Market map: 90 AI agents you can text
All 90 AI agents you can text, on two axes that are recorded fields rather than our opinion of anybody: across, what each one says it will do for you; up, how many of the 9 questions this site asks of every listing it answers. Open any product here and you can recompute its square.
Answers you
Researches, remembers and replies. Nothing of yours is connected to it.
Handles your accounts
Reaches into an inbox, a calendar or a task list you already own.
Acts in the world
Says it will book, buy or phone somebody who is not you.
Answers 7 to 9
of the 9 questions
Answers you, answers 7 to 9 of the 9 questions: 1 listings
Answers 5 or 6
of the 9 questions
Answers 0 to 4
of the 9 questions
Handles your accounts, answers 0 to 4 of the 9 questions: 5 listings
Acts in the world, answers 0 to 4 of the 9 questions: 9 listings
- Solid, heavier border
- Listed as a general assistant
- Solid, lighter border
- Does one job
- Dashed border
- Invitation or waitlist
- Question mark
- No price published
| Answers | Answers you | Handles your accounts | Acts in the world |
|---|---|---|---|
| 7 to 9 of 9 | 1 | 14 | 21 |
| 5 or 6 of 9 | 2 | 18 | 15 |
| 0 to 4 of 9 | 5 | 5 | 9 |
How to read it, and how not to
There is no diagonal. A grid with a top right corner gets read as a league table, so this one was tested before it was drawn. The rank correlation between how far a product goes and how much it publishes is 0.12, which is nothing: agents that handle your accounts average 6.16 of the 9 and agents that act in the world average 5.98. Doing more does not mean saying more, and it does not mean saying less. The map is here to show you who is in this market and how they group, not to rank them.
Across: what it says it will do. Read from the capability rows, which are what each product claims on its own pages rather than anything tested here. A listing is placed by the furthest it goes rather than by how many things it lists, so an agent that books a table sits under acts in the world whatever else it also does. 8 answers you, 37 handles your accounts and 45 acts in the world.
Up: how many of the 9 it answers. The same 9 mechanical criteria the leaderboard runs on, each one a thing a company could publish today at no cost: publishes a price; has a privacy policy; says whether it trains on you; addresses people who never signed up; says what happens if you cancel; names who carries the messages; names its model providers; says where to report a bug and says how you start. A low square means a company has not published something, never that a product is bad.
Price is a mark, not an axis. 26 of 90 publish no price at all and 41 of the rest start at zero, so a price axis would leave a third of this map empty and stack most of the remainder on one line. A chip carries a question mark when nothing is published. Which agents publish a price is its own page, because the refusal turned out to be the finding.
Nothing here is estimated. 9 listings sit behind an invitation or a waitlist and are drawn with a dashed edge rather than left off. A product that publishes nothing is recorded as publishing nothing rather than guessed into a band, and the order of names inside a square is alphabetical.
The square worth looking at
Bottom right holds 9: Air by WZRD, Alfi, Howie, Instinct, Mae, Miso, OpenInstinct, Orbits and szn. Each says it will book, buy or phone somebody on your behalf, and each answers four or fewer of the 9. That is not a judgement of any of them and none has done anything wrong. It is the one combination on this map where the gap between what a product will do and what its company has written down is widest, which makes it the square to open first.