AI agent vs chatbot.

The short version

A chatbot answers questions.

An AI agent takes actions.

Same underlying language model, often. What changes is whether the system has hands. A chatbot stops at the response; an agent continues to the next step in the world.

Concrete example

A user types: "Send a follow-up email to the prospect we met last Tuesday."

The chatbot drafts a follow-up email and shows it to the user. Maybe it asks for a few details. The user copies the draft, opens their email client, finds the prospect's contact, pastes, sends, then logs the activity in their CRM. The chatbot was helpful but the work happened outside it.

The agent queries the calendar to find which prospect met the user last Tuesday, looks up their contact details in the CRM, drafts the email referencing the meeting, sends it through the connected email service, and logs the activity back to the CRM. The user gets a Slack notification: "Sent. Logged to Salesforce."

When you actually need a chatbot

• FAQ-style customer questions where information is the answer
• Knowledge-base search disguised as conversation
• Casual user-facing chat where reading is the only outcome
• Internal Q&A over docs (handle with retrieval, not action)

Chatbots are simpler to build, cheaper to run, and carry less risk. If the workflow ends at "the user reads the answer," you don't need an agent.

When you actually need an agent

• Sales workflows: send the email, update the CRM, book the meeting
• Customer support that needs to look up an order, issue a refund, trigger a workflow
• Operations work: spreadsheet automation, scheduled reports, multi-step processes
• Coding work: read files, edit code, run tests (Claude Code, Cursor)
• Voice calls that need to look something up and act mid-call

If the workflow ends at "something changed in your systems," you need an agent.

The line is blurring

Modern chatbot platforms (Intercom, Zendesk, custom GPTs) have been adding agent-style tool-calling since 2024. A "chatbot" that can look up your order, issue a refund, and trigger a workflow is — by the definition above — an agent. The terminology has lagged the capability.

Practical advice: ignore what the vendor calls it. Look at what it can do. If it takes actions in your systems, it's an agent regardless of the product name.

Cost difference

Agents typically cost more per task than chatbots. They use larger models (more reasoning capacity), consume more tokens (multi-step workflows accumulate context), and run longer per interaction.

Order of magnitude: a chatbot answer might cost $0.001. An agent completing a five-step task might cost $0.01–$0.10. Use the cost calculator to model your specific case.

Which platform should you actually use?

For chatbot work (FAQ, knowledge-base): a tool like Stack AI or any vendor-native chatbot in your existing helpdesk stack.

For agent work (taking actions): the right answer depends on your use case. Run our Agent Picker — five questions, one recommendation.