What is the difference between an AI agent and a chatbot?

A chatbot produces text. An agent takes action in a system: it reads a calendar and writes an appointment, looks up an order and updates a record. The useful test is whether anything changes outside the conversation. If nothing does, it is a chatbot.

Which process should we automate first?

The one that repeats most often, has rules someone can write down, and touches systems that can be reached programmatically. In a 10–100 person company that is usually inbound enquiries or appointment booking, because both are high volume and low judgment.

How many processes can we automate at once?

One, to start. Every process automated in parallel multiplies the number of edge cases surfacing at the same time, and the team absorbing them is the same small team. Sequential is slower on paper and faster in practice.

Does an AI agent replace people?

In this company size it usually absorbs work nobody had time for rather than work someone is doing. The typical starting point is not an over-staffed team, it is a backlog: calls that ring out, emails answered two days late, a CRM nobody fills in.

What does every one of these use cases need to work?

A defined process that repeats often, rules that can be written down, systems reachable programmatically, and a named person for the cases the agent should not decide. All four, in every case on this page.

Use cases

What AI agents actually do in a 10–100 person company

An AI agent is software that takes action in your systems, not a chatbot that produces text. In a mid-sized company it earns its keep in six places: inbound calls, appointment booking, lead qualification, back office, internal knowledge and CRM upkeep. Each one has a different failure mode.

By Gorden WübbeAutomation, agents and search visibilityUpdated 10 August 2026

Agent or chatbot: the test that matters

The word “agent” is applied to almost anything with a language model behind it, which makes it useless as a buying signal. One question separates the two: does anything change outside the conversation?

A chatbot explains your opening hours. An agent reads the calendar, finds a slot, books it, sends the confirmation and writes the record. If nothing in a system changed, you bought a chatbot — which is fine, as long as that is what you wanted.

The six places it pays off

Use case	What the agent does	Connects to	Where it breaks
Inbound calls	Answers, identifies what the caller wants, handles it or routes it with context.	Phone system, CRM	Complaints and anything emotional
Appointment booking	Checks real availability, confirms, reschedules, chases confirmations.	Calendar, CRM	Multi-resource scheduling
Lead qualification	Replies within minutes, asks the qualifying questions, writes the record.	Forms, email, CRM	Judging strategic fit
Back office	Moves data between systems, prepares recurring documents, logs every step.	Accounting, storage	Exceptions with legal weight
Internal knowledge	Answers the questions your team keeps asking each other.	Wiki, drive, chat	Outdated source documents
CRM upkeep	Turns calls and emails into structured records without anyone typing them.	Email, calls, CRM	Anything never said out loud

The last column is the one worth reading twice. Every use case on this page has a specific point at which it stops being useful, and knowing that point in advance is the difference between an agent that runs for years and one that gets switched off after a bad month.

How to pick the first one

Not by which sounds most impressive. By which scores highest on four questions, each answerable in a minute:

How often does it repeat? Below roughly 20 to 30 times a week, the build costs more than it returns.
Can someone write the rules down? Ask the person who does the work. If the answer is three pages of conditions, good. If it is “it depends”, pick something else.
Can the systems be reached? A modern CRM with an API is a day of work. A nightly CSV export is a different project.
Who catches what the agent escalates? If no name comes to mind, the process is not ready regardless of how good a fit it looks.

In practice, companies of this size land on inbound calls or appointment booking first. Both are high volume, low judgment, and the improvement is visible within days rather than quarters.

Why one at a time

The temptation with a working first agent is to start three more. It rarely goes well at this company size. Every process in flight surfaces its own edge cases, and the people absorbing them are the same two or three people. Sequential looks slower on the plan and finishes earlier in reality.

What none of this replaces

In a company of 10 to 100 people, an agent usually absorbs work nobody had time for rather than work someone is currently doing. The honest starting point is not an over-staffed team — it is a backlog. Calls that ring out at lunchtime. Emails answered on Thursday that arrived on Tuesday. A CRM everybody agrees is out of date.

That distinction matters for how you introduce it internally. An agent presented as a headcount decision meets resistance from the people whose cooperation the project needs most.

The questions that come up first

What is the difference between an AI agent and a chatbot?: A chatbot produces text. An agent takes action in a system: it reads a calendar and writes an appointment, looks up an order and updates a record. The useful test is whether anything changes outside the conversation. If nothing does, it is a chatbot.
Which process should we automate first?: The one that repeats most often, has rules someone can write down, and touches systems that can be reached programmatically. In a 10–100 person company that is usually inbound enquiries or appointment booking, because both are high volume and low judgment.
How many processes can we automate at once?: One, to start. Every process automated in parallel multiplies the number of edge cases surfacing at the same time, and the team absorbing them is the same small team. Sequential is slower on paper and faster in practice.
Does an AI agent replace people?: In this company size it usually absorbs work nobody had time for rather than work someone is doing. The typical starting point is not an over-staffed team, it is a backlog: calls that ring out, emails answered two days late, a CRM nobody fills in.
What does every one of these use cases need to work?: A defined process that repeats often, rules that can be written down, systems reachable programmatically, and a named person for the cases the agent should not decide. All four, in every case on this page.

Which of these is costing you the most?

Thirty minutes on one concrete process, and you will know whether it is worth automating. If it is not, you will hear that instead.

Book a scoping call