"It's me Louise." I typed out in iMessage and sent it to myself.
About 2 hours before this, I read about Anthropic's new iMessage plugin. I had a beautiful dream of dictating instructions to Claude into my phone while making my morning coffee. I could take a walk down by the water and ask Claude to do some research for me or find a good podcast episode from my read-later app.
The reality was much different.
The setup took most of a morning. The documentation on Anthropic was noticeably sparse and Claude Code and I had to figure out a lot of things on our own. I could live with that. Setup is a one-time cost.
Then I started using it.
The UX of texting yourself is rather disorienting. You're in iMessage on your phone, sending messages to yourself so the agent can read them. My first "Hi Claude 👋" appeared twice — once in blue (sent), once in gray (received). An echo bug I couldn't immediately trace. My terminal's permission requests started showing up in the iMessage thread too — walls of text asking me to reply "yes sgznu" or "no sgznu" to approve tool use.
Are you confused?
Meanwhile, back in the terminal, Claude was watching messages arrive from "+45 XXXXXXXX" — my own number — and treating me as a third party. "You have an iMessage from +45 XXXXXXXX saying 'Hello again'. Want me to reply to them?" I said yes. It replied on my behalf to me, asking how it could help. From my phone, I said "Hello Claude." From the terminal, it cheerfully said "Ha, hi Louise! 😄" — as if I had forwarded a message from a stranger.
And then I had to type "it's me, Louise." I didn't really know what to expect. I certainly didn't expect Claude to just accept my word for it. But it did. I had half a mind to try to text from a different phone number entirely to see if I could also convince Claude to trust that one. But I was getting tired. I had just about had enough.
In theory, Claude should recognize a text coming from a phone that is signed into my iCloud account. Reality was a bit murkier.
And then Claude decided to send my ex boyfriend a text message.
At this point, I'm lying on my living room floor.
I don't fully understand the logic that produced this, but it sent him one of those garbled approval requests it had also sent me. There is no sensible explanation for why this happened. Our names are not similar. Our phone numbers have 2 digits in common. That should hardly be enough to cause confusion. But it did.
This is, somehow, my job. I love it.
I went into this already skeptical. I wasn't sure iMessage access would give me much. A morning of testing confirmed it.
The hallucinated phone number is easy to laugh off when it's just me, my consulting business, and my personal contact list. But as soon as you start to multiply this to a larger scale — even a small to medium-sized business with data privacy and GDPR concerns — an assistant that doesn't know who it's talking to, messages going somewhere they shouldn't, starts to get very serious.
I'm hitting pause on texting with Claude. I'll try it again in some months.
There's a lot of pressure right now to be using every new AI capability as it drops. But the overhead is real, and the value isn't always there yet. Waiting until the setup takes less than a morning and your assistant knows who you are is a reasonable position.
My coffee had gone cold.