The best interface for an AI agent might be the one you already use forty times a day. OpenAI seems to think so. The company has started rolling out texting for Dot, its personal agent, and it’s a smarter move than it first sounds.
Dot debuted late last month with a promise that users would eventually be able to text it. That preview has now arrived for US Pro subscribers. You can send your Dot text, photos and voice memos, and it can message you first with updates on tasks you’ve given it. That last part matters most. An agent that waits for you to open an app is a chatbot. One that pings you when the work is done is closer to an assistant.
Setup is simple. Open the Dot’s profile in ChatGPT, choose Texting, and make sure iMessage is on if you use an iPhone, or RCS if you’re on Android. OpenAI says it worked with Linq for the messaging layer and ran extensive penetration testing. Its pitch on X is blunt: only you can text your Dot, and your Dot can only text you. Memory is shared between ChatGPT and text, so context follows you from one conversation to the next.
The security claims deserve scrutiny. Putting an agent with memory and task access into a messaging channel widens the attack surface, and phone-based identity has a messy history. Pen testing is reassuring, but it isn’t proof. Developers and founders weighing this for real work should watch for independent reports.
The timing is the bigger story. The personal agent category got crowded fast after Anthropic joined last week with Claude Duet. In the Engadget writer’s testing, Dots feel more polished. Duet is still behind a US-only waitlist, though Anthropic plans to offer it to subscribers of its $20 per month Pro plan.
That’s where OpenAI has a gap of its own. Dots are not available to Plus users, so the cheaper tier is locked out for now. If Anthropic delivers Duet at $20, price becomes a real argument against OpenAI’s Pro-only approach.
So does this matter? Yes, modestly. Texting alone isn’t a breakthrough, since Google, Apple and plenty of startups have tried assistants in chat threads. But distribution is the fight now, and messaging is the most direct route to daily habit. Whoever makes an agent feel like a contact in your phone, rather than a tab in your browser, gets the stickier product. OpenAI just took a small, sensible lead on that front. Expect Anthropic to answer quickly.
