AI agents vs chatbots
By Matt Doughty, CEO & Co-Founder of Prefactor ·
A chatbot converses; an agent acts. Both may sit behind the same chat window and the same model, which is why the words blur — but a chatbot's output is a message for a human to act on, while an agent's output is the action itself: tool calls, records written, work completed. The window dressing is shared; the operational stakes are not.
| Dimension | AI agents | Chatbots |
|---|---|---|
| Output | Actions through tools — writes, calls, decisions | Messages — answers, drafts, suggestions |
| Who acts on it | The software itself | The human reading it |
| Shape of work | Multi-step loop toward a goal | Turn-by-turn exchange |
| Credentials needed | Its own identity and scoped system access | Often none beyond the knowledge source |
| Failure cost | Wrong actions on real systems, compounding | A wrong message, caught by the reader |
| Governance class | Operational actor: registry, gates, audit trail | Content tool: output review, usage policy |
The verdict
Classify by what happens to the output, not by the interface or the marketing. If a human stands between the response and any consequence, you have a chatbot — govern it as a content tool and invest in answer quality. The moment the system can write to anything — file the ticket, issue the refund, change the record — it has become an agent regardless of what the vendor calls it, and it needs the agent treatment: [its own identity, scoped permissions, gates on consequential writes, and an audit trail](/guides/secure-agentic-ai). The dangerous deployments are the in-between ones: chatbots quietly granted one small write capability and still governed like FAQ widgets. Re-classify on every capability addition, because the upgrade from chatbot to agent usually happens one harmless-looking tool at a time.