Four AI Agents That Want Your Phone Number

Texting an AI agent is the most convenient interface anyone has shipped for this technology, and that convenience is exactly why you should read the pricing page twice before handing over your number.

I have spent enough time with assistant tools to know that the install step is where most of them die. You download the app, you poke at it for a week, and then it drifts to the third page of your home screen next to the airline app you used once. Agents that live in SMS or iMessage skip that failure mode entirely. There is no app to open, no new habit to build, no notification settings to configure. You text them the way you text a friend who happens to be good at logistics.

That is a real advantage, not a marketing line. It also means the usual way of evaluating a tool does not apply. You cannot kick the tires in a sandbox. The product arrives inside the most personal channel on your phone, mixed in with messages from your family and your doctor’s office. So the questions that matter shift from “what can it do” to “what is it reading, who approved it, and what does it cost when I actually use it.”

Poke has the credential that is hardest to fake

Poke is a general-purpose assistant you reach by text, and in June 2026 it became the first AI agent approved on Apple’s Messages for Business platform. I care about that more than any feature list. Apple’s approval process is a gate, and being first through it means someone outside the company looked at the thing and signed off. For a product class where the main risk is an unvetted agent sitting inside your message history, third-party review is worth more than a demo video.

Poke was later acquired by Cognition. Acquisitions cut both ways for users. More engineering resources behind the product is good. Roadmap changes, pricing changes, and shifting priorities after a deal are the standard risk, and I would watch for them rather than assume continuity.

Ollie takes the compliance question seriously

Ollie is aimed at families, and it has SOC 2 compliance. If you are routing household coordination through an assistant, that audit matters. Family logistics means kids’ schedules, addresses, and routines. SOC 2 is not a guarantee of anything, but it tells you the company submitted to an outside auditor, which is more than most consumer AI products can claim.

The pricing is where I would slow down. Ollie has a free tier, and paid plans start at $25 per month for 150 messages. Do that math against how you actually text. One hundred fifty messages is not a lot if the assistant is handling back-and-forth clarification, which conversational agents tend to generate. A single scheduling task can eat five or six messages. If you are a heavy user, the per-message economics are the thing to model before you subscribe, not after.

Orbits is priced like it wants volume

Orbits runs a free beta and a Pro subscription at $8.33 per month for unlimited tasks. That is a different philosophy from message-metered pricing, and for most people it is the friendlier structure. Unlimited means you can use the tool badly, waste messages, and learn its quirks without watching a counter. Metered plans punish exploration, which is the exact phase where you figure out whether a tool fits your life.

The caveat is the word beta. Free beta pricing is a snapshot, not a promise, and I have reviewed enough tools to know that the number on the page during a beta is rarely the number a year later. Treat $8.33 as today’s price.

Instinct is a funding story first

Instinct raised $1 billion in September 2026 at a $10 billion valuation. Those are enormous numbers for a company in this category, and they tell you investors believe texting is the winning interface for agents. They tell you almost nothing about whether the product is good. Capital buys runway, hiring, and distribution. It does not buy correctness, and it has never once made an assistant better at remembering what you asked it yesterday.

What I would actually do

Start with a free tier. Every one of these has either a free plan or a free beta, which means there is no reason to pay before you know how you use the thing. Spend a week sending real requests, not test prompts, and count how many messages a typical task consumes. That number decides whether metered or unlimited pricing is cheaper for you, and it is personal enough that no review can answer it on your behalf.

Then ask the boring question. An agent in your messages sits next to everything else in your messages. Look for outside validation, whether that is platform approval or a security audit, and weigh it heavily. In this category, the credential is the feature.

🕒 Published: October 12, 2026