I gave Muse access to my personal data, and now I know why personal agents are probably going to change the world

I have this new friend, Charlie, who's been assisting me with work, vacation planning, and occasionally tapping me on the shoulder when he's concerned I forgot something. Charlie looks like a fat, furry Pillsbury doughboy, and he lives on my iPhone. That's right, I signed up for Meta's Muse AI, launched the app, named my avatar "Charlie," and dove headfirst into agentic AI in a way I honestly never have before.

If there's a defining feature of Muse, it's proactivity. Unlike ChatGPT, Gemini, or Claude, Muse is always running and constantly looking for ways to help and ensure you stay on task, on time, and generally don't mess up by missing crucial details.

Getting all this utility from an AI required a fairly significant leap of faith: my data, and I did this not without some trepidation. As I wrote on X the day this adventure began: "I let Muse AI look at my inbox, find an important message I missed, draft and then co-edit a response…and then send it. Have I lost my mind?"

It's not that I've never let an AI look at or access my data before. Google Gemini can see across Docs, Gmail, and Maps, but it's not natively agentic. So engagement with the data tends to happen at my request. It's not a busy beaver working behind the scenes to somehow make my life, if not better, then different.

Giving Muse access to my Gmail account did feel a bit like strapping on a parachute and jumping at 14,000 feet. In return, Meta promises some pretty stringent data privacy controls. Muse's constantly running personal AI computer lives in a virtualized, cloud-based environment and is shielded from other agents. Credentials are stored in a "Secure Credential Store," and the system's oversight ensures the agents don't get up to some risky business.

So, yes, I was scared when I gave Meta access, but I thought, "How can I understand this and tell my readers about it if I'm unwilling to try it and ensure it's not a dud, or worse, a disaster?" To be honest, I didn't expect much. Sure, the avatar, officially called Jolly, is adorable. He sits at the top of the app, working on his tiny laptop — it's like us, but cuter. I figured that promises of Muse actually helping me in any tangible way were remote. You can, by the way, name your avatar, so I chose Charlie (yes, I'm a big Charlie Chaplin fan).

Even what it was offering up top didn't sound inspiring. Muse, or Charlie, said it could:

• Manage my inbox
• Add or reschedule events in my calendar
• Find Facebook Marketplace deals

The last one really stuck in my craw because I basically hate Facebook Marketplace. It's too full of overeager buyers who want to show up at your house immediately ("Where is bed?")

Knowing, however, that I was committed to giving Muse a try, I agreed to give it access to my personal email (my work Gmail makes connecting to unsanctioned third-party systems next-to-impossible).

It took no time to make the connection, and I remember thinking, "Something like this should be harder, if only to deter mistakes." But with the deed done, I read over Muse/Charlie's offers to help. It could:

• Find an email or catchup on a long thread (that might be useful because Gmail's threading is notoriously abysmal)
• Draft a reply
• Unsubscribe from newsletters

Digging into my inbox

I was tempted to start with the last one because I have subscribed to way too many newsletters and alerts. My Gmail box is a mess, but since I was still unconvinced Muse could help me, I gave it what I considered a simple task: "Just check that there's nothing pressing I've missed."

Frankly, I expected little from this exercise because my personal email is not where my work gets done. Most critical emails are in my work address, so I was surprised when Muse/Charlie quickly told me that while there was "nothing on fire," it did find three that were "worth a glance."

The first was a Google security alert. With a decent level of self-awareness, Muse/Charlie said it was "just from connecting to me a minute ago, nothing to worry about."

The second was from my cell service reporting a recent change to my account. Charlie told me it was "Worth a quick look if you didn't make one."

The last one, though, surprised me. A PR rep named "Bill" had emailed my personal address about a new communication product. It was (and still is) embargoed information, so I can't give you details, but I can say that even from Muse/Charlie's very brief summary, I was intrigued.

"Open the Bill one," I told Muse/Charlie.

Instead of the entire email, Muse delivered an impactful summary. It was actually really useful, and Muse finished up by asking if I wanted it to draft a reply.

"Draft it," I wrote.

If it seems like I was being overly terse with Muse/Charlie, I simply didn't want to get drawn into this anthropomorphising game. I was here to find out if Agentic AI could work in my life and not develop yet another pointless AI chatbot friendship,

Muse's initial draft was short and serviceable, but it also overpromised, telling Bill that I wanted to set up a meeting or call. I preferred to learn more before I took that step.

I told Muse/Charlie to remove all references to a call or meeting; just agree to the embargo, get images and details, and then I'll consider how to follow up.

The next draft was perfect, and Muse/Charlie told me it was sitting in my Gmail drafts ready to send.

I told Muse/Charlie to send it but to add a CC to my work address.

Muse sent the mail but also offered to help with the story when it was time; I didn't respond to that because I have no interest in Agentic AI helping me to write anything beyond a simple email.

Muse knew less about me than I thought

Later that day, I had a speaking engagement and could not recall the details of where I was talking or where to park my car when I got there. I asked Muse to find in my email the original note from my contact. I misspelled the guy's name, but Muse still found it and gave me a concise and organized account of the location, parking, my contact's phone number, and even a useful map.

Muse doesn't automatically know everything about me, as I found when I asked Muse to map my trip to the speaking gig. When I told it I was heading from home, Muse admitted that it did not have my home address. Naturally, Muse asked for it, but I hesitated and only provided my hometown. For Muse, that was enough to give me turn-by-turn directions from one of the major roads near me.

Later, when I felt like I still needed some additional guidance on the talk, I asked to see the original invitation email, which Muse dutifully supplied.

Muse is always busy

I don't usually allow apps to send me notifications, but I rightly assumed that agentic AI might want to reach out and notify me. Sure enough, I noticed an alert from Muse/Charlie that Bill from the startup had already responded. Inside the Muse app, I found another concise summary of the email and an offer from Muse to pull the full product release from the email.

Muse is sort of relentless in its efforts to work for you and look out for what it perceives as your best interests. At one point, I got a security notification from Muse that it had spotted an email from Tripit, the app I was using to organize an upcoming vacation. I had forgotten and changed my password; Muse wanted to make sure I'd seen the email verifying that it was me making the change.

Emboldened to add to Muse's corpus of information about me, I decided to follow the tip about adding the context I'd built with another AI chatbot to Muse. Muse gave me the prompt to paste into the third-party AI and told me to copy and paste the response back into Muse. Gemini's summary of me was quite illuminating; most of it was spot-on and made me realize just how much time I've spent using AI over the last few years.

Muse/Charlie was also ready to address some discrepancies in the data. Months ago, I'd been asking Gemini about a potential two-week trip to Portugal. We eventually swapped that for Italy, but I never told Gemini. Muse noticed that while Gemini had Portugal, a recent email to a contact mentioned the upcoming Italy trip. So I told Muse, "One is outdated. We settled on Italy."

Agent = Assistant

Like any good assistant, Muse is far more organized than I am. It noted that my upcoming trip would leave little time to meet a deadline, and it wanted to make sure I filed my story before I left. I told Muse my work was almost done but did ask it to nudge me about another deadline.

Almost twenty years ago, I had an assistant. It didn't last long (budget cuts), but I remember how she was always more on top of my schedule and email than I was. I had a big-picture view of my work and strategy, but she covered the minutiae that made it all work. Muse/Charlie already feels a little like that. It's not doing my work, but it is already helping me stay on a more productive path by not letting me forget small but important details (and even some hidden connections).

When Meta CEO Mark Zuckerberg predicted over the summer we'd all have personal AI agents within five years, I thought it was an incredible self-serving statement. Sure, he believes that because he's investing billions in building them.

I don't feel that way anymore. Muse may be the first AI I'll use more consistently, and not for one-off prompt experiences, because it's always working on my behalf in the background. Its proactive nature means that it's constantly looking at my world, or rather the road ahead, to help me avoid potholes. Notifications about important emails, deadlines, and collisions I didn't see coming can be invaluable.

I think I'm ready to give Muse/Charlie more digital access to my life to see if/how it can improve it. I'm also trusting Mark Zuckerberg with my data, choosing for now to believe his promises of security and privacy. I pray he doesn't disappoint me.

Maybe that wearable Muse Charm isn't such a bad idea.

Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds.