Chatbot vs agent vs RPA
Three things customers conflate. Knowing the difference helps you scope correctly and not over-promise.
The one-line difference
| What it does | Can it decide what to do? | Can it act on systems? | |
|---|---|---|---|
| Chatbot (classic) | Answers from a script or FAQ | No — follows decision trees | Rarely (maybe creates a ticket) |
| RPA (robotic process automation) | Executes a pre-defined sequence on UIs / APIs | No — follows a recorded script | Yes — on whatever it was scripted for |
| Agent (LLM-driven) | Decides what to do based on the situation, then does it | Yes — reasons within constraints | Yes — via tools |
Pick which when
Why agents win for most Shipsy use cases
Logistics operations are messy. The same nominal task — “handle this delivery exception” — looks different every time depending on the order, customer, carrier, location, time of day. Chatbots and RPA collapse under that variability. Agents thrive on it.
That’s why Shipsy ships agents, not chatbots or RPA. See Architecture for how we build them.
What agents are still bad at
- Long-running, fully autonomous workflows with no human checkpoints (drift accumulates)
- Tasks where the cost of being wrong is very high and the agent’s confidence is hard to verify
- Things that don’t actually need reasoning (use the right tool — sometimes that’s RPA)
What CS folks hear from customers, decoded
| What the customer says | What they probably mean |
|---|---|
| ”We need a chatbot" | "We need to handle inbound queries cheaper” — propose an agent if scope is varied |
| ”We’ve already automated this with RPA" | "We’ve scripted the happy path” — look for the exception cases an agent could handle |
| ”Can your AI talk to my systems?" | "Can your agent use tools?” — yes, via MCP |
| ”Can it work alongside our team?" | "Can it escalate?” — yes, human-in-the-loop is built in |
Sources
- See Capabilities for what agents can compose
- Agent catalog for real examples
Changelog
- 26 May 2026: Initial draft.