Concepts

AI agent or automation

The word agent gets attached to everything. This is the difference that decides what you should build.

This page is written independently and also recommends things I do not sell myself. If you have a concrete question, ask it: you usually get an answer the same working day. Prefer to look for yourself first? Do the free process check.

The short answer

Automation runs a fixed route you defined and never does anything else. An AI agent decides the order and the steps itself to reach a goal. Fixed steps and a fixed outcome call for automation. Varying input and a judgement per case call for an agent, and then always with an employee who approves.

The difference in one table

The practical dividing line, not the marketing definition.
AutomationAI agent
Decides the stepsYou, in advanceThe model, per case
Same input, same outcomeAlwaysUsually, not guaranteed
Breaks onInput that does not fit the rulesA goal described too vaguely
Cost per casePredictableDepends on how many steps the model takes
Auditable afterwardsFully, the route is fixedOnly if you record every step
SuitsRetyping, routing, checking against fixed rulesResearching, summarising, weighing, drafting

Why this matters

An agent that decides the order itself can also decide the wrong order. As long as it only prepares, that is annoying at worst. Let it send, book or commit, and it becomes a risk you do not see coming, because it fails not with an error message but with a plausible wrong answer.

So every agent needs a boundary: what may it read, what may it do, who approves, when does it stop and what does it record. Those five questions sit on the page about control and security.

What you build in practice

Usually both, inside the same workflow. The automation fetches the mail, prepares the attachment and writes the result away. The model does the step that asks for a judgement: what does this say, does it match the source, which reply fits here. The employee decides.

Whoever starts with an agent for everything builds something nobody can check. Whoever starts with automation for everything gets stuck on the first case that deviates. Drawing the line is the real work.

Which of the two do you need

Answer one question: can someone who knows the work write down in advance which step follows when, without exceptions? If yes, it is automation and you need no model. If no because it differs per case, that is where the agent sits. The free process check walks that question through with you.

Frequently asked questions

Is an AI agent the same as a chatbot?

No. A chatbot answers a question. An agent receives a goal and takes several steps by itself to reach it, for example looking up information, comparing it and drafting a result.

Can an agent enter my system on its own?

Technically yes, sensibly no. In the way of working I build, an agent may prepare and may never send, book or commit on its own. An employee approves and decides.

Which one is more expensive?

Automation has predictable costs, an agent does not: it charges per step the model takes, and that varies per case. At large volumes that difference is worth calculating.

How do I know whether my process needs an agent?

If someone who knows the work can write the steps down in advance without exceptions, you do not need an agent. If it differs per case and a judgement is required, you do.

Read on

Free tools

Start without an appointment

Use the process check, calculator, baseline builder, prompt builder or control card builder. No account needed.

See all tools

Tamboer AI

Still unsure which way to go?

Email five lines about your work. I personally tell you what I would do, even when that means you do not need me.

Email your question