Follow material in an order that builds understanding.
inAiLoading page content...
Unit 01 · system classification
Plan for this page
After this unit, you should be able to
Terms you will use
These definitions prepare you for the reading; you do not need to memorize them.A learned component that produces an output from input; by itself it does not define the surrounding interface, tools, permissions, or process.
A specified sequence of steps, decisions, and handoffs that may contain AI components but usually follows a designed path.
A system that continues toward a goal across steps, uses observations or tools, and selects at least some next actions within defined boundaries.
A model is a capability inside a larger system. A chatbot is an interaction pattern, usually organized as turns of messages. A workflow coordinates steps. An agentic system continues work toward a goal and can select some actions from changing observations. One product can contain all four layers.
The categories are not a ladder from primitive to advanced. A fixed workflow can be safer and more useful than an agent for a stable task. A chat interface can call tools without gaining broad discretion. Classification should explain behavior, not award status.
Ask what happens after the first request. Does the system only return an answer, follow a predetermined branch, or observe results and decide what to do next? Then inspect whether it can act through tools, retain task state, retry, ask for approval, and stop.
Degrees matter. A system might select a document-search query but require a person to send the final email. Another might schedule, purchase, or modify records. Both may be called agents, yet the second carries more delegated authority and needs stronger controls.
A useful classification names the model or capability, interface, sequence, available actions, state, and decision points. It also notes what is not visible. A product page saying autonomous does not prove which operations occur without review.
When details are missing, state uncertainty instead of guessing. Ask for a system diagram, tool list, permission scope, logs, approval points, and failure behavior. The answer may be a hybrid: chatbot interface, fixed workflow for routine steps, and agentic selection inside one bounded stage.
Interactive model
The model or rule-based component that interprets, predicts, or generates.
Misconception to correct
Active checks
Explain back and transfer
Source context
Supports a product-neutral view of AI systems through inputs, outputs, objectives, and degrees of autonomy; it does not define every commercial use of the word agent.
Provides a concrete technical account of models, tools, observations, actions, and agent loops; it is ecosystem-specific and not a universal naming authority.
Supports examining a whole deployed system and its context when classifying risk; it is not a beginner taxonomy of product labels.
