I Tried Automating My Relationships Using AI Agents. Muse and Dots Overstepped

Most people revel in milestones. Turning the yearly page on birthdays, holidays and anniversaries is an opportunity for big parties, exchanging nice gifts and celebrating the ones we love. It can also be a minefield of missed connections and hurt feelings if you forget the dates.

If you’re ever sent a “happy belated birthday” message, neglected to recall your own anniversary or simply struggle to keep in touch with people, this guide might be for you. Granted, you’ll have to put your trust in an AI agent to do the heavy lifting, but you might never need to remember to send a card again.

While AI may not understand human connection and happiness, what it does well with is rote tasks. I tried to use ChatGPT and a few other AI tools to automate the upkeep of my personal and professional relationships — but here’s where it all went wrong.

More from CNET

(Disclosure: Ziff Davis, CNET’s parent company, in 2025 filed a lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.)

‘Happy belated birthday,’ no more?

The penalties for being terrible at remembering birthdays increase as one gets older. Soon it’s not just your friend’s birthdays that slip your mind, but also their kids’ birthdays and your grandkids’ birthdays and on and on as the wheel of reproduction turns. 

I used ChatGPT — which appeared to be the only AI tool among the three I tested that could automate this entire process for me — to identify key birthdays of friends and family and develop an automation system to make sure I never miss another important date. 

Claude was able to access and draft emails as well as create birthday cards and greetings, but not automate the sending.

Claude/Screenshot by CNET

Gemini seemed like it could only work with me in the chat to draft emails and generate custom cards, not actually work inside my inbox. Only some people have access to Google’s AI agent, Gemini Spark, right now.

Gemini/Screenshot by CNET

I authorized ChatGPT and its agentic offering, Dots, to look through my Google Calendar and Gmail to find and match dates to the birthdays of the people most important to me. You can do this by navigating to your profile icon in ChatGPT, clicking Settings and then opening Plugins.

ChatGPT/Screenshot by CNET

I also gave ChatGPT authorization to read, compose and send emails on my behalf. 

Then I asked ChatGPT to look through my calendar and Gmail and find as many important birthdays as it could.

ChatGPT was able to use my inbox to boil down my friends’ and family’s interests to a broad, yet specific, enough set of data points to create custom digital birthday cards that actually fit all their personalities. Even the one friend ChatGPT know much about still got a digital card that fit her pretty much to a T.

ChatGPT/Screenshot by CNET

Automating all my love

Now comes the most crucial element — setting up ChatGPT to automatically send digital birthday cards and greetings to all the important people in my life so I don’t screw it up. I tested this feature by asking ChatGPT to generate a digital card and message for my dog’s birthday and send the email to my husband.

ChatGPT/Screenshot by CNET

The chat limit was reached here, and ChatGPT also surfaced an alert that asked for my permission to send an email on my behalf. And while it can be good in general for an AI agent to ask for your permission before sending an email, I had tasked it with sending those specific emails by itself automatically each time.

Beyond my email task, I also asked Meta Muse, the social media giant’s AI agent, to send memes and check-ins to a few of my friends on Instagram. This ended up being more trouble than it was worth; the content Muse selected to send along was based on seemingly nothing specific, just random posts sourced from around the platform that might be construed as mildly amusing. 

The bottom line on AI agents overstepping

This is where things get a little fuzzy. The amount of work it took to optimize Dots and Muse so they would send relevant, plausibly functional correspondence was not worth it. If it takes more time to review, revise and approve a simple email or DM than it would for me to do it myself, the whole prospect and promise of agentic AI falls flat.  

Ultimately, you should also be aware that once you’re working with an AI agent, it may also start to overstep in the chat. I had to tell both Muse and ChatGPT’s Dots to back off — because while I’d given them permission to peruse my emails and social media accounts, I had only asked them to complete certain, very specific, actions while doing it.

ChatGPT/Meta/Screenshot by CNET

Dot took it upon itself, totally unprompted, to alert me to a timing snag in my evening, going through my emails and letting me know a movie I’d bought tickets for conflicted with my Japanese class that night. This might have been useful, had I not already sent an email to my teacher right after buying the tickets, letting them know I’d be missing class.

Muse started sending me unsolicited updates about the weather and a fire in Malibu. Both undertook relentless efforts to get me to use more and more of their features while I had the apps active. 

Overall, the experience didn’t leave me feeling more connected to anything but the AI companies.