I'm starting a series on personal agents.
I use AI more than 99% of the world. I've tried them all — Muse, OpenClaw, Grokbot, etc.
Part 1: three weeks living with one. What works, what flops, and some of my predictions for the future. https://x.com/i/article/2108962752668524544
The Personal Agent Series: Part 1
I'm starting a series on personal agents.
I'm a serial technologist, deep in AI. I try everything — if there's a new agent out there, I've probably already broken it.
Right now I'm chasing one question: what does it actually feel like to live with a personal agent — and how close are we to everyone having one?
I personally think it's the biggest thing in consumer AI. And I want to take you behind the scenes while I explore it.
This series is hands-on. I'll be testing the latest personal agents, sharing tips and tricks, walking through real use cases, and telling you honestly what I find success with — and what flops.
This is my first post: three weeks living with one.
Some context on where I'm coming from: I use AI more than 99% of the world. I'm a power user of @Claude and @OpenAI, and I teach people how to use these technologies at all levels. I've led 75+ workshops in the past year, and I've seen people adopt and struggle with all kinds of technology advancements.
OpenClaw was first. It gave me the feeling (around 8 months ago). That magical “holy crap, it just did the thing” feeling.
Then my claw would die. Memory would vanish. And there was always this low-grade security anxiety humming underneath - will everything be hacked?
So I never fully committed. I punted on the personal agent dream and just used AI for work, where I already live in it 24/7.
Then @Grokbot showed up last month. And Grokbot is genuinely good. I actually love it.
The specialty agents are the killer feature, and being able to add them into a shared channel to chat with each other is amazing.
And let's also give Grokbot credit where it's due. They were the first to really nail the shared VM experience for your agents (a shared computer in the cloud for all of your agents to work on), one that I think the rest of the industry is chasing to copy right now.
But most normal humans are never setting up specialty agents. And Grokbot is more expensive than some of its competition (starts at $20 but really need to be around $100/mo to use it a ton). IMO, Grokbot is great for work. It's backed by a fantastic company that is moving fast. They're a main player and will continue to do well.
But the one that has me most interested in becoming my personal agent is @Muse.
Full disclosure: I didn't like it at first.
Week one, I did a feature tour. Impressive — tons of features, simple on the surface, but offered plenty of options for advanced users like myself to explore and customize.
They do a great job building "trust" with their UI, and offering permissions at every level.
But one agent? I hated that at first. I wanted multiple agents, like Grokbot. It genuinely bothered me in the beginning.
Then I started using the one agent for everything. His name is Carl. And Carl delegates to his side chats (sub agents).
That's when the vision snapped into focus. One agent for everything is simpler. It's the right model for a personal agent IMO. At least right now.
The part nobody tells you: the training period
Every personal agent needs 2-4 weeks of training. During that window, results are mediocre.
And because results are mediocre, you don't normally invest more time or data. Which keeps results mediocre. Adoption death spiral and so most won't reach this mark. Honestly, it makes sense given where agents are in their adoption period now.
I cheated. I had videos coming up and events on the calendar that focused on Muse, so I forced the adoption on myself. Best thing I could've done. It got me over the hump, and then I started seeing what a fully connected agent actually does.
That's where I am now.
My use cases are embarrassingly simple. My wife reminds me of this regularly
Every morning: Carl does weather check and based on that, suggests outfits for me to wear for the day (pulling from 40+ photos of real clothes in my closet)
Pro tip: take these photos the next time you're doing laundry and have to put them all away. Even low quality photos work.
Is the weather cool? Yes. Is the artifact it generates cool? Yes. Is an agent indexed on my clothing cool? Obviously.
Am I actually taking its fashion advice? ...No. Not yet. But the shape of the thing is what excites me.
Beyond this use case, think about the synergy that comes with an agent knowing what clothes you own. Now all of sudden, things such as "Packing lists" and "shopping for new clothes" becomes much easier, and a task that your agent can do for you.
But first, your agent needs more details. Your agent needs to know where you are, every step of the day. Get past the creepy part and this is where it gets good.
The boring use case that matters most: the calendar
Yesterday I was driving to Connecticut for a wedding. I'd requested the time off at work.
Mid-drive, Carl messages me (image below)
Honestly, it made me so happy. That's exactly the type of response I was looking for.
- Awareness of the new event
- Awareness that I couldn't attend it
- And suggesting the proper follow up for my approval.
That's exactly the type of behavior I'm looking for in a personal agent.
Then there's travel
I gave it my hotel info and the agenda. Before I even left, it handed me: things to do with the kids nearby, places to eat, the weather, pool hours to watch out for.
Three weeks ago, if you'd asked me about that wedding weekend, my answer would've been: “No clue. Haven't had time to think about it. I don't even know where we're staying. I'm just doing what my wife says."
Now? After one message I feel like a local! It's like having an executive assistant in my corner, and for the first time I'm feeling more ahead on these small, personal details than I would have previously. And I enjoy that!
It's also building personal software for me. And this is the biggest deal.
Here's the thing I keep coming back to. I'll say it plainly: this is the biggest deal for Muse so far.
Any idea you have, Muse can probably build it for you. Inside the app. Already connected to your data and everything it knows about you.
Think about Claude and OpenAI. Both let you create artifacts. But on the go, in the app, it doesn't feel like an extension of you.
Not with Muse. Because of the UX, it feels like you're building your own set of apps and tools inside the Muse ecosystem. I never played Minecraft. But if I did, this is what I imagine it felt like, to build your own game, within the game.
With a couple of prompts, Muse built me a "Watchlist" tracker that lives inside the app. It learned my taste, curates a weekly feed from theaters and the streaming services that I subscribe to, and I upvote/downvote to train it further on my preferences within the app.
It even builds its own skills. I had it create one for movie ticket booking — check calendar, confirm times, remember my seat preference, buy via the Link integration, add it to my calendar, ping me when it's time to leave.
Now all I need to do is click "buy ticket" and it confirms date / times with me and handles the rest.
Again, not a huge thing. But it adds up over time.
Oh, it's also turned into my personal DJ.
I gave Carl access to my SoundCloud, it studied my history, and built me a high-tempo work playlist. Then it refreshes it with new tracks every morning.
Update: I've since moved the whole music setup to @Spotify (@soundcloud started to block my bot attempts. Do better, Soundcloud).
Carl also follows me to new devices
Here's the part that got me even more excited.
Muse released their official SDK last week. So I put Carl on an ESP32 device. A $30 device with a built in touchscreen, mic, speaker, memory, etc. And I connected it to the OpenAI API, and gave him a voice.
The same agent I talk to in the app now lives in my pocket.
Some more pro tips:
- Configure the action button on your iphone so when pressed, it opens your Muse in dictation mode.
- Or, hit a shortkey on your keyboard, mine is "fn" and it opens dictation for your muse agent there too.
Either way, your agent is always one tap away now.
But what if Muse was also Operating System for your phone, computer, or Car?
Let's daydream for a second.
You open your phone. Your agent is waiting for you.
Not Siri. Your main agent. The one that knows everything about you. For me, that's Carl.
You talk to it and it controls your phone for you. Fully. Want a new app? You don't go to the App Store. It builds it for you.
You plug into CarPlay and your agent comes with you. Call it up for a running audio conversation. Turn on video mode and show it what you're seeing.
You chat. It does things. It reports back. Everything's connected.
That world? That's the world I want to live in. And it doesn't feel distant anymore.
Put yourself in Meta's shoes for a second. They have their sunglasses. It's only a matter of time until Muse lives in them. They're shipping SDKs for every kind of device and widget you can think of. Within a year or two, your agent moves across all of it, following you throughout the day.
Here's a fun visual: you unlock your phone and instead of a static wallpaper, your agent is there. Moving around in the background. Working on what you gave it. It welcomes you back with "How can I help you?"
There is no endless list of apps. Just a helpful agent ready to be your 24/7 companion in life.
Honestly, we're not that far away.
The trust question
I know you're probably all rolling your eyes about now. So lets talk about it.
Security and giving all your data to a personal agent company like Meta, Xai, or Instinct.
The truth is, in order to get results that you're actually happy with, you need to give whatever platform you pick, a lot of your data.
So who do you trust? A startup that launched within the last year like @instinct? Or a company that's been around 15+ years, hardened through election cycles and attack after attack like @meta?
I'm picking Meta for now.
And look, this will always be sensitive. I'm not trying to convince you of anything. You might say: I'm not doing personal agents, I don't want to hand over my data like that. Totally fine.
Or, I hate Meta and I'm not giving them more of my data. Fine as well.
But before you make your decision, think about the internet ten years ago.
If you were around for that moment of time back then, you probably heard a lot of people discussing “Are you going to use the internet too?” / "are we really going to upload our business data there?"
All of which sounds hilarious now that we've seen how things have played out.
That's where I think we are with AI and personal agents at the moment.
For me? I'm too interested in this technology not to give it a try.
Pro tip: toggle of "help improve our models" so they can't train on your data, add your secrets to the "secure store" and review all of the permissions you give to your agent.
And let's be real. Meta already knows everything about me. I mine as well benefit from it.
And did I mention that Muse is free?
So I'm going to keep giving it a real shot for now. After all, I think this is the future we will all be living in soon.
So where does this land?
Muse is my favorite right now for my personal life. Grokbot is really good for work related tasks that need automation, specialistic, and simplest agents. Claude and OpenAi still dominate my presence at work, specifically for knowledge and engineering work.
But how impactful is Muse really in my daily life?
After 10+ hours of onboarding, I'd say I'm 30-40% of the way there.
I mean that as a compliment. It's pretty damn good. I only score it that low because I can see how far this goes, and how early I am.
But really, no matter what platform you're using — they're all good, and it's great that we have options.
But here's the thing you need to know: to get the results you're looking for, you have to give your agent a lot of information about yourself. A lot.
If you're okay with that, pick the agent you feel comfortable with. If you're not, it might not be worth your time until you are. The results won't come until it knows everything about you. And then suddenly you have an executive assistant living each day with you.
One more honest note: I've never had an agent stick with me for more than a month, for which I'm approaching now.
OpenClaw felt like this early on but died two weeks in and after multiple attempts at resurrection, it never stuck. Grokbot I still use, but it moved to work stuff and is no longer my personal agent.
But there's something different about this one. It feels like a real personal companion.
We'll see. I'll report back in the next post.
Follow me for more things on personal agents, using AI in your work, and generally, for the vibes 😎
Additional Muse Resources:
2) Getting started with Muse guide: https://tenex-labs.github.io/muse-personal-agent-start-kit/
3) Setting up Muse on your own device: https://github.com/JJenglert1/esp32-muse-agent
4) Subscribe to the #1 AI newsletter: https://www.tenex.co/ultrathink