We get asked some version of "so is this an agent or a bot" almost every week, usually by someone who's been through three vendor calls that each used the words differently. It's not a pedantic distinction. Buying the wrong one of these three things is one of the most common expensive mistakes we see teams make before they come to us.
What a chatbot actually is
A chatbot is fundamentally a conversation interface. It answers questions, usually by retrieving information from your documentation or support history and phrasing it clearly. The good ones know what they don't know and hand off cleanly. What a chatbot generally doesn't do on its own is take action in another system — it can tell a customer their order shipped, but it isn't the thing that reroutes the shipment.
That's not a limitation so much as a scope. Chatbots are the right tool when the job is genuinely informational: pre-sales questions, policy lookups, the first line of support triage.
What an RPA bot actually is
RPA stands for robotic process automation, and the "robotic" part is accurate — it's rule-based, not reasoning-based. An RPA bot operates an existing interface exactly the way a person would: click here, copy this field, paste it there, next record. It's deterministic. Given the same input, it does the same thing every time, and it has no real ability to handle a situation nobody explicitly programmed for.
That makes RPA excellent for high-volume, stable, repetitive work — especially against legacy systems with no API — and a poor fit for anything that requires judgment or varies meaningfully from one instance to the next.
What an AI agent actually is
A custom AI agent is the one built to handle ambiguity. It reasons about a situation, decides what action to take from a defined set of options, executes that action across one or more systems, and recognizes when a case is outside what it should handle alone — at which point it hands off to a person with context attached, rather than failing or guessing.
The trade-off is that agents are more work to build responsibly and need real guardrails, because unlike RPA, they aren't fully deterministic. You're not buying certainty that it does the exact same thing every time — you're buying judgment within boundaries you define.
A quick way to tell which one you actually need
Ask one question: does this task ever require handling a case nobody explicitly wrote a rule for? If the honest answer is no — the process is stable, well-defined, and the same every time — RPA is usually the cheaper, more reliable answer. If the honest answer is "sometimes, and when it does, someone has to think about it," that's agent territory. If the job is mostly answering questions rather than taking action, it's a chatbot, possibly with an agent underneath it for the cases that need to go further.
Most real operations end up using a mix of all three. The mistake isn't picking one of these technologies — it's picking the wrong one for a specific task because a vendor used the exciting word instead of the accurate one.