SMS was designed in 1984. It has a 160-character limit, no read receipts, no typing indicators, no native media, and no way to verify who sent it. For 40 years, that has been the protocol layer underpinning billions of business-to-human interactions — and the one that AI agents inherit by default when they text a user.
RCS (Rich Communication Services) was supposed to change that. The GSMA published the first specification around 2007. It has taken roughly 17 years to reach escape velocity. As of late 2024, it finally has.
This matters more for AI agents than for any other kind of sender.
What RCS actually is
RCS is the carrier-level successor to SMS. It's baked into the default messaging app on Android and iPhone — no separate app required. When a message is sent over RCS it gets:
For a human-to-human conversation, this looks like iMessage or WhatsApp but built into every carrier's default messaging layer. For business-to-consumer messaging (called A2P RCS or RCS Business Messaging), it looks like a branded, interactive message thread instead of a plain-text SMS from an unfamiliar number.
Why it took 17 years
RCS had a coordination problem so severe it became a textbook case.
Every carrier had to implement the standard independently. Different carriers shipped different subsets of the spec. Interoperability between carriers required bilateral agreements. The result was a patchwork where an RCS message sent from a T-Mobile subscriber might degrade to SMS when it hit a Verizon subscriber, or fail entirely when it crossed international borders.
But the deeper problem was Apple.
The US iPhone market share sits above 55%. Until every iPhone supported RCS, any business or developer building on RCS was building on a channel that couldn't reach most American consumers. The "green bubble" problem wasn't cosmetic — it meant the channel had a structural ceiling.
The GSMA pushed. Google pushed harder, making RCS the default in Google Messages starting in 2018–2019. Sprint, then T-Mobile, AT&T, and Verizon all came online. But none of it was enough to move the needle without Apple.
The inflection point
In November 2023, Apple announced it would support RCS. In June 2024, at WWDC, Apple confirmed it would ship in iOS 18 that fall. On September 16, 2024, iOS 18 launched.
The numbers shifted immediately.
| Year | Monthly active RCS users |
|---|---|
| 2022 | ~421 million |
| 2023 | ~930 million |
| 2024 | ~1.1 billion |
| 2025 (projected) | 2.4–2.8 billion |
| 2028 (projected) | 3.6 billion |
RCS Monthly Active Users
millions — dashed line indicates projected figures
*(Source: Juniper Research, GSMA, Infobip Messaging Trends Report)*
After iOS 18 launched, Infobip's data showed a fivefold (500%) increase in global RCS traffic year-over-year. North America alone saw a 14x surge. These aren't projections — they're observed traffic numbers from a major messaging platform in the months immediately following the iOS 18 release.
For A2P RCS (business messaging), the channel is now live in 57 countries, running on Google's RCS for Business platform. Omdia projects A2P and P2P RCS traffic will grow from ~150 billion messages in 2024 to over 450 billion by 2029, generating $4.2 billion in annual A2P revenue.
Why this matters for AI agents specifically
AI agents that communicate with humans today are doing it over SMS. That means:
RCS changes every one of these.
Suggested replies and action chips mean an agent can send a confirmation message with tap targets: "Confirm" and "Reschedule." The human doesn't have to type anything — they tap. Completion rates for structured interactions go up. Misinterpretations go down.
Read receipts give the agent a signal it currently doesn't have. If a message was delivered but not read after 24 hours, the agent can decide whether to follow up. If it was read and not replied to, that's a different signal. Today, agents are flying blind.
Verified sender identity is significant for trust. One of the core problems with AI agents communicating via phone is that humans have been trained by spam and robocalls to distrust unknown numbers. A verified RCS business profile shows the company name and logo, verified by the carrier. That's a fundamentally different trust posture than a 10-digit number from an area code the recipient doesn't recognize.
Rich media matters for specific agent categories. An AI real-estate agent can send a property card with photo, address, and viewing time. An AI scheduler can send a calendar invite directly in the message thread. An AI support agent can send a screenshot of the resolution steps. None of this is possible over SMS.
The part nobody talks about: A2P RCS is still complicated
Here's where the RCS story gets less clean.
Person-to-person RCS — an iPhone user texting an Android user — works automatically in iOS 18. There's nothing to set up.
A2P RCS — a business or application sending RCS messages to users — is a different process. It goes through Google's RCS for Business (RBM) platform and requires:
This is the same category of problem as 10DLC — a registration and approval process, carrier dependencies, fragmented tooling, and compliance overhead that most developers aren't equipped to navigate alone. The underlying channel is better, but the access path is not simpler.
What the gap looks like today
An AI agent built today sends SMS. It can't send RCS because:
1. The developer doesn't have a registered RCS brand agent in the Google RBM console
2. The SDK or platform they're using (Twilio, custom, etc.) may not have a clean RCS abstraction
3. Fallback to SMS when RCS isn't available needs to be handled explicitly
4. The richer capabilities (action chips, read receipts, suggested replies) need new interaction patterns that most agent frameworks don't support
The result is that the channel upgrade that Apple just made available to 1+ billion devices is sitting unused by the vast majority of AI agents in production.
What "RCS-native" agent communication looks like
An agent built on a communication layer that abstracts RCS properly can do things that feel qualitatively different from current agent interactions.
Instead of:
Agent: "Your appointment is confirmed for Thursday at 2pm.
Reply 1 to confirm or 2 to cancel."An RCS-native agent sends:
That's a fundamentally different interaction quality — one that humans associate with high-trust, well-resourced services.
The channel is ready. The addressable audience is now over a billion users and growing toward 2.7 billion by end of 2025. The remaining work is on the infrastructure side: clean abstractions that let agents send RCS without the developer managing brand registration, fallback logic, and carrier compliance separately.
That work is what separates "RCS is technically available" from "agents can actually use it."