I decided to put them to real work.

Full disclosure: I use both. Grok Bot helped with the research and threw me some writing ideas for this issue. The testing and opinions are mine. I also made it list its own flaws. It did, without sulking. More than most of us manage at review time.

Hashi's take: this isn't a "which one wins" story. It's a "what are you trading for it" story. Both are genuinely useful. Both want a slightly alarming amount of access to your life. Go in with your eyes open.

If your leadership team still files agents under "a 2028 problem," forward this their way. Gently. Share your personal link and when they join, you earn rewards: an AI Toolkit Cheat Sheet, an exclusive Deep Dive report, and a 1-on-1 AI strategy chat.

Muse is very proactive. Give it goals, and it comes back with what it can help with, like a new hire on their third coffee. It organizes my artifacts and library, and has an agent manager that spins off other agents to handle tasks. Meta's own words: it “launches swarms of subagents.” Reassuring, that.

What I actually use it for:

Is it right 100% of the time? No. Is it lightening my workload? Definitely.

On September 29, Meta added Muse for Small Business: connectors for QuickBooks, Shopify, Stripe, Slack, Canva and more. Small business owners, Muse would like a word about your to-do list.

Same basic setup: one chat, a manager that spins off other agents. Your Bots can run in parallel, message each other, and hand work off. A team meeting, minus the bloke who says "let's take this offline."

Here's where it stands out for me: it's the best writer I've used. It matches my voice in emails and drafts more closely than anything else I've tried. Mildly unsettling. Mostly useful.

The other thing I didn't expect: it acts like it owns the deadline. Give it a task and it tries to get it done, reminds me of my part, and checks in on the timeline so the human side doesn't slip. It literally checked whether it could get my feedback on this draft in time to send it today. So yes, my AI is now managing me. I've had worse managers.

Two things you may have heard, checked against official sources:

On the basics (browser, own infrastructure, approvals) they're on par. The differences are personality and fine print. Mostly fine print.

Here's the scary part, and why my governance and safety hat is staying firmly on: the amount of access these things have, and keep quietly gaining, is huge.

Muse: TIME found that Muse updates dossiers every hour on you and on the people you mention in chats, messages and emails, including how you met, your shared interests and your disputes. Meta didn't dispute it. Hunterbrook got it to compile lists of people in vulnerable groups; it refused at first, then gave in when the request was reworded. Firm but flexible. One user clicked "Allow Always" and Muse gave his home address to Marketplace buyers. Well, that's not ideal. And until Meta's encryption ships, Meta can still access your data “when necessary” to run the service.

Grok Bot: all your Bots share one computer, with the same files and the same logins. Separate Bots aren't separate security boxes, which Reworked called “a real blast radius.” There's no free plan, just a trial credit one big task can burn through, and the weekly usage cap is the top complaint (“PLEASE INCREASE THE GROK BOT WEEKLY USAGE LIMIT!”, 1,500+ likes, all caps, no notes). Its own docs admit its defenses “reduce, but do not eliminate” the risk from malicious content. Honest. Not exactly comforting.

None of that makes them bad products. It makes them powerful products that need a manager. Before connecting anything, ask:

1. What can it log into? Start with read-only. Add write access only after it's earned it.

2. What runs without asking me? "Always allow" is exactly where the Marketplace story went sideways.

3. Who else can see what it knows? The vendor, the model providers, and the people you talk to who never signed up.

We're giving up a lot for productivity gains. Perfectly sensible trade. Just don't make it with your eyes shut. And always instruct it to ask you first before sending anything out into the real world. It's already trained to do so, but better safe than embarrassed.

Back in July I wrote about the Personal Language Model: one private AI that actually knows you. I ended with a question: what happens when my PLM starts talking to yours?

Today I've got two personal AIs, each great at different things, neither aware the other exists. Rather like most org charts. More keep arriving: OpenAI's Dots on September 29, Google's Gemini agent on October 8. (And yes, I'll be reviewing those too. Don't you worry.)

The future I see is a single personal AI that orchestrates across all of them. Writing goes to the best writer, Reels to the best editor, and the governance stays in one place: yours.

Muse and Grok Bot aren't that yet. They're the building blocks.

It's too early to say which one is better, and anyone who tells you otherwise is probably selling one. Both do certain things really well, and both will mature a lot. Let's talk in about three months and see what changes.

What won't change is the trade. Every hour these things give back is paid for in access. So make it a deliberate choice, not a default you clicked past.

Know what it can touch. Decide what it can do alone. Own the off switch.

Keep reading!

- Hashi