Lesson 1 of 5, 11 min with the exercise

Do you need an agent at all?

Pick the simplest thing that does the job.

An agent is a model running in a loop: it looks at the situation, picks a tool, sees the result, and decides what to do next. That flexibility is what makes it useful, and also harder to predict, slower and more expensive.

• The steps are always the same: use ordinary automation, with no AI.
• The steps are the same but one needs reading or judgement: use automation with one AI step.
• The next step depends on what it finds: consider an agent.

Most business tasks fit the first two. A support ticket that must be classified and routed is one AI step. Investigating a billing dispute across three systems might justify an agent.

See it in practice

Invented example: refund requests. If the rule is "under $50 and within 30 days, approve", that's plain automation. If someone needs to read the email to find the order number and the reason, that's automation with one AI step. If each case means checking orders, shipping and past complaints in whatever order the evidence suggests, that's where an agent might pay off.

Common mistake: Starting with an agent because it's the exciting option. You'll spend your time debugging decisions that a simple rule would never have got wrong.