A chatbot replies. An AI agent acts. That's the difference in one sentence. This post explains what that means for the products you actually buy, with specific examples and pricing reality.
The 30-second definition
| Chatbot | AI Agent | |
|---|---|---|
| What it does | Replies in text | Takes actions in your systems |
| Integration depth | Knows what's in its prompt | Calls APIs, reads databases, writes records |
| Resolution rate (support) | 20–40% tier-1 | 60–75% tier-1 |
| Pricing | Per-seat / message | Per-resolution / per-task / per-seat |
| Setup time | Hours to days | Weeks |
| Examples | ELIZA, classical IVR, basic ChatGPT chat | Sierra, Decagon, Intercom Fin |
A chatbot is bounded by its prompt and training. An AI agent is bounded by the tools you give it.
What chatbots actually are
A chatbot is a conversational interface — usually text, sometimes voice — that responds to inputs with pre-trained or templated responses. Classical chatbots use rule-based logic (decision trees, intent matching). Modern chatbots use generative AI — an LLM generating responses on the fly.
What chatbots don't do: take action outside the conversation. They tell you the answer; you act on it.
Examples:
- Pre-2023 customer service bots ("press 1 for billing")
- ChatGPT in default chat mode — answers your question, doesn't take action
- A FAQ bot on a website that suggests articles
- A pre-trained intent matcher inside an IVR
What AI agents actually are
An AI agent is a system built around a generative model that operates in your environment. It uses tool use — function calls into your business systems — to read data, take actions, and verify outcomes.
A customer support AI agent doesn't just say "your order is on its way." It looks up the order in your warehouse system, checks the shipping carrier API, sends a follow-up note if delayed, and updates the ticket status.
Examples:
- Sierra — branded customer-facing agent that resolves tickets end-to-end
- Decagon — chat-first agent with deep helpdesk integration
- Intercom Fin — outcome-priced agent inside Intercom
- Cursor Agent — coding agent that edits files and runs tests
The technical difference: chatbots end at the LLM's response. Agents add the agentic loop — observe → reason → act → observe — that lets them take real-world actions.
When chatbots win
Simple FAQ deflection. "What are your hours?" "Where's my receipt?" If 30% of your tickets are answerable from a knowledge base alone, a chatbot is the cheapest path.
Strict guardrails. Highly regulated workflows where the cost of a wrong action is asymmetric. A chatbot that says "please contact support for refunds" costs less than an agent that processes refunds wrong.
Cost-sensitive deployment. Chatbot tiers start at ~$50–200/month. Agent deployments often start at $20–40K/year. If volume is low, the chatbot math is overwhelming.
Quick wins. Chatbots ship in days. Agents take weeks of integration work.
When AI agents win
Real deflection targets. Want to deflect 60–75% of tier-1 tickets? You need an agent that touches your order system, refund flow, and account database. A chatbot caps out around 30–40%.
High volume. Once you're processing 5,000+ tickets/month, agent per-resolution pricing (typically $0.50–2.00) beats the loaded cost of human reps.
Cross-system workflows. When the answer requires data from three systems and an action in a fourth, only an agent can complete the flow end-to-end.
Voice + chat parity. Agents like Parloa, Sierra, and Vapi work across channels. Chatbots are typically text-only.
A practical decision tree
Ask yourself three questions:
1. Does your most common workflow require data from your own systems?
- Yes → AI agent
- No → Chatbot is enough
2. What deflection rate do you need to hit?
- Under 30% → Chatbot
- 30–50% → Chatbot with deep KB integration, or low-tier AI agent
- 50%+ → AI agent
3. What's your support volume?
- Under 1,000 tickets/mo → Chatbot (volume doesn't justify agent setup)
- 1,000–5,000 → Either, lean toward agent if budget allows
- 5,000+ → AI agent (per-resolution math becomes overwhelming)
For specific buyer guidance see our customer support agent buyer's guide.
Pricing reality
Chatbots
- $50–500/month (small business tier)
- Setup: hours to days
- Total first-year cost: $1,000–10,000
AI agents (customer support)
- Intercom Fin — $1/resolution. Pay only for what deflects.
- Decagon — typically $40–120K/year mid-market
- Sierra — enterprise contracts, $150–500K/year
- Setup: 2–8 weeks (depends on knowledge base maturity)
The 10–50× cost gap between chatbot and agent is real. So is the throughput gap once the agent is deployed: well-implemented agents handle the work of 5–15 human reps at less than the loaded cost of one.
The gray area: "chatbots" that are actually agents
Several vendors market "AI chatbots" that are technically agents — they call APIs, look up data, take actions. This is mostly a positioning choice; "chatbot" is the term buyers search for, even when they want an agent.
How to tell them apart in a demo:
- Ask "what specific actions does the system take in our systems?"
- Vague answer ("it understands your knowledge base") → it's a chatbot
- Specific answer ("it queries your order API, processes refunds via Stripe webhook") → it's an agent
Related distinctions
- Agentic AI vs Generative AI — the broader framing
- Autonomous vs copilot agents — the agent-autonomy spectrum
- Customer support agent buyer's guide — specific to support
The verdict
For most teams in 2026:
- Chatbots are still the right answer when volume is low or the workflow is genuinely simple
- AI agents win at volume, on cross-system workflows, and when deflection targets are real
- The cost gap shrinks as outcome-based pricing (per-resolution) replaces seat-based pricing
If your tickets are routine and infrequent, a $200/month chatbot is enough. If they're cross-system and high-volume, an AI agent will pay for itself inside a quarter.