MODULE 04 — AGENTIC AI · 13 AI Concepts

What makes AI “agentic”

Nothing was added to the model. The same next-word machine was wired into a circle, and allowed to keep going until it decides it is finished.

Loops0 Tool calls0 On the desk0
Goal

Find the cheapest supplier for 500 units.

Press Step. Watch which stage repeats — that repetition is the entire difference.

The problem it explains

Same model, different wiring

A PLAIN CHAT prompt the model answer one pass · it stops whether or not it worked AN AGENT goal the model byte-for-byte identical a tool what happened, fed back in …then it goes again The difference is one arrow. Nothing inside the box changed.

How it works

Three moves it repeats

“find the cheapest supplier” 1 · check each price list 2 · compare at 500 units 3 · write it in the note
PlanIt writes the steps down as text, because text is all it can produce.
open_price_list(“Bregan”) your code runs it
ActOne step at a time, through the same tool handoff as page 12.
$3.95 / unit done yet? yes · stop no · loop
Observe and decideThe stopping rule is the agent’s hardest and least reliable judgement.

What it costs

Every loop re-reads everything

THE DESK, AFTER EACH LOOP loop 1 — goal + one result loop 2 — everything above, again loop 3 loop 4 loop 5 — and the edge is coming A mistake in loop 1 is read back as established fact in every loop after it.

So what

An agent is not a smarter model. It is the same model, given a tool, a loop, and permission to decide when to stop.