An AI agent and a chatbot are not the same purchase — here is the test that tells them apart
· 6 min read · Mona Technologies
Every vendor call now opens with the word 'agent'. Most of what gets demoed is still a chatbot: a language model that reads your message and writes a reply. That is not a criticism — a good chatbot is a legitimate product — but it is a different purchase from an agent, and buying the wrong one shows up six months later as a project that never left the pilot stage.
The test: does it act, or does it only answer?
A chatbot takes an input and produces text. Ask it to check an order status and, at best, it tells you where to look. An agent is a runtime: the model sits inside a loop with tools, memory and a planner. It decides what to do, calls a tool — look up the order, write to the CRM, trigger a refund — reads the result, and decides again. The output of an agent is a changed state in some system, not just a sentence.
- Chatbot: answers a question from a knowledge base or script
- Agent: reasons about a goal, chains multiple tool calls together, and adapts when a step fails
- Chatbot: one turn, one response, human does the rest
- Agent: multi-step, with an explicit escalation path back to a human when it hits its limits
Why the distinction is not pedantry
Gartner's assessment of the market is blunt: of the thousands of vendors marketing agentic AI, it estimates only around 130 offer something genuinely agentic — the rest is a retrieval tool or a scripted flow with a conversational skin on top. If you are buying on the promise of automation and get a chatbot, the ROI case collapses the moment someone asks it to actually do the task rather than describe it.
“Over 40% of agentic AI projects will be canceled by the end of 2027, due to escalating costs, unclear business value or inadequate risk controls.”
Where each one actually earns its cost
A chatbot is the right tool for a narrow, high-volume, low-risk question set: opening hours, order status lookup, basic troubleshooting from a fixed script. It is cheap to build, cheap to run, and predictable — the same input reliably gives the same class of answer.
An agent earns its cost where the task has steps, the steps touch real systems, and the value is in the action completing without a person doing it manually — processing a return, reconciling an invoice against a PO, qualifying and routing a lead. Gartner projects that agentic AI will autonomously resolve 80% of common customer service issues without human intervention by 2029, with a corresponding 30% cut in operational cost for the issues it handles. That number describes agents doing the resolving, not chatbots doing the describing.
“By 2029, agentic AI will autonomously resolve 80% of common customer service issues without human intervention, leading to a 30% reduction in operational costs.”
How to check what you are actually being sold
- Ask to see the tool calls, not the chat transcript — an agent's log shows systems being read and written to, in sequence
- Ask what happens when a tool call fails — a chatbot has no answer; an agent has a retry, a fallback or an escalation rule
- Ask whether it can complete a task with zero messages from a human mid-way — if a person has to relay information between steps, it is a chatbot with a wrapper
- Ask for the cost per completed task, not per conversation — agents that call tools and hold context cost more per interaction than a scripted chatbot, and that number should be known up front, not discovered on the first invoice
The short version
Buy a chatbot when the job is answering a question well. Buy an agent when the job is completing a task across systems without a person carrying state between the steps. Most failed 'AI agent' projects were never agents — they were chatbots asked to do a job only an agent can do, and nobody checked which one was actually being built.