Texting an AI Agent on iMessage: What to Expect

01

Getting started

Most iMessage agents start with one text: you sign up on the maker’s site, get a number, and say hello. Some put you on a waitlist first, and the number arrives when you’re let in.

The first thing worth doing is saving it as a contact. Some agents send you their contact card in the first few messages; either way, saving it means it shows up with a name instead of a bare phone number. That matters more than it sounds: a named contact is easy to find in search, easy to share with a friend, and easy to add to a group chat later.

Expect a short hello rather than a form. Good agents ask one question that unblocks your first job and skip the questionnaire. Some will ask you to connect an account, such as your email or calendar, before they can help with it. That step usually opens a sign-in page in your browser, not a password typed into the chat.

A good agent never asks you to text it a password. If one does, stop.

02

What it’s like day to day

You text it a goal, it tells you it’s on it, and the result comes back as a message. The difference from a chatbot is the middle part: the agent opens websites, fills forms, makes calls or sends emails while you get on with your day.

A typical errand looks like this. You text “move my dentist appointment to next week”. A few seconds later you get a short reply saying it’s looking. Then, a few minutes later, either a done message with the new time or a question it needs you to answer, like “Tuesday 9 or Thursday 2?”. You reply with one word and it finishes.

Agents differ a lot in how far they go on their own:

• Does the errand itself. Tab does the errand instead of just answering: it books, calls businesses and buys things with your approval, then texts you the result. Asmi calls, texts and emails providers to cancel memberships, dispute charges and book appointments, and keeps following up until it’s done.
• Texts you first. Kiwi reads your calendar and email and messages you at the right moment, often with the small job already lined up, like a car booked when a flight is late.
• Works your inbox and calendar. Howie schedules meetings when you CC it on an email or text it, then handles the back and forth and sends the invite. Dunnit sorts your inbox and calendar overnight and sends a morning brief.
• Handles research and automations. Poke covers email, calendar, research and automations from one chat thread.

Expect some jobs to take minutes, not seconds. A real errand involves real websites, and those are slow. A good agent tells you roughly how long, and says something if it gets stuck instead of going quiet.

03

Group chats

Some iMessage agents can join a group chat, so everyone sees the plan as it comes together. This is where texting an agent beats a separate app: the plan lives where the people already are.

Adding one usually works like adding a friend. You add its contact to the group, and from then on anyone in the chat can ask it something. Jarvie researches trips and stays, finds restaurants, keeps track of dates, splits receipts and summarizes the chat. Alfi joins as a member of the group, plans social events, books dinner reservations and sets reminders for everyone. Caddy and Tomo also work in group chats.

Two things to know before you add one. First, everyone in the group can see what the agent says, so don’t ask it about anything private there. Second, check what the agent remembers from the group and whether that stays separate from your one-on-one chat with it. The agent’s own page or privacy policy should say.

04

Photos, voice notes and files

Many agents can read the photos and voice notes you send, which is often faster than typing. A photo of a receipt, a screenshot of an error or a quick voice note while you walk is usually enough to start a job.

Support varies, so check before you rely on it. Caddy works with voice memos. Tomo lists web, video and photo tasks. Jarvie handles images one-on-one. Agents that build things, like Tab, can send finished files back into the thread: a document, a video or a link to a page.

A few habits make this work better:

• Add one line of context. A photo of a menu plus “book a table here for Friday at 8” beats the photo alone.
• Keep voice notes short. One job per note is easier to act on than a long ramble.
• Expect links as cards. Booking pages and maps usually arrive as a preview card you can tap.

05

Approvals and spending

A good iMessage agent shows you what it’s about to do and waits for your yes before it spends money or speaks for you. This is the single most important habit to look for.

In practice the approval is a message: here’s the item, the total and who gets paid, reply yes to buy. Tab shows exactly what it’s buying before spending, and nothing is paid until you approve it. Ori finds the product, confirms the price and orders only after you say yes. Kiwi books and orders when you say yes. Arlo sends emails, books and calls after your approval.

If an agent spends or sends something you didn’t approve, treat it as a reason to stop using it. You should never have to guess what it did on your behalf.

06

Privacy

iMessage protects messages on the way between devices, but the agent on the other end reads everything you send it, so its maker’s privacy policy is what really counts. Texting an agent is like texting a company: your messages reach the service that runs it.

Before you connect your email or calendar, look for a few plain answers on the maker’s site:

• Who can read your messages. Is anyone at the company reading them, or only the software?
• Whether your data is sold or used to train models. The good ones say this in one line.
• Whether you can delete it. Tomo says you can permanently clear your data from its dashboard and that it never sells personal conversations.
• What it connects to. An agent that reads your inbox can see a lot. Connect only what the job needs.

For a fuller checklist, read the best private AI agents.

07

What iMessage agents still get wrong

They are fast at simple jobs and slower or shakier on long, messy ones. Knowing the weak spots saves frustration.

Websites change, logins expire and some sites block automated visitors, so an errand can stall halfway. A good agent says so plainly and suggests another route, like calling instead. Agents can also misread a vague request, so “book dinner Friday” works better as “book a table for two near me Friday at 8, under what we usually spend”. Long threads can also lose details, so restate anything that really matters.

If you’re new to all of this, what is an AI agent explains what’s happening behind the chat, and AI agent vs chatbot explains why some only answer while others act.

08

Common questions

Do I need an iPhone to text an AI agent?

No, not always. Many agents that work in iMessage also work over plain SMS, and some over RCS on Android. Caddy works in iMessage on iPhone and RCS on Android. Check the agent’s page for where it runs.

Do I need to install an app?

Usually not. Most iMessage agents work from a phone number you save as a contact. Some offer an optional app for settings or connecting accounts.

Can an AI agent on iMessage join my group chat?

Some can. You add the agent’s contact to the group like any friend, and then anyone in the chat can ask it things. Jarvie and Alfi are built for group chats.

Can it understand photos and voice notes?

Many can, but not all. Send a short test, like a photo of a receipt or a quick voice note, and see what comes back before you rely on it.

Is it safe to give an iMessage agent my email or calendar?

It can be, if the maker says clearly who reads your data, that it isn’t sold, and how to delete it. Connect only what the job needs, and never text a password into the chat.

Will it spend money without asking?

A good one won’t. Look for an agent that shows the item, the total and who gets paid, and waits for your yes before buying.

09

Find one that fits

The easiest way to learn what texting an agent is like is to try one on a real errand from this week, something with a website and a follow-up. AI agents on iMessage lists every one that lives in Messages, and the full directory shows them all side by side.