If you've shopped for "AI" in the last year, you've probably been pitched a chatbot. Maybe one that lives in the corner of a website, answers FAQs, and hands off to a human when it gets stuck. That's not what we mean when we say agent.
The shortest distinction we use in calls:
Chatbots talk. Agents do.
A chatbot's job ends at the response. An agent's job starts at the response, takes an action, then comes back to you with the result. The difference shows up in what each one can actually accomplish for your business.
Five examples, in order of "obvious" to "weird"
1. The lead qualifier
Someone fills out a form. A chatbot would say "thanks, we'll be in touch." An agent reads the submission, checks your CRM to see if they're already a contact, scores them against your ideal customer profile, books a 30-minute slot on your calendar if they qualify, and updates their record with the call details. By the time you check your inbox, the meeting is on the books.
2. The proposal drafter
Discovery call happens. Notes get dropped into a shared doc. An agent reads the notes, pulls the relevant package details from your pricing system, drafts a proposal in your template, and sends it for your review — usually in under five minutes after the call ends. You edit, you send. The 48-hour-proposal promise becomes 4-hour reality for most clients.
3. The content pipeline
A new case study gets approved. An agent generates a LinkedIn post, a Twitter thread, an email newsletter blurb, and a website carousel image — all in your brand voice — and stages each one for approval before publishing. You don't write four versions of the same announcement; you approve four.
4. The data janitor
New client gets onboarded. An agent watches for the welcome email, pulls relevant onboarding documents, creates folders in your project management system, sends calendar invites to the team, populates a kickoff agenda, and DMs you a one-line summary. This is the boring kind of agent — the kind nobody demos at conferences — and it's the one that saves the most hours per week.
5. The watchdog
Site goes down. Agent notices within 60 seconds, checks the status pages of your hosting provider, pulls the last 50 error logs, drafts a postmortem-in-progress, and pages whoever's on call with a one-line summary plus a link to the working draft. You wake up not to "something broke" but to "something broke and here's what we know."
What these all have in common
Notice none of them are conversational. The user doesn't talk to the agent. The agent just does the work. The conversation interface — if there is one — is for you, not the customer.
That's the second clearest mental model we use:
Chatbots are customer-facing. Agents are usually employee-facing.
This isn't a hard rule (sometimes you want both in the same system), but it's a useful starting point. If you're picturing a chat bubble, you might want a chatbot. If you're picturing "I wish someone would just handle this thing every time it happens," you probably want an agent.
What they cost
Off-the-shelf chatbots: $50–$500/month subscription, deploy in hours, capability ceiling hits fast.
Custom AI agents like the ones above: $3K–$15K one-time build depending on integration depth, then $20–$200/month in API costs depending on volume. Total cost of ownership undercuts the off-the-shelf chatbot once you exceed about 12 months and need actual integrations with your systems.
We built a comparison page on this if you want the full breakdown.
What to ask before you commission one
Three questions in order:
- What action does it take? If the answer is "it responds to messages," that's a chatbot. If the answer involves a verb that touches one of your systems (your CRM, your calendar, your database, your email), that's an agent.
- What does success look like in numbers? "Saves Cole 6 hours per week" is a better target than "improves customer experience." Vague targets produce vague agents.
- What happens when it gets confused? Good agents have a clear escalation path — back to a human, with full context. Bad agents hallucinate confidently. Ask to see how the escalation logic is designed before you sign anything.
If you're in the market and want a sanity check on whether you actually need an agent (versus a chatbot, or honestly, a Zapier flow), drop us a line. We'll tell you honestly.
— Cole