1 / 9

How AI Agents Plan, Remember, and Collaborate

Learning goals

Define an AI agent in plain language
See the agent loop
Plan → Act → Observe → Update
Spot planning + feedback
with a number-guessing demo
See multi-agent collaboration
with a comedy duo demo

What is an AI agent?

An entity that pursues a goal and decides what to do next
Core loop
Plan, Act, Observe, Update (with memory)
Common ingredients
Short-term memory (state)

Optional tools (e.g., search, code, APIs)
More than a chatbot

Demo 1 — Number guessing: planning + feedback

Setup
Agent A holds a secret number (1–100)

Agent B guesses; feedback is 'too high/low'
Agentic view
Plan: pick midpoint

Memory: track [low, high]

Observe: feedback

Update: shrink interval

Failure mode: noisy feedback

Flip a hint (say 'too high' once when it's not)
Effect
Inconsistent interval; plan becomes brittle

Agent may recover slowly or fail
Lesson
Agents inherit the reliability of their signals

Demo 2 — Comedy duo: coordination + roles

Setup
Two agents alternate with a theme (e.g., coffee)

Short-term memory: build on prior line
Simple strategy
Turn-taking + style guardrails = coherent output
Agentic view
Roles steer behavior

State = last few lines

Coordination emerges without hard-coding jokes

From toy to real workflows

Same loop scales to
research helpers

data pipelines

classroom tutors, graders, lab partners
Design levers
Explicit roles and goals

Tight feedback signals

State logging and resets

Responsible use & guardrails

Validate feedback channels; detect contradictions
Constrain roles, tone, topics (age-appropriate themes)
Log interactions; reset state between runs

Try it yourself

Web app
Number guessing + comedy duo