Agent Systems: Terminology and Building Blocks

This page defines the objects used throughout the series.

Outline

1. Define model, agent, prompt, context, state, tool, memory, evaluator, and human checkpoint.
2. Distinguish a model from an agent system: a model produces an output; an agent system manages a loop, state, tools, and evaluation.
3. Introduce the basic contract: input → context → inference → action → observation → updated state.
4. Explain the design-pattern vocabulary: recurring problem, stable interface, variable implementation, forces, trade-offs, and failure modes.
5. Explain why a small interface is preferable to exposing every internal detail.

Design question

What should remain stable when the model, tool, prompt, or memory implementation changes?

Suggested figure

01-building-blocks.svg in the Agent design pattern illustration folder.