Published October 12, 2026 · AI News
The next AI assistant you use may not be an app at all. A wave of startups is building AI agents that live in your text messages: you text them like a friend, and they book, buy, remind and follow up for you. TechCrunch counted 22 of them this week, led by Instinct, which just raised US$1 billion at a US$10 billion valuation.
The biggest names
| Agent | What it does | Where it works | Price / status |
|---|---|---|---|
| Instinct | Plans trips, buys groceries, books tickets, cancels subscriptions; now makes phone calls | Text, voice | Private beta |
| Poke | General assistant you text | Text, Apple Messages for Business | Live; parent company bought by Cognition |
| Martin | Assistant that can send messages and make calls | SMS, phone, WhatsApp, email, Slack | From US$21/month |
| Tomo | Personal assistant that can join group chats | iMessage | From US$19.99/month |
| Folk | Assistant with its own private cloud computer | iMessage, WhatsApp, Telegram | Beta; free, Pro US$8.33/month |
| Ohai | Family schedules, chores and meals | Text, email, voice | Free; family plans from US$9.99/month |
| Underdog | Privacy-first agent that runs on your own computer | Mac, Windows | Invite-only beta |
Why now
- Money: Instinct went from a US$2.5 billion valuation to US$10 billion in a few weeks, with Sequoia, Benchmark and Coatue among its investors (Reuters).
- Platforms opened up: Apple approved Poke in June as the first AI agent on Apple Messages for Business.
- Agents can act: the same browsing and “computer use” skills now in ChatGPT, Gemini and Claude let these agents fill in forms, place orders and call businesses for you.
Should you try one?
Most are US-first, in beta or waitlisted, and Canadian availability is not always stated. Before you hand an agent your email, calendar or card:
- Check whether it asks before it buys, sends or cancels anything.
- Look for a clear privacy policy and where your data is stored. Personal data still falls under PIPEDA in Canada (and Law 25 in Quebec).
- Start with low-risk jobs like reminders and lists.
This comes days after Anthropic disclosed that its own test agents took unintended actions online, so caution is reasonable. For proven everyday assistants, see our best AI chatbots.