An SMS AI agent for business is a large language model wired to your business phone number. It reads customer texts, answers in plain language, books or reschedules appointments, qualifies leads and passes the thread to a person when it should. In 2026 you can buy one inside a helpdesk or CRM, assemble one on a messaging API such as Twilio, Sinch or Infobip, or build your own.

Prices are as of November 2026. This guide is based on vendor pricing pages, documentation and launch coverage; we have not run these products ourselves.

What is an SMS AI agent, and how is it different from an autoresponder?

Older business texting tools matched keywords: text "HOURS" and get a canned reply. An AI agent reads free text ("can you come Thursday afternoon, the kitchen sink is leaking again") and works out the intent, the details and the next step.

The practical differences:

  • It understands messy messages. Typos, several questions in one text, follow-ups that refer back to earlier messages.
  • It can act. With tool access it can check a calendar, create a booking, look up an order or write a CRM note.
  • It knows when to stop. A well-configured agent hands off to a human on complaints, payments, medical or legal questions, or anything outside its instructions.

Which channels count as "SMS" here?

Plain SMS and MMS are the baseline because they reach every phone. Many platforms now run the same agent across WhatsApp, RCS and web chat. Twilio's Agent Connect SDK, generally available since May 6, 2026, targets SMS, WhatsApp, RCS and voice from one codebase (Twilio changelog). For a channel-by-channel breakdown, see AI agents by channel.

What do businesses actually use SMS AI agents for?

Vendor materials consistently emphasise four jobs:

  1. Appointment booking and reminders. Podium's AI Scheduler and similar tools let customers book, confirm or move a slot by text. See AI SMS appointment booking.
  2. Lead qualification and follow-up. The agent replies to a web form or missed call within seconds, asks a few qualifying questions and books a call. See AI SMS lead qualification.
  3. Tier-1 customer support. Order status, opening hours, returns policy, password resets, with handoff for anything harder. See AI SMS customer support.
  4. Order and delivery updates. Usually template-driven, with the AI handling replies such as "can you leave it with my neighbour".

Industry-specific guides cover home services, clinics and dentists and real estate.

How widespread is this?

Most adoption figures come from vendors, so treat them as directional. Sinch reports that 62% of organisations it surveyed have an AI customer-communications agent live (Sinch). The often-repeated "98% SMS open rate" has no traceable original source and cannot be measured, because SMS has no read tracking. We look at where that figure comes from in is the 98% SMS open rate real?.

Buy, assemble or build: which route fits?

There are three broad routes, and the right one depends mostly on how many messages you handle and whether you already run a helpdesk or CRM.

Route Examples Pricing model Best for
Buy inside a helpdesk or CRM Fin (Intercom), Zendesk AI agents, Salesforce Agentforce Per outcome, per resolution or per credit, plus seats Teams already on that helpdesk
Buy an SMB texting suite HighLevel Conversation AI, Podium AI Employee Monthly add-on plus SMS usage Local businesses, agencies
Assemble on a messaging API Twilio Agent Connect, Sinch Agent Builder, Infobip AgentOS Per message plus platform usage Mid-size teams with a developer
Build your own Twilio or Sinch number plus your own LLM code Per segment plus LLM tokens Developers, high volume

Prices as of November 2026. For the vendor detail, see SMS AI agent platforms compared. If you want to build, start with how to build an SMS AI agent on Twilio.

The trade-off in one line each:

  • Packaged tools go live in days with handoff, reporting and often opt-out handling built in, but per-resolution fees add up at volume.
  • Building your own gives the lowest per-message cost and full control of prompts and data, but you own 10DLC registration, STOP handling and security.

What does an SMS AI agent cost?

There are two layers: the cost of sending and receiving texts, and the cost of the AI.

Twilio US SMS, per segment each way
$0.0083
Fin, per outcome
$0.99
Zendesk, per automated resolution
$1.50-$2.00
Agentforce, per conversation
$2

Messaging layer (US, as of November 2026). Twilio lists $0.0083 per segment for outbound and inbound SMS, plus carrier fees such as $0.0035 for AT&T and $0.005 outbound for Verizon, and a $0.001 fee on failed messages (Twilio pricing). Sinch lists $0.0078 per 10DLC message each way (Sinch pricing). A 10DLC brand registration costs $4.50 or $46 one time on Twilio, campaign vetting $15, and the campaign fee $1.50 to $10 a month.

AI layer. Packaged tools bill per result: Fin charges $0.99 per outcome with a 50-outcome monthly minimum when used with another helpdesk (Fin pricing); Zendesk charges $1.50 per committed or $2.00 per pay-as-you-go automated resolution (Zendesk pricing); Agentforce sells Flex Credits at $500 per 100,000 (about $0.10 per standard action) or $2 per conversation (Salesforce).

A one-segment exchange (one text in, one out) on Twilio works out to roughly $0.024 to $0.03 including carrier fees, before AI costs. That figure is our arithmetic from Twilio's list prices, not a quoted rate. Full worked examples are in what an SMS AI agent costs.

Which rules apply before you switch it on?

Compliance is the most common reason SMS projects stall. The short version for the US and UK:

  1. Register the numberA2P 10DLC brand and campaign, or toll-free verification, before sending.
  2. Collect consentClear opt-in for anything marketing; keep a record of when and how.
  3. Handle STOP firstProcess opt-out keywords in code before the LLM sees the message.
  4. Disclose the AITell customers they are texting an automated assistant and how to reach a person.
  5. Set handoff rulesRoute complaints, payments and sensitive topics to staff.

For what the law says about telling customers they are texting an AI, see do you have to disclose AI in text messages?.

United States

  • 10DLC. Carriers require brand and campaign registration through The Campaign Registry for business texting on local numbers; unregistered traffic is filtered or blocked. See 10DLC registration for AI chatbots.
  • TCPA consent. Customers can revoke consent by any reasonable means unless you name an exclusive opt-out method, and you must honour it within a reasonable time of no more than 10 business days. The FCC's amended rules take effect November 25, 2026 under FCC 26-67. They let senders designate exclusive opt-out methods, require the seven reply words stop, quit, end, revoke, opt out, cancel and unsubscribe to be honoured, and narrow "revoke-all" so an opt-out from one informational category covers only that category. Details in TCPA rules for AI texting.

United Kingdom

PECR requires prior consent for marketing texts, with a "soft opt-in" exception for existing customers. The Data (Use and Access) Act 2025 raised maximum PECR fines to £17.5 million or 4% of global turnover. See UK PECR and AI text messages and, for the wider picture, AI agents for UK small businesses.

What about WhatsApp?

Since January 15, 2026, Meta's WhatsApp Business terms bar general-purpose AI assistants, but businesses using AI for their own customer service are still allowed. See WhatsApp AI agents for business.

What can go wrong?

  • Prompt injection. A crafted message can try to override the agent's instructions. Security researchers showed this with a WhatsApp-connected agent in 2025. See prompt injection via messages and the deeper explainer at getpersonalaiagents.com.
  • Spoofed sender numbers. Never treat the "From" number as proof of identity for sensitive actions.
  • Hallucinated answers. Restrict the agent to your own knowledge base and price list, and test with real past messages before launch.
  • Long replies. One GSM-7 segment is 160 characters; a chatty agent can triple your per-message bill.

Bottom line

For most small businesses, the fastest route is the AI agent inside a tool you already pay for, such as your helpdesk, CRM or booking system, as long as per-resolution fees make sense at your volume. Businesses with a developer and thousands of messages a month usually save money by assembling on Twilio, Sinch or Infobip. Either way, register your number, handle STOP in code, disclose the AI and keep a person on call for handoffs.