It is 10:47 PM. Moshe, the owner of an online store in Petah Tikva, sends a message to Danny, a customer from this morning:
“Hi Danny, I saw that you still have not received your package. I arranged an emergency courier for tomorrow at 9:00 AM. Please confirm.”
Danny replies two hours later: “Perfect, thank you!”
Four seconds later, Tuki AI jumps in:
“Hello! I am Tuki AI, the smart sales agent for this store. How can I help?”
Moshe yelled at the screen.
This is not a bug. This is an AI sales agent that was designed a little too well.
The problem: when the system follows the rules, but the rules are not enough
Tuki AI was built for one clear purpose: detect an incoming customer message, understand the intent, and respond. 24/7, in natural language, without getting tired. That is exactly why online store owners add it to their website, WhatsApp, WooCommerce store, Shopify store, or WordPress based shop.
They want an AI sales agent for online stores that answers customers even when the team is not available.
But there is one type of conversation that general automation rules do not handle well: conversations started by the business.
When a customer opens a conversation, the intent is clear. The customer needs help, and Tuki AI has a role.
When the business opens the conversation, the context is different. It might be a price negotiation, an apology for a delayed shipment, a sensitive order, or a personal follow up from a human representative. When the customer replies, Tuki AI sees an incoming message and starts working. Suddenly, the human agent is competing with the AI assistant inside the same conversation.
The question we had to solve was not: how do we make Tuki AI answer better?
The real question was: how does an AI sales agent know when not to answer at all?
The ideas we rejected
On the way to the solution, we tested four directions. Three of them failed for the same reason: they required the human agent to remember something during the workday.
Any solution that depends on human memory during a busy day eventually fails in real life.
1. A manual button: “I am taking over this conversation”
The representative clicks a button before every conversation they start. Simple in theory. In practice, it fails exactly when it matters most: during pressure, between tasks, or at the end of a long day.
2. Detecting WhatsApp replies
WhatsApp webhooks can show when a message is a direct reply to a previous message. It sounds elegant. But most customers do not use the reply feature. They simply send a normal message.
3. Using AI to detect a personal tone
We considered asking a language model to detect sensitive or personal conversations and decide when to stay silent. We rejected it because it breaks two important rules: it is expensive in tokens, and it is not deterministic. You cannot promise a business consistent behavior if the system makes a new judgment every time.
4. Automatic timeout plus a secret code
This is what worked. Two modes. Zero manual action by default. Manual control only when the representative needs more time.
The solution: two modes and a quiet handover
Regular mode: 3 days of silence
When a human representative sends the first message to a customer who did not contact the business first, the system automatically detects it and moves the conversation into human agent mode. Tuki AI stays silent.
Every new message from either side resets the timer for another 3 days. After 3 full days with no communication, Tuki AI returns to normal activity automatically.
This covers most real business cases. In most conversations started by an online store, 3 days is long enough for the representative to handle the customer without the AI sales agent interrupting.
Extended mode: 14 days with a secret code
For the remaining cases, such as long negotiations, complex orders, sensitive support cases, or high value sales, we added a manual escape route.
The representative adds ### at the beginning of the message. The system detects the code, removes it before sending, and opens a quiet window of 14 days.
Example. The representative writes:
### Hi Joseph, following our conversation yesterday about the large order...
The customer receives:
“Hi Joseph, following our conversation yesterday about the large order…”
That is it. No settings. No menus. No complex workflow. Three characters at the beginning of a message give the human representative two weeks of silence from the AI assistant.
What the representative sees in the dashboard
Inside the Tuki AI conversation dashboard, the representative sees a clear blue banner:
“This conversation is controlled by a human agent. Tuki AI will return automatically on May 20 at 2:30 PM.”
No guessing. If the representative wants to bring the bot back earlier, one click on “Activate bot” inside the conversation restores Tuki AI immediately.
Quick usage guide
| Situation | What to do | Result |
|---|---|---|
| You sent the first message to a customer | Nothing. It happens automatically | 3 days of silence |
| Long or sensitive conversation | Add ### at the beginning of the message |
14 days of silence |
| You want to bring the bot back early | Click “Activate bot” inside the conversation | Immediate return |
| The quiet period ended | Nothing. It happens automatically | Tuki AI returns by itself |
What we learned
The hardest part of this feature was not the code. It was understanding that in an automated system, the most important question is sometimes not when to act. It is when not to act.
An AI sales agent that always answers is simple. An AI sales agent that knows when to step back is a real product.
That is especially important for online stores using Shopify, WooCommerce, WordPress, WhatsApp, and website chat, where the sales process often moves between automation and human relationships.
This is what Tuki AI is designed to do: help online stores sell more, answer faster, and protect the human touch when it matters.
Want to see how it works in your online store?
Start a free Tuki AI pilot: an AI sales agent for WooCommerce, Shopify and WordPress stores. Setup takes 2 minutes, no credit card required, and your customers get natural answers from the first message.
Frequently asked questions
What happens if a representative forgets to add ### in a sensitive conversation?
If the representative sends the first message, regular human agent mode is activated automatically for 3 days. The ### code is needed only when the representative wants to extend the quiet period to 14 days.
Does the customer see the ### code?
No. The system removes the code before the message is sent. The customer only sees the regular message.
How do I know when Tuki AI will start responding again?
Inside every conversation in human agent mode, the dashboard shows a clear banner with the exact date and time when Tuki AI will return.
Can this feature be turned off?
There is usually no need to turn it off. It is activated only when a human representative starts the conversation. In conversations started by customers, Tuki AI continues working normally.
Does this work for Shopify, WooCommerce and WordPress stores?
Yes. Tuki AI is built as an AI sales agent for online stores and can support sales and service workflows for WooCommerce, Shopify, WordPress based stores, WhatsApp, and website chat.
Does it work on Instagram and Facebook Messenger?
The feature currently works on WhatsApp and Web Chat. Expansion to Instagram and Facebook Messenger is planned.
