Called an AI chatbot, built like a script: what a real AI employee does
Open almost any WhatsApp tool’s website today and you will see the letters A and I. Then try the demo. Too often, the “AI” turns out to be a menu: Reply 1 for pricing, 2 for location, 3 to talk to us.
This post is the evidence piece. Our concept piece explains the difference between a chatbot and an AI employee. Here we show what misleading “AI” looks like in the market, without naming anyone, so you can recognise it yourself.
This isn’t just us saying it
In June 2025, research firm Gartner warned about “agent washing”: vendors rebranding existing products, such as AI assistants, robotic process automation and chatbots, “without substantial agentic capabilities”. Gartner estimated that only about 130 of the thousands of agentic AI vendors are real.
The US Consumer Financial Protection Bureau, reviewing bank chatbots in 2023, described rule-based bots as using decision trees or keyword databases to trigger preset, limited responses, and documented customers stuck in “doom loops” with no route to a human.
Different markets, same pattern. Here is how it tends to show up for Malaysian SMEs.
Six patterns to recognise
1. The numbered menu with an AI badge
The customer types a real question. The bot replies with a list: “1. Pricing 2. Location 3. Promotions”. Whatever they type next, they get the list again. The brochure says “AI-powered”. The bot is a phone-tree in chat form.
2. Keyword triggers dressed as understanding
The message contains “price”, so the bot sends the price list. But “how much for 3 pax ah?”, “brp harga?” or “价钱多少” don’t contain the keyword, so the customer gets “Sorry, I don’t understand.” Real Malaysian messages mix BM, English and Chinese, with typos and short forms. Keyword lists break on exactly that.
3. The flow builder renamed “AI agent”
You are shown a drag-and-drop canvas of boxes and arrows, and told it is your “AI agent”. Every branch has to be drawn by hand. Ask a question nobody drew a branch for, and the agent has nothing to say. That is a decision tree, the thing Gartner’s “agent washing” warning describes.
4. “AI” that only helps your staff type
The AI turns out to be a button in the inbox that drafts a reply for your staff. Useful, honestly. But customers at 11pm still get the menu, because nobody is there to press the button. If the AI never talks to customers, it isn’t covering your customer service.
5. The loop with no exit
The customer types “agent”, “human”, “nak cakap dengan orang”. The bot shows the menu again. Or it says “please WhatsApp 01X-XXX XXXX”, starting the conversation over on another number. This is the CFPB’s doom loop, and it is where customers give up.
6. Confident answers from nowhere
The opposite failure. Some products do plug in a large language model, but don’t tie it to the business’s facts or tell it when to stop. It sounds fluent and happily invents a promo, a delivery date or a refund policy. Customers treat that as your word. We cover why in What an AI CS employee should never invent.
Menus aren’t the enemy
To be fair: buttons, menus and flows are useful. ThinkChat has them too. Guiding a customer step by step through a booking is good design. The problem is when a script is all there is, and it is sold as intelligence.
What a real AI employee does
- Understands what people actually say. Free text, mixed BM, English and Chinese, typos, voice notes and photos, answered from your own knowledge.
- Answers honestly. When the answer isn’t in your knowledge, it says so instead of inventing. See When “I don’t know” is the right answer.
- Keeps its place. Mid-booking, a customer asks “parking ada?” It answers, then continues the booking.
- Hands off with context. When a person is needed, your team takes over in the same chat, with the history in front of them. See Handoff to a human without losing the chat.
Test any vendor in 5 minutes
Open their demo on your phone and try this:
- Type like a real customer. “hi boss ada slot sabtu x? 2 org. how much ah” If you get a menu or “I don’t understand”, it’s a script.
- Ask something it can’t know. “Do you sell gift cards?” A real AI employee admits it doesn’t have that answer. A bad one invents.
- Change topic mid-flow. Start booking, then ask about parking. Does it answer and carry on, or reset?
- Ask for a person. Type “nak cakap dengan orang”. Does a human path appear in the same chat?
- Ask the vendor two questions. “What happens after it misses an answer twice?” and “Can I see a record of every conversation?”
Bonus: send a voice note. Customers do.
How ThinkChat measures up
ThinkChat answers from your own Q&A, documents and website in English, Chinese and Malay, transcribes voice notes and describes images before replying. It follows the paths you draw in AI Flows and answers side questions without losing its place. When it doesn’t know, it says so; after a second miss in a row it can bring in your team, who take over in the same inbox. Every conversation stays in the Conversation Ledger.
Don’t take our word for it either. Run the five-minute test on us.
Soft next step
If you want to hire a ThinkChat AI CS employee, Talk to Lyia on WhatsApp: +60 11-5515 9318.
Plans stay simple: Junior RM299 · Senior RM599 · Manager RM899 a month. Pick the level that matches how much of your customer service you want covered.