What does it mean to orchestrate AI agents across SMS, WhatsApp, web chat and voice? Four levels
Four levels of what 'orchestrate' means across SMS, WhatsApp, web chat and voice, how to tell which one your process needs, and seven questions for any vendor.
Iván Itzcovich
Co-founder
A platform orchestrates AI agents across SMS, WhatsApp, web chat and voice when one conversation, one memory of the customer and one set of actions serve all four channels. Many platforms only do the first step, which is connecting the channels. Ask which of four levels a vendor means: shared channels, shared inbox, shared memory or shared actions.
The question keeps showing up in search because the word is doing too much work. Every vendor in this market says orchestration, and they mean different things by it.
What does 'orchestrate' mean when a vendor says it?
There are four levels, and each one includes the one before it. A demo that looks the same at level one and level four is the reason this gets decided too late.
| Level | What it means | What you can check in a demo |
|---|---|---|
| 1. Shared channels | One vendor sells you access to SMS, WhatsApp, web chat and voice | Can you switch a channel on without a new contract? |
| 2. Shared inbox | Every channel lands in one place where your team reads it | Does a person see the SMS and the WhatsApp message in the same thread? |
| 3. Shared memory | The agent knows it is the same customer on every channel | Text from one number, then message from another: does it connect them, and how? |
| 4. Shared actions | The agent does the same work in your systems from any channel | Start a refund by chat and finish it by phone. Is it one refund? |
Which level does your process need?
None of the four is better in general. What decides it is where your process ends.
If it ends when someone answers a question, level one or two is enough. If it ends when a record changes in your systems, a refund issued or a visit booked, only level four closes it, and level three has to hold for the agent to know whose refund it is.
Level one is infrastructure and a good business on its own. Level four is where a process gets resolved instead of answered, which is the difference our earlier post draws between the layers of the market.
Why are levels three and four the hard ones?
Because every channel identifies a person with a different key. A phone number reaches SMS and voice. WhatsApp reaches the customer through their WhatsApp account. A web chat session is just a browser session until the visitor identifies themselves.
Joining those keys is the work. A platform that says it has shared memory should be able to tell you which key it uses on each channel and what happens when two keys point at the same person.
Level four is hard for a different reason. Actions need credentials, permissions and a record of who did what. Doing the same action from four channels means writing it once, in your systems, and letting the channels call it. Writing it four times means four versions drifting apart.
What does each channel's own rulebook add?
The platform has to respect rules it does not control. On WhatsApp, Meta states the one that shapes every integration:
When the window closes, you can only send pre-approved template messages.
The 24-hour window is a WhatsApp rule, and SMS has consent rules of its own that vary by country. An orchestrating agent needs to know which rulebook applies before it sends. We cover the SMS and voice rules in a separate post, and the choice of channel for each kind of conversation in another.
Seven questions to ask any vendor
- Is a customer who texts and then messages on WhatsApp one record or two?
- When the conversation moves from chat to a call, what does the agent know about the earlier messages?
- Where do the actions run: once in your systems, or once per channel?
- Which channel rules does the platform enforce for you, and which does it leave to you?
- How does a handoff to a person work on each channel, and does the context travel?
- Is quality measured per conversation or per channel, and can you see the same customer across both?
- Who owns the numbers and accounts if you leave? The five assets of an agent have different owners.
A vendor who answers all seven without changing the subject has built at least level three.
How do named platforms measure up?
That is a different post. We read the public channel pages of five vendors and of our own product and set them side by side in a comparison of platforms for SMS, WhatsApp, web chat and voice. Use the seven questions above on every row of it.
Frequently asked questions
What does it mean to orchestrate AI agents across channels?
It means one conversation, one memory of the customer and one set of actions serve every channel. A platform that only connects SMS, WhatsApp, web chat and voice to the same dashboard has solved the first of four levels. The harder levels are recognising the same person on every channel and running the same actions in your systems whichever channel started the request.
Which of the four levels does my process need?
It depends on where the process ends. If it ends when someone answers a question, shared channels or a shared inbox can be enough. If it ends when a record changes in your systems, such as a refund or a booking, you need shared actions, and shared memory so the agent knows it is the same customer on every channel.
Is a shared inbox the same as shared memory?
No. A shared inbox puts every channel's messages where your team can read them. Shared memory means the agent itself knows the SMS and the WhatsApp message came from the same customer. That needs a rule for joining the identifiers each channel uses, and a vendor should be able to tell you what that rule is.
Iván Itzcovich · Co-founder, StudioChat
Want agents like these working for your team?
Talk to us now