Agent or Workflow? Choosing the Right Design
Decide in ten minutes whether your idea needs an agent, a fixed workflow or a single good prompt
A taste of a lesson
We want an agent that reads support emails, tags them and forwards them to the right team. Is an agent the right call?
Probably not, and that is good news. Your steps are the same every time: read, tag, forward. The model only needs to make one judgment, the tag. That is a workflow with a classification step: code receives the email, one model call returns a tag from a fixed list, and code does the forwarding. It is cheaper, faster and easy to test with a set of past emails. Quick check: can you think of one email type where the next step would depend on something the system has to go and find out?
Written by the teacher as an example. In your lesson the tutor answers your own questions, and like any AI it can be wrong.
What you will be able to do
- Explain the difference between a single call, a workflow and an agent in plain words
- Place any proposed AI task on the autonomy spectrum and justify the choice
- Name the cost, speed and testing trade offs that come with more autonomy
- Spot vendor claims that use the word agent loosely and ask the right question
Lesson plan
- 1 Three designs: one call, a workflow, an agent Tell apart the three basic designs by asking one question: who decides the next step. Start
- 2 What makes a task agent shaped Recognise the signals that a task genuinely needs a model to decide steps on the fly. Start
- 3 What makes a task workflow shaped Recognise tasks where a fixed sequence is cheaper, faster and easier to trust. Start
- 4 The price of autonomy Estimate what an agent costs in money, time and testing effort compared with a workflow. Start
- 5 Worked cases: sorting five real tasks Practise choosing a design for realistic tasks and defending the choice. Start
- 6 Talking about agents with your team Explain your design choice to colleagues and question loose vendor claims. Start
Try asking
About this tutor
For anyone planning an AI project who keeps hearing the word agent and is not sure it applies. You will learn the difference between a single model call, a fixed workflow where code decides the steps, and an agent where the model decides what to do next in a loop. Each lesson takes a real looking task, places it on that spectrum and explains the trade offs in cost, speed, predictability and failure. By the end you can look at your own idea, choose the simplest design that will work, and explain your choice to a colleague or a manager without jargon.
Reviews
4.7
3 ratingsSample
- Sophie R.Sample
Helped me push back on a vendor pitch with actual questions instead of just a bad feeling.
- Marta L.Sample
I came in sure we needed an agent for our onboarding emails. Two lessons later I had a simple workflow sketched and a clear reason why. The 'who decides the next step' question is now on a sticky note on my monitor.
- Kwame A.Sample
Clear and calm. The five worked cases were the best part. I wanted a little more on hybrid designs, but the basics are solid.
About the teacher
I teach how AI agents are built: the loop, the tools, the memory, and when a plain workflow is the better choice
9 tutors 310 lessons taught Sample
I build and teach the inner workings of AI agents. Most of my working life has been spent on backend systems, so I approach agents the way I approach any distributed system: what runs, in what order, what can fail, and what it costs. I like to start every topic with a drawing of the loop on a whiteboard and...
See Hiroshi's profile and tutorsMore like this
Other tutors on the same or nearby topics.