AI agent vs AI assistant vs chatbot: which one do operators actually need?

The three terms get used interchangeably, sold interchangeably, and priced wildly differently. They aren't the same thing. Picking the wrong one for your business is how you end up paying $4,995 for something a $29/month tool would have done, or paying $99/month forever for something that should be a one-time build.

Here's the actual difference, in plain English, plus a decision tree for which one your business needs.

The three terms, in one sentence each

Chatbot: Waits for a question and answers it. Reactive. Linear. "If user types X, respond with Y."

AI assistant: Helps you do a task by suggesting, drafting, or completing parts of it. Still requires you to be in the loop driving. Think Microsoft Copilot or ChatGPT.

AI agent: Acts on its own across multiple tools. Reads, decides, executes, reports back. You give it a goal, it figures out the steps. The big shift from assistants to agents is that agents do the work, not just help you do the work.

That's the whole distinction. Everything else is marketing.

What each one actually does for a 12-person business

Chatbot use case: Your website has a chat widget. A visitor asks "what are your hours?" The chatbot answers from a FAQ. The visitor asks "can I book a consultation?" The chatbot hands them off to a Calendly link. Useful. Limited. Costs ~$50–$200/month for a tool like Intercom or Drift.

AI assistant use case: A team member uses ChatGPT to draft a difficult client email. They write a rough version, paste it in, ask for tightening, get a better draft back, edit, and send. The human is in every step. Useful for individual productivity. Doesn't change how the business runs. Costs $20–$30/month per seat (ChatGPT Plus, Claude Pro, Microsoft Copilot).

AI agent use case: Your meeting-notes agent joins every Zoom call, transcribes, writes a structured summary, extracts the action items, pushes them into HubSpot, drafts the follow-up email, and queues it for review. You don't ask it to do this. It just does it after every call. Costs $4,995 once.

The agent isn't faster than the assistant. It's a different category. The assistant is a smarter pen. The agent is a smarter intern.

The decision tree

Three questions to figure out what you actually need:

Question 1: Is the task one-shot or recurring?

If you do it once, an assistant is fine. (One-time copy edit, one-time research, one-time analysis.)

If you do it 5+ times per week, look at an agent. The repeated work is where the agent's setup cost pays back.

Question 2: Does the task touch more than one tool?

If the task lives entirely in one tool (writing inside Google Docs, ideating in ChatGPT, summarizing inside Notion), an assistant covers it.

If the task spans tools (read email → check CRM → update spreadsheet → send Slack message), you want an agent. Agents are defined by their ability to act across systems. Assistants can't do that on their own.

Question 3: Does the task have a clear outcome that doesn't require human judgment?

If the outcome is "draft something a human will edit," assistant.

If the outcome is "complete the workflow end-to-end with light review," agent.

If the outcome requires real human judgment at every step (legal opinion, medical diagnosis, hiring decision), neither — that's a person's job, with AI helping.

Picks by vertical

Quick map of which one a growing team in each space usually needs first:

Dental practice: Agent. Phone calls, no-show rebooking, recall lists, insurance verification — all recurring, multi-tool, clear-outcome work. AI Receptionist productized at $4,995 + $497/mo.

CPA firm: Agent. Meeting notes to CRM, document intake from clients, follow-up nudges. Recurring multi-tool work, especially during tax season. Meeting Notes → CRM at $4,995.

Law firm: Mix. Inbox triage and engagement letter drafting are agent work ($2,995 + $3,995). Substantive matter analysis stays as assistant work (Claude or GPT-4 with the attorney in every loop).

Marketing agency: Agent. Proposal drafting from past wins, meeting notes, CRM hygiene. The proposal-drafter SKU alone usually pays back in two weeks at agency billing rates. Proposal Drafter at $3,995.

Consulting firm: Agent for client-facing work (meeting notes, follow-ups), assistant for thinking work (analysis, decks, custom research).

E-commerce or retail operator: Chatbot for customer service triage, agent for inventory and supplier follow-ups, assistant for product description writing. All three have a place.

The trap to avoid

The trap is paying agent prices for assistant work, or paying chatbot prices for agent work.

Paying $4,995 for an "AI agent" that's really a fancy chatbot answering form questions is overpaying. Paying $99/month for "AI scheduling" that's actually a calendar integration with a chat widget is underpaying for what it actually is.

If a vendor uses all three terms in the same paragraph and won't tell you which one they're actually building, that's a tell. Real agents do specific multi-tool work and the vendor can describe exactly what tools the agent will read and write to. Real assistants are a single chat interface plus integrations. Real chatbots are decision trees with NLP on top.

Where to start

If you want help figuring out which one your specific business needs, a 30-minute audit walks through your workflows and outputs a recommendation: which 1–2 things should be agents, which 2–3 should be assistants, which 1 might be a chatbot, and which work to leave alone for now.