Two customers, two hundred customers: how a WhatsApp AI agent handles simultaneous conversations, and where the limits are
A person can hold one WhatsApp conversation at a time, which is why Saturday mornings feel like drowning. The natural question about an agent is whether it has the same limit. It does not, but it has different ones, and knowing them is the difference between a calm peak and a confused one.
Quick answer
Every conversation is handled on its own: a customer's message is read with that chat's history and your knowledge, and the reply is written and sent independently of every other chat. Fifty customers at once get fifty separate replies within seconds of each other. The limits are not simultaneity but pacing: each reply waits 3 to 12 seconds to look human, messages from the same person within six seconds are answered together, and each number is capped at 30 replies an hour and 200 a day. A person sees all of it in one shared inbox and can take over any chat without affecting the others. ReplyKit runs this way on every plan.
Each conversation is its own thread
The agent does not have a queue of customers waiting their turn. When a message arrives, the system looks up that customer's conversation, reads the recent history of that chat only, retrieves the relevant parts of your knowledge, writes a reply, and sends it. Another customer's message arriving at the same moment goes through the same steps separately. Nothing from one chat leaks into another; the isolation is per conversation and per account, as explained in is an AI agent safe for customer data.
The six-second batch
People type in bursts: "hi", "got slot tmr?", "afternoon". If the agent answered each line, it would interrupt itself three times. So it waits six seconds after each message from the same person and answers everything that arrived in that window as one reply. The customer experiences a thoughtful answer to their whole question. This is why the reply count on your meter is lower than the message count in the chat; the counting is explained in the cost guide.
Pacing and the per-number caps
- Human pacing. Each reply waits 3 to 12 seconds, with the typing indicator shown. Fifty customers still all get a reply within about fifteen seconds.
- 30 replies an hour, 200 a day per number. A hard cap that protects the number, because ReplyKit pairs by QR code rather than the official API; the reasons are in our ban guide. New numbers warm up at 60 then 120 a day.
- When a cap is hit. The agent pauses on that number, new chats are marked "Needs you" in the inbox, and you get one email. Nobody is ignored; they wait for a person or for the cap to reset.
- More volume. A second number on the Growth plan doubles the ceiling; the daily limits guide covers planning for a peak.
What a person sees
One inbox with every conversation, marked by who is handling it: the agent, waiting for a person, or taken over. Replying in any chat pauses the agent on that chat only; the other forty-nine continue. Handing back is one tap. On a peak day the human's job is the "Needs you" column, not the whole inbox. The routine is in AI-to-human handover.
The viral day
If a post goes viral and 800 people message one number in a day, the agent answers the first 200 (fewer during warm-up), then pauses and the rest wait for a person. That is the honest ceiling of one paired number. Businesses that expect peaks pair extra numbers in advance and put different numbers on different posts, which also tells them which post worked. The planning is in seasonal businesses and peak days.
Frequently asked questions
Can the agent talk to many customers at the same time?
Yes. Each conversation is handled separately and in parallel; fifty customers get fifty replies within seconds of each other.
Why does it wait a few seconds before replying?
It waits six seconds to batch rapid messages from the same person into one reply, then 3 to 12 seconds to look human. Both are deliberate.
Is there a limit on how many it can handle?
Per number: 30 replies an hour and 200 a day, with a warm-up for new numbers. Beyond that the agent pauses and chats wait for a person. Add numbers for more.
Can information leak between two customers' chats?
No. Each conversation is read and answered on its own, and accounts are isolated from each other.
What happens if I reply in one chat during a busy period?
The agent pauses in that chat only. Every other conversation continues.
Every customer answered, none at the expense of another
Separate conversations, one inbox, limits that keep the number safe. 7-day free trial, no card.
Sources and further reading
- Harvard Business Review, The Short Life of Online Sales Leads (Oldroyd, McElheran, Elkington, 2011) · firms that responded within an hour were about seven times more likely to qualify the lead
- ReplyKit, Will WhatsApp ban my number for automation? · our own account of the risk and the limits we enforce
ReplyKit is an independent product and is not affiliated with or endorsed by Meta or WhatsApp. This article is general guidance for running a small business, not legal, financial or regulatory advice.