Last Thursday OpenAI shipped an Apple Messages plugin for the Mac ChatGPT app
Last Thursday OpenAI shipped an Apple Messages plugin for the Mac ChatGPT app. It can search your iMessages and SMS, summarise them, draft replies, and, once you approve, send them. Free tiers included.
It didn't surprise me, because my own machine already works that way. Every morning an agent reads my email and sorts it: this needs you, this one I've handled.
The unnerving part wasn't day one. It was day three.
On day one you check everything it did. By day three you don't. You start trusting it, and then you stop noticing what it is reading. The shift happens faster than you can observe it happening, and by then it has stopped being a technical question.
So when I teach this, the longest part isn't wiring up the tools. It's one line:
Reading is not sending.
Letting it read is convenience. Letting it send is letting it speak as you. Those deserve separate approvals, and the sending one should never come with a "don't ask me again" box.
You can trust an agent very far. The trust just has to be something you chose, rather than something you drifted into.
Would you let an AI read your messages? Or read yes, send no?
