Lesson 1 of 7 · About 3 minutes
From answering to acting
Distinguish a model, a workflow, and an agent by who chooses the next step.
01 / Learn the idea
Start here
A large language model (LLM) generates text using patterns learned during training. It works with tokens: pieces of words, whole words, or punctuation. A fluent reply can still be wrong. Writing a convincing answer is therefore different from checking that an action actually happened.[1]
A workflow follows steps arranged by its developer. An agent can use a model to choose its next step and tools as the task unfolds. We use Anthropic's distinction here; the word agent is used differently elsewhere. Both approaches can include AI, and neither is automatically the better choice.[2]
Imagine a homework helper. Explaining fractions is an answer. Always reading a worksheet, summarizing it, and saving a draft is a fixed workflow. Checking what you misunderstood and choosing a suitable practice question illustrates an agent making a next-step choice. Ask what the system can do, not what its name promises.
Original school example
The library helper
A reply says a book is available. A tool-enabled helper checks the library catalogue first. If the book is unavailable, it can look for another edition. The catalogue result changes the next step.
A chat interface alone tells you little. Look for what chooses the next action and what tools the system can actually use.
02 / Explore the flow · Guided simulation
One request. Three ways to handle it.
Find a room for 24 students. Compare who decides the next action in each design.
Original fictional scenario. Decisions are authored; calculations and checks run in your browser. No live AI or external action.
↳ A failed check can send the flow back to another action—or to a human.
Step 1 of 3 · Visible system event
Request
Find a room for 24 students on Friday at 15:00.
Inspect the visible trace (1 events)
- Request
Find a room for 24 students on Friday at 15:00.
This log shows authored actions, tool results, and checks. It is not a model’s private reasoning.
03 / Check your understanding
A school app always summarizes a form and then saves a draft. Its developer fixed both steps. What best describes this design?
Where these ideas come from
Primary sources, explained in our own words. Research findings and engineering guidance are different kinds of evidence; neither is a guarantee that an agent will succeed.
Supports: Token generation; Learned patterns and model limitations.
Limit: An introductory model explanation, not evidence that any particular agent is reliable.
Supports: Workflow versus agent distinction; Choosing simple systems.
Limit: An engineering taxonomy, not a universal definition. The page notes that its tooling has changed since publication.
Stemtiq original lesson · Sources checked September 12, 2026 · Editorial review pending. School examples and guided simulations are original simplifications, not replications of the cited systems. No research affiliation or endorsement is implied.