Texting AI agent lists now go out of date within a week. On 3 October 2026 TechCrunch published a roundup of 18 AI agents that you reach by text message instead of through an app. On 10 October it republished the piece under a new headline, “Here are the top AI agents that can live in your text messages”, with four new entries: Comma, Szn, Tab and Underdog. That takes the list to 22, and two of the four newcomers had launched or come out of stealth only days before.
We compared the two versions line by line, using archived copies of the original, and checked each new entry against the company’s own website and the reporting behind it. This article explains what changed, profiles the four newcomers, maps every texting AI agent by the messaging apps it supports, and picks the best option for each job, from family admin to private on-device help. It also flags five details in the list that still differ from what the companies publish, plus three new ones.
If you want the full profile of the original 18, with prices, funding and privacy questions, read our earlier guide to every AI agent you can text. This piece builds on it. It focuses on what is new, which texting AI agent ranks highest for each job, and what a UK business should put in place as staff start texting agents that can read work email, a question that now belongs in any AI strategy.
Table of contents
- What Changed in TechCrunch’s Texting AI Agent List
- The Four New Texting AI Agent Entries
- Every Texting AI Agent by Messaging App
- How We Picked the Top Texting AI Agent for Each Job
- Top Texting AI Agent Picks by Job
- Three Designs Behind Every Texting AI Agent
- What a Texting AI Agent Costs After the Refresh
- The Money and Momentum Behind the Texting AI Agent Race
- Where the Updated Texting AI Agent List Still Needs Correcting
- Safety Checks Before You Use a Texting AI Agent
- What a Texting AI Agent Means for UK Businesses
- Frequently Asked Questions
- References
What Changed in TechCrunch's Texting AI Agent List
AI agents that live in a messaging app are one of the fastest-moving corners of consumer technology this autumn, and the TechCrunch list is the closest thing the category has to a public register. The original article went live at 14:00 UTC on 3 October. Archived copies from that afternoon and from 5 October carry identical text, so the list did not change during its first week. The 10 October version replaced it at a new address, and the old address now returns a “page not found” error.
Four new entries in one week
The four additions are Comma, an open-source texting AI agent; Szn, which runs errands from a single conversation thread; Tab, which emerged from stealth on 6 October at a $300 million valuation; and Underdog, a privacy-focused assistant whose invite-only beta opened on 5 October. Each one is profiled in the next section. Their arrival means 22 services now count as a texting AI agent in TechCrunch’s eyes, up from 18, an increase of 22% in seven days.
The new headline matters too. “All the AI agents that can live in your text messages” became “Here are the top AI agents that can live in your text messages”, and the introduction now describes “the most notable options so far” rather than “some of the most notable options”. TechCrunch has quietly turned a complete-looking census into a curated shortlist, which is a fairer description of a market that is adding services faster than any list can track.
Small edits to existing entries
Only two of the original 18 entries changed in substance. Fambot’s integrations grew: the 3 October version said it connected to Gmail and Google Calendar, with Outlook and Apple Calendar planned, while the 10 October version lists Gmail, Google Calendar, Outlook and WhatsApp, with only Apple Calendar still to come. And Fo, Wajo’s first consumer agent, now correctly launched publicly in September 2026 rather than September 2025.
The other changes are copy edits: “cancelling” became “canceling”, “Hermes agent” became “Hermes Agent”, and a clumsy sentence in the Miso entry was tidied. None of them changes a price, a funding figure or a feature.
One correction made, five still standing
Our 3 October guide found six details in the original list that differed from the companies’ own pages. The Fo date is the only one TechCrunch has fixed. The other five carry over unchanged into the refreshed list, and the table below summarises where each stands.
| Item | 3 October version | 10 October version |
|---|---|---|
| Headline | All the AI agents that can live in your text messages | Here are the top AI agents that can live in your text messages |
| Number of agents | 18 | 22 |
| New entries | None | Comma, Szn, Tab, Underdog |
| Fambot integrations | Gmail, Google Calendar | Gmail, Google Calendar, Outlook, WhatsApp |
| Fo public launch | September 2025 (wrong) | September 2026 (fixed) |
| folk and Martin prices | Yearly rates shown as monthly | Unchanged |
| Ollie security claim | “SOC 2 compliance”, type not given | Unchanged |
| Skye funding | About $3.6m “seed” | Unchanged (filing says pre-seed) |
The Four New Texting AI Agent Entries
Each of the four newcomers makes a different bet about what a texting AI agent should be. Tab gives the agent its own identity and wallet. Underdog keeps the model on your computer. Comma publishes its code. Szn sells the follow-through rather than the conversation. The table summarises them before the detail.
| Agent | Where you reach it | Big idea | Price | Status |
|---|---|---|---|---|
| Tab | iMessage, WhatsApp | Own phone number, computer and wallet | Not published | Private beta; $300m valuation |
| Underdog | Mac and Windows app | Model runs entirely on your device | Free, no ads | Invite-only beta from 5 Oct |
| Comma | Signal, Telegram, desktop | Open source (AGPL-3.0), self-hostable | $0 Value plan; Pro adds credits | Public |
| Szn | One text thread | Calls, emails and follow-ups done for you | Not published | Public sign-up |
Tab: a texting AI agent with its own phone, computer and wallet
Tab announced on Tuesday 6 October that it had emerged from stealth at a $300 million valuation, according to TechCrunch, though the company declined to share how much it raised. Its investors include SV Angel, Valar Ventures and American Spirit. The founders, Brennan Erbz, Stafford Schlitt and Ammar Amdani, started work on it earlier this year. You text it on iMessage or WhatsApp, and it handles tasks such as ordering groceries or finding birthday presents.
Tab’s own site describes a texting AI agent with “its own computer and browser” that can open websites, make files, use the accounts you connect and phone a hotel, a doctor’s office or a restaurant, wait on hold and report back. Amdani told TechCrunch it “has its own phone number, computer, and wallet”. The trust rules are unusually specific: your data is never used to train AI, your card sits in a vault Tab never sees, and before it spends you see what it is buying, the total and who gets paid. Nothing is paid until you approve.
Underdog: private AI that runs on your own computer
Underdog comes from Conway Research, founded by Thiel Fellow Sigil Wen, and opened an invite-only beta on Monday 5 October. Its model runs wholly on the user’s device, currently a Mac with Apple silicon or a Windows PC in beta, with Linux, iPhone and Android promised. Andreessen Horowitz led its first round, which Underdog says closed in April and a16z announced on 2 October, alongside Khosla Ventures, Hummingbird, SV Angel and the Anthology Fund run by Menlo Ventures and Anthropic. Angels include Stripe’s Patrick Collison.
The flagship model, Underdog 27B, is built on Qwen3.8 27B. Underdog’s site claims a score of 33.7 against 31.9 for Claude Opus 4.6 Max on the Artificial Analysis Intelligence Index, but its own footnote says those are the base Qwen model’s scores. A smaller version, Saluki 27B, is 7.89 GB and fits a 16 GB laptop. The business model is novel: the app is free and will never carry ads, and Underdog takes a small cut when it buys something for you using a one-time card.
Comma: the free, open-source option
Comma calls itself “a free, open source personal AI agent” that is “always on, proactive and sessionless”. Its code is published under the AGPL-3.0 licence, and you can self-host it with Docker Compose so your data stays in your own databases, with model calls going to whichever provider you choose. The hosted Value plan costs $0 with unlimited chat if you bring your own API key, and a Pro plan adds credits for heavier use.
The design is built around tasks rather than chats. You talk to a Router, which hands each task to a Worker. You can pick a model for each role, or let OpenAI’s Codex or Anthropic’s Claude Code act as the Worker. Comma works on its own cloud computer, so tasks continue when your Mac is closed, and you check in from Signal or Telegram. Decisions wait for you in a “Needs Review” column, and connected computers stay read-only until you allow more.
Szn: one thread for calls, emails and follow-ups
Szn’s pitch is that “so much of getting something done is everything after deciding to do it”: the email, the phone call, the follow-up, finding a time. You send it a text, voice note, photo or forwarded email in one thread, and it makes calls, sends emails and chases replies, stepping back to you only for decisions. TechCrunch cites the company’s figures of more than 7,500 businesses in 16 countries, over 10,000 jobs and 95 hours of calls.
Those numbers need context, which we explain in the corrections section below. Szn’s site does not name a pricing plan or a specific messaging app, so it is the hardest of the four to compare on cost or channel. It fits the list on behaviour rather than branding: it is a texting AI agent whose value is in the dull follow-through that most chat assistants leave to you.
Every Texting AI Agent by Messaging App
The first question for any buyer is not which texting AI agent is cleverest but which one you can actually reach. iMessage only works on Apple devices, RCS and SMS reach Android, and WhatsApp, Telegram and Signal work anywhere. The table maps all 22 entries. For the original 18 we use the channels each company publishes, as checked for our earlier guide; for the newcomers we use their own sites.
| Agent | Where you text it | Works without an iPhone? |
|---|---|---|
| Caddy | iMessage, RCS | Yes (RCS) |
| Comma | Signal, Telegram (WeChat per TechCrunch) | Yes |
| Fambot | SMS plus app and web | Yes |
| folk | iMessage, Telegram; WhatsApp rolling out by region | Yes (Telegram) |
| Instinct | SMS and voice | Yes, if invited |
| Iris | iMessage | No |
| Martin | SMS, WhatsApp, phone, Slack | Yes |
| Miso | iMessage | No |
| Ohai | Text, email, voice | Yes |
| Ollie | Text and group chats | Yes |
| Orbits | Text plus app | Yes |
| Pally | Text | Yes |
| Poke | iMessage, SMS, Telegram, WhatsApp | Yes |
| Rene | iMessage, Telegram, WhatsApp | Yes |
| Skye | iPhone home-screen widgets | No |
| Stanley | iMessage, Telegram, web | Yes (Telegram) |
| Szn | One text thread (app not named) | Not stated |
| Tab | iMessage, WhatsApp | Yes (WhatsApp) |
| Tomo | iMessage | No |
| Town | iPhone app; WhatsApp inbound only | Partly |
| Underdog | Mac and Windows app; no chat channel listed | Yes (desktop) |
| Wajo (Fo) | Text, calls, email | Yes |
How many apps each texting AI agent reaches
Counting only named messaging apps (iMessage, RCS, SMS, WhatsApp, Telegram, Signal and Slack), Poke leads with four. Martin and Rene reach three each. The chart counts the channels in the table above, so a service listed only as “text” is left out because it does not name the app, and folk’s regional WhatsApp rollout is not counted. Bar widths are each count divided by Poke’s four.
iPhone-only texting AI agent picks
Four services in the table need an iPhone: Iris, Miso and Tomo work only in iMessage, and Skye is a set of iPhone home-screen widgets. Apple’s own Messages for Business programme works only on Apple devices, and it is where Poke became the first approved AI agent in June. If everyone in your household or team uses an iPhone, the iMessage-only options are worth testing. If anyone uses Android, a texting AI agent built only for iMessage will leave them out of shared chats.
Texting AI agent options that work on Android
Android users have more choice than the iMessage-heavy coverage suggests. Caddy uses RCS, the modern messaging standard on Android. Poke, Martin, Fambot and Instinct work over SMS. Poke, Rene, Tab and Martin support WhatsApp, and Poke, Comma, folk, Stanley and Rene support Telegram. Comma is the only service in the list that names Signal, the encrypted messenger, which matters to anyone who wants the chat itself end-to-end encrypted.
Agents that are not really texted
Three entries stretch the definition. Skye is a home screen, not a chat. Town’s WhatsApp and Telegram links only take messages in. And Underdog’s own website describes a desktop app for email, calendar, meeting notes, dictation and errands, without naming any messaging channel at all. Underdog belongs on any list of personal agents, but by its own description it is not yet a texting AI agent in the plain sense.
How We Picked the Top Texting AI Agent for Each Job
TechCrunch’s list is alphabetical, and its new headline promises the “top” agents without ranking them. Ranking all 22 on one scale would be misleading, because a family organiser and a content assistant are not competing for the same job. So we picked a winner for each common job instead. Every choice rests on four tests, using only what the companies and their reporting publish, as the table shows.
| Test | What we checked | Why it matters |
|---|---|---|
| Reach and availability | Named messaging apps; public sign-up or invite-only | An agent you cannot join or reach is not a choice |
| Ability to act | Calls, bookings, purchases, files, not just reminders | Separates a real agent from a chat reminder |
| Privacy and security evidence | Audits, training policy, approval steps, where the model runs | These agents read email and can spend money |
| Price you can check | A price on the company’s own page, billed monthly | Unpublished pricing is a risk for a paid habit |
Reach and availability
A third of the list is still gated. Instinct and Tab are in private beta, Underdog needs an invite, and Skye has not set a release date. We prefer a texting AI agent that is open for sign-up today, and we say so where a pick depends on an invite. Reach also covers the people around you: a texting AI agent that only works in iMessage cannot join a group chat with Android users.
Ability to act, not just remind
Almost every texting AI agent here can set reminders and read a calendar. The ones that matter can also act: Instinct, Tab, Fo, Martin and Szn place phone calls, and Tab, Fo and Underdog can pay for things, through a wallet, the agent’s own card or one-time cards. We weight that capability heavily, because it is what the category is selling, and we weight the safeguards around it just as heavily.
Evidence on privacy and security
A texting AI agent that reads your inbox should show its work. Ollie has a SOC 2 Type 1 report, a point-in-time review of its controls. Underdog has passed the App Defense Alliance’s CASA Level 1 assessment with Leviathan Security Group, issued on 10 September 2026. Tab, Comma and Rene state that your data is not used to train models. Services that publish none of this are not disqualified, but they cannot win a privacy-sensitive category.
Price you can check
We use the price on each company’s own page, billed monthly, as our earlier guide did. Where a figure appears only in TechCrunch’s list, such as Pally’s and Tomo’s, we say so. A price billed yearly but shown per month, as with folk’s $8.33 and Martin’s $21, is converted back to the monthly rate.
Top Texting AI Agent Picks by Job
Here is how the four tests play out for each texting AI agent. The table gives the pick and runner-up for each job, and the sections below explain each call. No texting AI agent on the list has a long public track record. Most launched this year, so treat every pick as provisional and test it with a low-risk task first.
| Job | Top pick | Runner-up | Deciding factor |
|---|---|---|---|
| All-round personal help | Poke | Rene | Four messaging apps, free tier, Apple approval |
| Errands end to end | Tab | Instinct | Own phone and wallet with approval before paying |
| Family organising | Ollie | Fambot | Published audit and prices |
| Maximum privacy | Underdog | Comma (self-hosted) | Model runs on your own machine |
| Free and open source | Comma | Iris (built on open-source Hermes Agent) | AGPL-3.0 code you can run yourself |
| Group chats | Rene | Tomo | Agent-to-agent planning across three apps |
| Work | Town | Martin | Built for recurring professional tasks |
| Calls and follow-ups | Szn | Martin | Follow-through is the whole product |
| Travel | Miso | Fo | AI planning backed by a human travel team |
| Social content | Stanley | None comparable | The only creator-focused entry |
Best all-round texting AI agent: Poke
Poke remains the default texting AI agent to try first. It reaches four messaging apps, has a free tier with Pro at $19 a month, and in June became the first AI agent Apple approved for its Messages for Business platform. It had relayed about 100 million messages by then, and in July its maker, The Interaction Company of California, was bought by AI coding company Cognition. We covered the startup in our Poke AI agent profile. Rene is the runner-up for its three apps and free start.
Best for errands end to end: Tab or Instinct
Instinct is the best-funded texting AI agent on the list, valued at $10 billion after its $1 billion Series C, as we reported in our Instinct Series C story. It has its own email addresses, places calls and now joins group chats. But it remains in private beta. Tab takes the pick because its approval rule is explicit: you see the item, the total and the payee, and nothing is paid until you say yes. Both are invite-gated today, so Fo is the open alternative.
Best family texting AI agent: Ollie, with Fambot close behind
Ollie is the only family-focused texting AI agent on the list with a SOC 2 report, albeit Type 1 rather than the stronger Type 2. Its prices are on its own page: free for 50 messages, $25 a month for 150 and $100 for 1,000. Fambot is the runner-up and the better fit for Android-heavy households, since it sends a nightly SMS digest of the next day’s events and is free during its beta. Its new Outlook and WhatsApp connections were the main change to an existing entry in the refreshed list.
Best for privacy: Underdog, with a caveat
If the question is “where does my data go”, Underdog has the strongest answer on the list: the model runs on your own computer, mail syncs directly with your provider, mail on disk is encrypted with AES-256, and a “Naval mode” stops it calling home. The caveat is the one above: by its own description it is a desktop assistant, not something you text. For a privacy-first texting AI agent today, a self-hosted Comma is the closest match, at the cost of running your own server.
Best free and open-source texting AI agent: Comma
Comma is the only entry whose full code is public under an open-source licence, and the only one you can self-host in full. That makes it the natural pick for developers and small teams who want a texting AI agent without a subscription. Iris is the runner-up: it is a hosted product, free with a $19 Plus plan, but it is built on Nous Research’s open-source Hermes Agent.
Best texting AI agent for group chats: Rene and Tomo
Rene’s site shows two users’ agents negotiating a dinner date “Rene to Rene”, and it supports iMessage, Telegram and WhatsApp, so it can sit in mixed iPhone and Android groups. We covered its launch as a multiplayer AI agent in September. Tomo also joins group conversations, but only in iMessage, and TechCrunch puts its basic plan at $19.99 a month.
Best texting AI agent for work: Town and Martin
Town is the only entry built around professional life. Its “Townie” learns how you work and handles recurring tasks across email, calendars, documents and Slack, and it raised a $55 million Series A from Andreessen Horowitz and Forerunner in June. Martin is the runner-up because it works over SMS, WhatsApp, Slack and the phone, though its Basic plan costs $35 a month billed monthly, the highest published entry price on the list.
Best for travel and content: Miso and Stanley
Two specialists win their categories by default. Miso pairs AI trip planning in iMessage with a human travel team and takes loyalty points into account. Stanley is the only creator-focused texting AI agent: it connects to your Instagram, turns voice notes into scripts and captions, and keeps a content calendar. Neither publishes a price, so ask before you commit.
Three Designs Behind Every Texting AI Agent
Every texting AI agent on the list looks alike from the outside, because each lives in a chat thread. Underneath, they follow one of three designs, and the design decides what can go wrong. The four newcomers make the split clearer than before, since each represents a design in its purest form.
| Design | Examples | Strength | Main risk |
|---|---|---|---|
| Cloud agent using your accounts | Poke, Caddy, Ollie, Rene, Szn | Easy to start; sees your real inbox and calendar | Broad delegated access held by a startup |
| Agent with its own identity | Tab, Fo, Instinct | Acts in the world without your passwords | Spends money and speaks for you in public |
| Local or self-hosted agent | Underdog, self-hosted Comma | Data stays on hardware you control | Smaller models; you run the security |
Cloud agents that borrow your accounts
Most of the list works this way. You connect Gmail, Outlook or a calendar, usually through delegated access so the service never sees your password, and the texting AI agent reads and acts from the company’s servers. It is the easiest design to start with, and the most useful on day one. The risk is concentration: a young company now holds standing access to many people’s email, and its security becomes your security.
Agents with their own identity
Tab, Fo and Instinct take a different path. Each agent gets its own phone number, email address or payment card, so it can sign up for services, call businesses and pay without handling your credentials. Fo’s maker Wajo built its first consumer agent around that idea, as our Wajo launch story explains. The risk moves rather than disappears: a texting AI agent with a wallet can buy the wrong thing, and one with a phone number can say the wrong thing to a real person.
Local and self-hosted agents
Underdog runs its models on your computer, and Comma can run on a server you manage. Both keep personal data on hardware you control, which is why privacy-minded buyers are watching them. The trade-off is capability and effort. On-device models are smaller than the frontier models in data centres, and our local LLM hardware guide explains what a laptop can realistically run. Self-hosting means patching and securing the server yourself.
What a Texting AI Agent Costs After the Refresh
Price is where the texting AI agent list is least complete. Of the 22 services, 12 are free to start, two list only paid plans in TechCrunch’s text, and eight publish no price or are not open to the public. The chart shows the split. Widths are each group’s share of 22.
A free texting AI agent is the norm
More than half the list costs nothing to try, and the refresh tilted it further that way. Two of the four newcomers, Comma and Underdog, are free, while Tab and Szn have not published prices. Among the original 18, paid plans start at about $10 a month for folk and Ohai and run to $35 for Martin, billed monthly. For most people, trying a free texting AI agent for a week is the cheapest way to learn whether the habit sticks.
New business models: transactions, not subscriptions
Underdog’s plan is the most interesting change. Because the model runs on your hardware, it has almost no inference bill, so it says it will stay free and earn a small cut of purchases made with its one-time cards, which TechCrunch compares to an interchange fee. Comma’s model is open-core: free software, with paid credits for those who want hosted capacity. Both are a response to the cost problem we described in the ugly economics of consumer AI.
Read the billing period
The two pricing errors from the original list are still there. folk’s $8.33 is its $99.99 yearly plan divided by 12; the monthly price is $9.99. Martin’s $21 is its Basic plan billed yearly; billed monthly it is $35, and Pro is $49. Before you pay for a texting AI agent, check whether the headline price commits you to a year.
The Money and Momentum Behind the Texting AI Agent Race
Investors are still pouring money into the texting AI agent market, and the newcomers show how high the opening bids have become. Tab’s debut valuation of $300 million matches what Poke was valued at after its extension in April, when Poke was already a known brand. The chart compares the three disclosed valuations. Widths are each figure divided by Instinct’s $10 billion.
One giant and a crowded field
Instinct’s two rounds in five weeks total $1.35 billion, and the $10 billion valuation sits 33 times above anyone else’s disclosed figure. Below it, the money is spread thinly: Town’s $55 million Series A, Poke’s $25 million, Fambot’s $3.5 million pre-seed and Skye’s roughly $3.6 million. Conway Research did not disclose the size of the round Andreessen Horowitz led for Underdog, and Tab did not disclose its raise.
The directory is growing more slowly
The Agent Store, an independent directory of agents you reach in iMessage, listed 139 agents in mid-September and 144 on 3 October. On 11 October it showed 145, alongside 15 infrastructure providers. That is five new listings in the first 17 days and one in the following eight. One directory is not the whole market, but it suggests the rush of new launches is levelling off while funding for the leaders keeps climbing.
The real competition is the assistant already on your phone
Start-ups are not competing only with each other. Meta’s Muse and OpenAI’s Dots are being built into apps people already use, and we asked whether either can be trusted in Can you trust Muse or Dots to run your life?. Comma pitches itself directly as “an open source alternative to Muse, Dots, Instinct and Town”. An independent texting AI agent has to be clearly better at something specific, which is why the newcomers lead with privacy, openness or follow-through.
Where the Updated Texting AI Agent List Still Needs Correcting
The refreshed list is still the best single map of the texting AI agent category, but it repeats five errors from the original and adds three new problems in the newcomer entries. None changes the overall picture. Each matters if you are choosing on price, trust or channel.
| Agent | The list says | The primary source says |
|---|---|---|
| folk | Pro is $8.33 a month | $9.99 monthly; $8.33 only on the $99.99 yearly plan |
| Martin | Starts at $21 a month | $21 is Basic billed yearly; $35 billed monthly |
| Ollie | Achieved SOC 2 compliance | SOC 2 Type 1, a point-in-time design review |
| Skye | About $3.6m in seed funding | More than $3.58m in pre-seed funding (SEC filing) |
| Town | Works with Slack and other platforms | WhatsApp and Telegram are inbound-only |
| Szn (new) | Has “served” more than 7,500 businesses | The figure counts businesses its users’ errands reached |
| Comma (new) | Reachable on Signal, Telegram and WeChat | Its own FAQ names only Signal and Telegram |
| Underdog (new) | Listed among agents that live in your texts | Its site describes a desktop app with no chat channel |
Why the Szn figures need context
TechCrunch says Szn “has served more than 7,500 businesses”. Szn’s own page frames the counter differently: “Every booking, order, and errand connects a real person with a real business.” The businesses are the restaurants, shops and offices that Szn contacted on its users’ behalf, not paying customers. The call figure is also smaller than it sounds: 95 hours across more than 10,000 jobs works out at no more than 34 seconds of calling per job on average, so most jobs involve no call at all.
Why Underdog is an odd fit
Underdog is a significant launch, and its on-device approach is the clearest privacy alternative in the category. But nothing on its site, its security page or in TechCrunch’s launch report describes texting it from iMessage, SMS, WhatsApp or any other messaging app. It is on the list as a personal agent and privacy benchmark. Readers looking specifically for a texting AI agent should treat it as one to watch until it ships a phone app.
Safety Checks Before You Use a Texting AI Agent
Giving a texting AI agent access to your email, calendar or card is a real decision, however casual the chat feels. We covered the wider privacy debate in AI agent makers are promising privacy. Four checks take a few minutes and cover the biggest risks for any texting AI agent.
Approval before money moves
The best policies in the list are specific. Tab says nothing that can be undone happens without your “yes”, and shows the item, total and payee before paying. Underdog’s browser agent “asks before it buys, pays, sends, or deletes anything”. Comma parks decisions in a review column. If a texting AI agent cannot tell you exactly when it will ask permission, do not connect a card.
Delegated access, not passwords
Rene says email and calendar use delegated access, so it “never sees or stores your password”, and Underdog says it never sees your password or your messages. Prefer services that connect through the official sign-in screens of Google or Microsoft, and review the permissions they request. You can revoke delegated access from your Google or Microsoft account settings at any time.
Know where the model runs
Most entries send your messages to a cloud model, sometimes through several providers. Underdog keeps processing on your computer, and a self-hosted Comma sends model calls only to the provider you choose. Tab, Comma and Rene say your data stays out of model training. Read the training clause in the terms before you forward anything sensitive.
Texting a US number from the UK
Many texting AI agent services are US start-ups, and one reached over SMS may give you a US phone number. Texting it from a UK mobile can count as an international message under your tariff. iMessage, WhatsApp, Telegram and Signal travel over the internet instead. Check whether a texting AI agent is available outside the US before you build a routine around it.
What a Texting AI Agent Means for UK Businesses
The texting AI agent list is aimed at consumers, but the effects land in offices. Staff connect personal agents to work accounts, and agents increasingly phone businesses on their users’ behalf. Neither needs a procurement decision to happen. Strong cybersecurity basics help, but the faster fix is a short, clear policy.
Staff will connect work accounts
Every texting AI agent on this list asks for email and calendar access, and nothing stops an employee granting it to a work Microsoft 365 or Google Workspace account. Both platforms let administrators control which third-party apps users can consent to. Review those settings now, and decide whether any agents are acceptable for work data. Any IT governance review should now include exactly this audit.
Agents will phone your front desk
Instinct, Tab, Fo, Martin and Szn all place calls, and Instinct and Muse both added calling in September, as we covered in rival AI agents add calls. Reception, booking and support teams will increasingly talk to software acting for a customer. Decide what a texting AI agent may change on a customer’s behalf, such as a booking, and what still needs the customer in person, such as a payment change or a security question.
Write a short texting AI agent policy now
A one-page policy covers most of it: which agents, if any, are approved for work data; that no agent may hold work passwords; that agents may not make payments on company cards; and how staff report a texting AI agent that acted without approval. Review it quarterly, because this list changed by four entries in a single week and will keep changing.
Frequently Asked Questions
Which texting AI agent is best overall?
For most people, Poke. It reaches iMessage, SMS, Telegram and WhatsApp, has a free tier, and was the first AI agent Apple approved for Messages for Business. Rene is a strong alternative, especially for group chats.
What changed in TechCrunch’s updated list?
On 10 October TechCrunch added Comma, Szn, Tab and Underdog, taking the list from 18 to 22. It also updated Fambot’s integrations, corrected Fo’s launch year to 2026 and retitled the article “Here are the top AI agents that can live in your text messages”.
Which texting AI agents work on Android?
Caddy uses RCS; Poke, Martin, Fambot and Instinct use SMS; Poke, Rene, Tab and Martin use WhatsApp; and Poke, Comma, folk, Stanley and Rene use Telegram. Iris, Miso and Tomo need an iPhone.
Is there a free, open-source texting AI agent?
Yes. Comma publishes its code under the AGPL-3.0 licence, can be self-hosted with Docker Compose, and has a $0 hosted plan if you bring your own API key. You reach it through Signal or Telegram.
Which texting AI agent is most private?
Underdog keeps its model on your own computer, but by its own description it is a desktop app rather than a chat. If you want a texting AI agent in the strict sense, a self-hosted Comma keeps the most data under your control.
References
Here are the top AI agents that can live in your text messages (TechCrunch, 10 October 2026)
All the AI agents that can live in your text messages, archived 5 October 2026 (Internet Archive)
Meet Tab, which emerged from stealth with a $300M valuation (TechCrunch)
Tab: meet your AI in Messages (Tab)
Silicon Valley’s AI wunderkind launches Underdog (TechCrunch)
Underdog: the most loyal AI (Underdog)
Investing in Conway, the creator of Underdog (Andreessen Horowitz)
Comma: your 24/7 AI personal assistant, free and open source (Comma)
szn: the work behind your life (Szn)
Rene: multiplayer personal superintelligence (Rene)
folk product summary and channels (folk)
Fambot introduces an AI chief of staff for families (TechCrunch)
Apple approves Poke as the first AI agent on its Messages for Business platform (TechCrunch)
Why Cognition bought Poke (TechCrunch)
Viral AI agent Instinct raises $1B Series C at a $10B valuation (TechCrunch)
Ollie is betting privacy can win the AI assistant race (TechCrunch)
Town raises $55M Series A from a16z and Forerunner (Yahoo Finance)
The App Store for iMessage Agents (Agent Store)
Configure how users consent to applications (Microsoft Learn)
Control which apps access Google Workspace data (Google Workspace Admin Help)
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.
