What an Agent Is · ai-skills

Tools, Workflows and Agents · Beginner to Intermediate

An agent is a loop. A model chooses an action, a tool carries it out, and the result feeds back into the next choice.

Approach	Who picks each step	Example
Single prompt	Nobody: one input gives one answer.	Draft a paragraph from a fixed brief.
Fixed script	The author, ahead of time: steps run in a fixed order.	Lint, then compile, then report, every night.
Agent loop	The model, at run time: each result shapes the next choice.	Read a failure log, then decide which check to run.

Goal-state to model to tool to result and back, with a stop edge to done.

The agent loop: each result updates the state behind the next choice.

Lesson	What it covers
Prompts, scripts, and agents	Single prompts and fixed scripts compared with a choosing loop.
The agent loop step by step	Read, choose, run, record: the four events and the step budget.
Agent components	Model, tools, memory and state, goal, and stopping rule.
Agent loop simulator	A runnable loop with a fake decision function and two fake tools.
When a script wins	Fixed steps, cost arithmetic, and when to skip the loop.

Check yourself A program calls a model once, then runs three fixed tool calls with no feedback. Is it an agent?

No. The model chose nothing after the first call, and no result shaped a later choice. It is a single prompt followed by a fixed script.