The one-line difference: a chatbot answers, an agent acts. A chatbot is a conversation interface over knowledge — it tells a customer your refund policy. An agent is software that pursues a goal across your systems — it actually processes the refund, updates the CRM, and emails the confirmation.
Vendors blur these terms constantly ("agentic chatbot", "AI assistant"), usually to charge agent prices for chatbot capability. Here is the distinction that matters when you are deciding what to build.
The practical difference
| Chatbot | AI agent | |
|---|---|---|
| Core job | Answer questions in conversation | Complete tasks toward a goal |
| Connects to | Your docs / knowledge base | Your actual systems (CRM, email, billing, APIs) |
| Acts without a human? | No — it only responds | Yes — within the guardrails you set |
| Typical wins | Support deflection, lead qualification, FAQ | Ticket resolution end-to-end, data entry, order processing, scheduling, reporting |
| Failure mode | Wrong or unhelpful answer | Wrong action — which is why guardrails and audit logs are non-negotiable |
| Typical cost (our tiers) | $2–4K, 1–2 weeks | $4–10K, 3–5 weeks |
When a chatbot is all you need
- The same 20–50 questions eat your support inbox, and the answers live in docs you already have.
- You want 24/7 first-line response with clean handoff to humans for the hard cases.
- You need to qualify inbound leads before they reach a salesperson.
- Your systems of record should stay human-operated for now — you want answers, not automation.
A well-built chatbot grounded in your real content (see RAG — that is the retrieval technique underneath) resolves a large share of routine inquiries and, importantly, knows when to say "let me get you a human." That last part is where cheap off-the-shelf bots fail.
When you actually need an agent
- The bottleneck is not answering questions — it is the repetitive doing: copying data between tools, processing routine requests, chasing status updates.
- A task follows rules a competent new hire could learn in a week. That is the automation sweet spot.
- The work spans multiple systems — email in, CRM update, invoice out — which no single tool’s built-in automation covers.
- Volume is growing but you do not want the next three hires to be data-entry roles.
The honest caveat: agents are only as good as the guardrails around them. A production agent needs defined permissions, human-approval gates for consequential actions, and an audit trail. Anyone selling you a "fully autonomous employee" with none of those is selling a demo, not a system.
Cost and build reality in 2026
Our own pricing, since we publish it: a custom AI chatbot with analytics runs $2–4K and ships in 1–2 weeks; AI agents wired into your org’s systems run $4–10K over 3–5 weeks, with the extra time going into integrations, guardrails, and testing against your real workflows. SaaS chatbot subscriptions look cheaper ($50–500/month) until you need your own data, your own integrations, or your own behavior — the usual reason companies come to us after outgrowing one.
A 60-second decision checklist
- Write down the top 10 things you wish were off your team’s plate.
- Mark each one A ("answering something") or D ("doing something").
- Mostly A → chatbot. Mostly D → agent. A real mix → chatbot first; it is cheaper, ships in days, and its knowledge layer becomes the foundation the agent uses later.
- For every D item, ask: could a new hire learn this task’s rules in a week? If yes, it is automatable. If it needs judgment calls with real consequences, keep a human in the loop by design.
Still unsure which side your workload falls on? That is literally what our free AI audit is for — 30 minutes, no pitch, and we will tell you if the answer is "neither yet."