Someone in a WhatsApp group for Singapore SME owners recently asked if they should "get an AI agent" for their customer service team. Three vendors replied within the hour. All three sent decks. All three used the phrase "AI agent." And if you looked closely at what they were actually selling — two of them were flogging a WhatsApp chatbot with a few canned responses and a hand-off button. The third was selling a slightly fancier version of the same thing, dressed up with a ChatGPT logo and a lot of confidence. This is the problem. The word "agent" has been kidnapped, and SME owners in Singapore are paying the ransom.
A Chatbot Is Not an Agent. Stop Letting Vendors Tell You Otherwise.
Let's be blunt about what a chatbot is. A chatbot takes your input and returns an output. That's it. You type "what are your opening hours," it replies "9am to 6pm." You ask "do you have this in blue," it says "let me connect you to a human." Maybe it's scripted with a decision tree. Maybe it's powered by GPT-4 and sounds remarkably articulate. Doesn't matter. If the system waits for you to say something, responds, then stops — that is a chatbot. A very nice chatbot, perhaps. But a chatbot.
An AI agent is a fundamentally different thing. An agent is autonomous. It doesn't just respond — it acts. It takes a goal, breaks it into steps, uses tools to complete those steps, remembers what it's done, and adapts when something doesn't go to plan. It doesn't wait for you to guide it through every micro-decision. You point it at an outcome and it figures out the path. That distinction — reactive versus autonomous, single-turn versus multi-step, stateless versus stateful — is the difference between a fancy FAQ and an actual operational system.
Here's a concrete example of a chatbot doing its best chatbot impression: A retail company in Toa Payoh deploys a "customer service AI agent" on their website. A customer types in "I want to return my order." The bot asks for the order number. Customer provides it. Bot says "Thank you, our team will be in touch within 2 business days." End of interaction. That is a chatbot. A polite one, sure. But it didn't check the order status, didn't verify if the return window was still valid, didn't raise a return request in the backend system, didn't notify the warehouse. It collected information and stopped. A human still has to do everything that matters.
Now here's what an actual AI agent looks like in the same scenario. Customer says "I want to return my order." The agent pulls the order record from your e-commerce system. Checks the purchase date against your return policy. Verifies the item is eligible. Generates a return authorisation number. Sends the customer a prepaid shipping label via email. Logs the return in your inventory system. Flags it to the warehouse team. Updates your CRM. Done. No human touched it. That's not a chatbot. That's an agent — and as we broke down in our deep-dive on building a 5-agent AI workflow, the real power comes from chaining these capabilities into systems that handle entire operational sequences, not just single interactions.
Why Singapore Vendors Keep Getting Away With This
Because the terminology is new, the buyers are busy, and the demos look impressive. A well-prompted GPT chatbot can sound very, very smart. It can handle nuanced questions, maintain a conversational tone, and give you the feeling that something sophisticated is happening under the hood. Vendors know this. So they add a few buzzwords — "agentic," "autonomous," "multi-modal" — and suddenly a WhatsApp bot with an OpenAI API key becomes an "AI agent solution" with a five-figure price tag.
Singapore SME owners are particularly vulnerable to this because they're genuinely curious about AI, they don't have time to go deep on the technical definitions, and the grant landscape (yes, some of this gets packaged as EDG-eligible — more on that at our EDG page) creates a sense of urgency that clouds due diligence. When someone tells you "this is grant-fundable, but we need to move fast," your critical thinking takes a coffee break.
The result? Businesses spend real money on systems that don't actually reduce workload. The chatbot answers questions. Humans still process everything. Six months later, the business owner is frustrated, the vendor has moved on, and "AI" gets a bad reputation it doesn't deserve. The technology isn't the problem. The mislabelling is.
This is exactly why we've been vocal about the gap between what gets sold and what gets built — we even turned down three projects last quarter because the briefs were asking for theatre, not transformation. If you haven't read that post, it's worth five minutes of your time before your next vendor meeting.
Three Questions That Instantly Expose What You're Actually Buying
You don't need a computer science degree to cut through the noise. You need three questions. Ask them before you sign anything. The answers will tell you everything.
Question one: "Can you show me a live demo where the system takes a multi-step action without a human in the loop?" Not a slide. Not a video. A live demo, with your data or a realistic simulation, where the system autonomously completes more than one task in sequence. A chatbot cannot do this. It will respond to your prompt and wait. An agent will keep moving. If the vendor fumbles, changes the subject, or shows you a Loom recording of someone clicking through a workflow manually — you have your answer.
Question two: "What tools does the system have access to, and how does it decide when to use them?" A real AI agent uses tools — APIs, databases, external systems, communication platforms. It calls the right tool at the right moment based on context. If the vendor can't name specific integrations, or if their answer is essentially "it uses ChatGPT to generate replies," you're looking at a chatbot. Tool-use and decision-making about tool-use is the core of what makes an agent an agent. No tools, no autonomy. No autonomy, no agent.
Question three: "How does the system handle a failure mid-task, and what state does it retain between sessions?" This is the one that separates the serious builders from the demo merchants. Agents have state — they remember what they've done, where they are in a process, and what's pending. If the system crashes halfway through raising a purchase order, a real agent knows it was halfway through raising a purchase order. A chatbot has no memory of the conversation the moment you close the window. Ask what happens when something breaks. Ask how it recovers. If the vendor looks confused, or says "the user would just start again," you're not buying an agent.
These three questions work because they probe the actual technical requirements of agentic systems: multi-step autonomy, tool integration, and persistent state. A chatbot, no matter how well-dressed, cannot honestly answer all three. A real AI agent, built properly, answers them without breaking a sweat. Same team. Bigger output — but only if the system you're buying actually does the work your team shouldn't have to do manually.
If you're not sure what you've been pitched, or you want a second opinion before committing to a build, talk to us. We'll tell you straight — whether what you've been shown is an agent, a chatbot, or just a very expensive autocomplete. No deck required.
