A WhatsApp multiple agents setup is a model in which a team answers customers from the same number through a shared inbox. In 2026, WhatsApp with multiple agents works best on the WhatsApp Business API, connected to a platform where AI agents and human agents share the same queue, with full history.
This guide covers the app’s limits, the available paths, coexistence, AI working alongside humans, organization, metrics, data protection, and costs.
Key Takeaways
- The WhatsApp Business app was designed for one person at a time; a shared inbox on the official API is what allows multiple agents and AI on the same number.
- With coexistence, the business keeps its phone and history and starts serving customers as a team, with AI agents, through the API.
- From October 1, 2026, Meta charges for API service messages above 1,000 free per phone number per month, which weighs on high-volume teams.
What Is a WhatsApp Multiple Agents Setup?
The idea is simple: the customer talks to a single number, and on the other side several people and AI agents take turns without the customer noticing the switch. What changes is how conversations arrive, are split up, and are recorded.
One Number, One Shared Inbox
In practice, every incoming message lands in a single list, visible to the team according to each person’s permissions. Each conversation has an owner, a status, and a complete history, which prevents duplicate replies and forgotten customers.
This model depends on a software layer between WhatsApp and the team. That layer receives messages, distributes them among agents, keeps the context, and sends replies through the company’s official number.
Why the Regular App Does Not Work for Multiple Agents
According to WhatsApp Business itself, the app is free, aimed at small businesses, supports one number per account, and was not built for teams with many users. The API-based platform, in Meta’s words, supports thousands of agents and bots.
In the app, whoever opens the conversation first replies, and there is no assignment, queue, or clear record of who said what. With few messages, that works. With higher volume, it leads to conflicts, delays, and lost opportunities.
Where Artificial Intelligence Comes In
In 2026, the most common question is no longer just how to split a number among people. The question now is which platform brings together a WhatsApp inbox with multiple agents and AI in one place, without switching systems at each step.
In this design, the AI agent answers frequent questions, qualifies leads, and books appointments, while humans take on complex cases. The topic is covered in depth in the guide to human and AI customer service.
Paths to Team-Based Service on WhatsApp
There are basically two legitimate paths for putting more than one person on the same number. Each has clear limits, and the choice depends on volume, use of automation, and the need for control.
WhatsApp Web on Several Devices and Its Limits
The first path is to connect WhatsApp Business to a few linked devices, such as computers running WhatsApp Web. The number of devices is limited, and all of them see the same conversations at the same time, with no split by person.
This fixes the improvisation of a small team, but it offers no round-robin assignment, internal notes, per-agent reports, or integration with AI agents. Extensions and tools that imitate WhatsApp Web fall into the unofficial WhatsApp API category, which violates WhatsApp’s terms.
Official API With a Shared Inbox
The second path is to register the number on the Cloud API and use a platform with a shared inbox. All incoming messages arrive by webhook, as described in Meta’s documentation, and the platform distributes them among the team.
This is where automatic assignment, permissions, reports, approved templates, and AI agents connected to the CRM come in. The risk of a ban is lower than with unofficial solutions, although it still depends on following Meta’s policies.
Comparing the Paths
The table below sums up the differences that matter most for businesses that need multiple agents.
| Criterion | App with linked devices | Official API with shared inbox |
|---|---|---|
| Designed for teams | No, according to Meta | Yes, with multiple agents and bots |
| Assignment and ownership | Manual | Automatic, rule-based |
| Integrated AI agents | No | Yes, with handoff to humans |
| Per-agent reports | No | Yes, depending on the platform |
| Message costs | Free | Meta’s price list, per message |
Coexistence: App and Shared Inbox on the Same Number
For years, moving to the API meant giving up the app and the history. WhatsApp coexistence changed that by letting the same number work in the app and on the API at the same time.
How Coexistence Works in Practice
According to Meta’s coexistence documentation, the business connects its WhatsApp Business app account to the Cloud API through a partner. It sends at volume through the API, keeps chatting one-to-one in the app, and keeps the history in sync.
The sync brings in messages from the last 180 days and the contacts. For businesses already serving customers from the phone, this makes it possible to adopt a shared inbox and AI without changing numbers or losing past conversations.
What Stays Outside the API
Some features remain app-only. Groups, voice and video calls, the catalog, orders, status, and app labels do not work on the Cloud API, and broadcast lists become read-only.
That is why the team needs to agree on where each thing happens. A common rule is to keep groups and calls on the phone and concentrate one-to-one service in the shared inbox, where there is logging, assignment, and an AI agent.
Linked Devices and the Inactivity Rule
Under coexistence, Meta supports up to four compatible linked devices, except WhatsApp for Windows and Wear OS. On activation, all devices are disconnected, and compatible ones can be linked again.
The primary phone must stay in use: after about 14 days without use, the API connection drops. Linked devices idle for 30 days are also removed. It is worth adding this check to the routine of whoever owns the number.
How AI and Human Agents Share the Queue
The real gain of WhatsApp with multiple agents appears when AI and people work in the same inbox. The customer does not have to repeat information, and each conversation goes to whoever can best resolve it.
What the AI Agent Handles on Its Own
AI agents typically handle first contact, frequent questions, lead qualification, and scheduling. When they are connected to the company’s systems, as shown in the post on AI agent CRM integration, they look up data and record information without human intervention.
While the agent is handling a conversation, it does not need to enter the human queue. This frees the team for cases that require negotiation, empathy, or a decision, and keeps service running outside business hours.
Handoff to a Human With Context
The WhatsApp Business Messaging Policy allows automation, but requires prompt, clear, and direct paths for the customer to talk to a person. Every AI on WhatsApp needs a visible exit to human service.
A good handoff carries a summary of the conversation, the reason for the transfer, and the data already collected. That way, the agent starts from the right point, and the customer does not feel sent back to the end of the line.
Taking Over and Handing Back the Conversation
Sometimes a person needs to step into the AI agent’s conversation, for example to close a sale. Ideally, they can take over with one click, pausing the agent in that conversation.
Once the human part is done, the conversation can go back to the agent for simple tasks, such as confirming a time or sending a link. This back-and-forth should be logged, so metrics can separate what the AI and the team resolved.
Organizing the Shared Inbox
A shared inbox without rules turns into a disorganized group chat. Organization defines who sees what, who replies, and how fast, following the same WhatsApp customer service best practices used on other channels.
Queues, Labels, and Assignment
Queues separate topics or numbers, such as sales, support, and billing. Labels mark the stage or type of request. Automatic assignment gives each new conversation to an available agent, without anyone having to choose.
- Round-robin: rotates conversations among available agents, balancing the workload.
- Per-person limit: keeps one agent from piling up too many conversations at once.
- By number or queue: restricts each conversation to the group that handles that topic.
Internal Notes and Canned Responses
Internal notes sit inside the conversation but never reach the customer. They pass context between colleagues, record agreements, and guide whoever takes over next, without leaving the history.
Canned responses store frequent texts, such as hours, addresses, and return policies, triggered by a shortcut. They standardize the brand’s tone and reduce errors, but they should be reviewed often so they do not go out of date.
Wait Time, SLA, and Business Hours
An SLA defines how quickly each conversation must get a reply. To meet it, the inbox needs to show how long the customer has been waiting and allow sorting by longest wait instead of by most recent message.
After hours, an away message says when the team will be back, and the AI agent can keep handling simple requests. Conversations that depend on a future follow-up can be snoozed to reappear on the right date.
Metrics for WhatsApp Multiple Agents Teams
Without numbers, the team cannot tell whether it needs more people, more automation, or more training. Customer service metrics show where the bottleneck is and whether the AI is actually helping.
First Response Time and Resolution Time
First response time measures how long the customer waits until someone, a person or the AI, replies. Look at the median and at the slowest cases, because the average hides customers who waited hours.
Time to resolution and the reopen rate show service quality. Conversations resolved quickly but reopened soon after point to incomplete answers, not real efficiency.
Per-Agent Productivity and AI Performance
Per-agent reports show volume, response time, and resolved conversations for each person. They should be used to balance the workload and train the team, not only to hold individuals accountable.
For the AI, the core indicators are conversations resolved without a human and the handoff rate. A satisfaction survey at the end of the conversation completes the picture with the customer’s own opinion.
The 24-Hour Window, Opt-In, and Data Protection
Serving customers as a team through the API means following Meta’s rules and data protection law. These precautions apply to every agent and to AI agents as well.
The 24-Hour Window and Templates
According to Meta, when a customer sends a message, a 24-hour customer service window opens, and it renews with each new message from them. While the window is open, the team replies freely, including with media, buttons, and lists.
Once the window closes, only an approved WhatsApp message template can be sent. The shared inbox should show the window status, so agents do not try to send free text to someone who went quiet days ago.
Opt-In, Opt-Out, and Number Quality
Meta’s policy only allows contacting people who provided their number and gave WhatsApp opt-in. Opt-out requests must be honored by everyone, which requires the platform to block new sends after the request.
Unwanted messages lower the WhatsApp quality rating. According to the opt-in guidelines, a prolonged low rating limits sending, and the policy enforcement rules provide for temporary blocks in case of violation.
Data Protection, Access, and Retention
Under Brazil’s data protection law (LGPD, in Portuguese), the controller must be able to prove consent when that is the legal basis (Article 8, §2), and individuals can request deletion of their data (Article 18). Similar rules apply under the GDPR in the EU. With multiple agents, access control carries more weight.
Good practices include limiting each person to the numbers and conversations they work on, defining how long messages are kept, and logging who did what. Legal questions should go to the legal team or the data protection officer.
Official API Costs for Team-Based Service
The cost of team-based service has two parts: the service platform and the messages Meta charges for. From October 2026, the second part weighs more for businesses that reply to many customers.
Charges From October 1, 2026
Meta charges per delivered message, based on the category and the recipient’s country, as shown on its pricing page. From October 1, 2026, according to the page on non-template messages, Meta charges for service messages at the market’s utility and authentication rate, with 1,000 free service messages delivered per phone number per month. In Brazil, for example, that is R$ 0.035 per delivered message, while marketing costs R$ 0.3217.
From the same date, Meta also charges for utility templates sent inside the window. Under coexistence, messages sent from the app are not charged, but they also do not open or extend windows on the API. The WhatsApp API pricing guide details the model.
How to Estimate Operating Costs
The calculation includes marketing, utility outside the window, and authentication. From October onward, it adds service messages above each number’s free allowance and utility templates inside the window.
Conversations that start from click-to-WhatsApp ads can open a free 72-hour window, which lowers the cost of campaigns that lead into service.
For Brazilian numbers, scenarios can be simulated in the cost calculator (Brazil, in Portuguese). Terms such as WABA and customer service window are explained in the WhatsApp API glossary.
What to Look For in a Shared Inbox Platform
A platform for team-based service should run on the official WhatsApp API, support coexistence, and let human agents and WhatsApp AI tools share the same conversations.
Shared Inbox, Round-Robin, and Access Control
Good automatic assignment takes each agent’s presence into account: it gives the new conversation to whoever is available and has gone longest without one, within a limit of open conversations. Each number can have its own list of agents.
With restricted visibility, each team member sees only the numbers they work on and the conversations assigned to them. Hours, greetings, and away messages should be configurable per number, and conversations should be easy to snooze until a set date.
AI and Humans on the Same Screen
When the AI agent hands off, the conversation should move to a “waiting for human” lane with a summary in an internal note. The agent can take over at any time, pausing the AI, and hand the conversation back later.
The screen should show the 24-hour window and offer templates when it closes, along with internal notes, canned responses, media, buttons, and lists. A report should bring together first response time, resolutions, AI handoffs, and per-agent performance.
How ConverZap Handles Team-Based Service
ConverZap works on the official WhatsApp API, supports coexistence, and has its own inbox where human agents and AI agents share the same conversations, with the features described above. Each one is explained in the team inbox tutorial in the help center.
Best Practices and Common Mistakes
No tool makes up for poorly defined processes. A few simple practices make the difference between an organized shared inbox and a new source of complaints.
Best Practices to Get Started
- Map the topics: separate what the AI resolves, what goes to sales, and what goes to support.
- Set realistic SLAs: by queue and by time of day, with a clear away message.
- Train the team: internal notes, canned responses, and how to take over from the AI.
- Review metrics every week: first response, reopens, and handoffs.
It also pays to connect the inbox to the CRM, so sales and service see the same customer history.
Common Mistakes That Stall Service
The costliest mistake is using unofficial extensions to split the number, which can lead to a banned WhatsApp number. Another is deploying AI without an exit to a human, which goes against Meta’s policy and irritates customers.
Other common mistakes: giving full access to every agent, ignoring the 24-hour window and opt-in, and not tracking the cost of service messages from October 2026.
In short, WhatsApp with multiple agents in 2026 combines the official API, coexistence when it makes sense, AI on the front line, and humans for cases that require judgment. To set up this operation, talk to the ConverZap team.
WhatsApp Multiple Agents FAQ
Look for a platform that uses the WhatsApp Business API, has a shared inbox with automatic assignment, and lets AI agents hand conversations to humans with context. Round-robin assignment, per-number access control, and AI agents on the same number are the key features to compare.
In the app, the limit is the number of linked devices, and under coexistence Meta supports up to four. On the official API, the number of agents depends on the platform, not on devices linked to the phone, since Meta describes the API as able to support thousands of agents and bots.
No. Everyone replies from the company’s single number, and the platform records who handled each conversation. To separate departments, a business can have more than one number in the same account: a portfolio starts with up to 2 numbers and can reach 20 after verification.
Yes. With coexistence, the number stays in the app and also works on the API, with up to 180 days of history synced. The other option, deleting the app account and registering the number on the API, means the message history is lost, according to Meta.
They should. The WhatsApp Business Messaging Policy requires clear paths to talk to a person, and announcing the handoff avoids confusion. Under Brazil’s LGPD, individuals can also request a review of decisions made solely by automated means that affect their interests, and similar rules apply under the GDPR in the EU.
The cost adds up the service platform and the messages Meta charges for. From October 1, 2026, Meta charges for service messages above 1,000 free per phone number per month. For Brazilian numbers, the cost calculator (Brazil, in Portuguese) helps simulate scenarios.
