AI Workshop • 30 minutes

Build Your
Agentic Workflow.

Most people use AI like a search engine — one prompt, one answer, inconsistent results. We’re going to flip that. You’ll leave with a reusable workflow that turns AI into a team.

Hook
Bad usage
System
Demo
Action
AI?
Part 1 • Hook

Would you use one tool
for everything?

Obvious in real life

One tool to build a house.
One tool to cook.
One tool to fix a car.

Different jobs need different tools, different methods, and different kinds of thinking.

Build
a house
Cook
Fix a car

So why are you doing that with AI?

Name the shape of the fix

Two words
that change everything.

Before we fix how you use AI, let’s name what you’re building. The rest of the workshop is just filling these in.

Agentic workflow
AI that breaks a task into roles and passes outputs between them — instead of answering in one shot.
Mini swarm
A small team of roles — planner, builder, critic, improver — coordinated by you to produce stronger results.
Punch

What makes you different from AI?
Nothing — unless you structure how you think.

Part 2 • Destroy bad usage

Be honest:

You ask once
You take the answer
You move on

That’s why it feels
hit or miss.

Hard truth

Most people using AI right now
are just faster at being average.

Why?

no structure
no iteration
no system

AI isn’t confused.
Your instructions are.

Part 3 • Teach the system

AI works best as a team.

Stop treating it like one magical brain that should somehow know what you meant.

Planner

Figures out the approach before action.

Builder

Creates the first working version.

Critic

Finds flaws, holes, and weak reasoning.

Improver

Makes the final version sharper.

SYSTEM
Planner
Builder
Critic
Improver

The loop:

Plan → Build → Critique → Improve

This is the part people remember because it explains why better output is not magic.

This loop is intelligence.

Why this works:

multiple passes
refinement
structure
1 pass
gives an answer.
4 passes
build a result.

A company wouldn’t hire
one person to do everything.

Even if they could...
they wouldn’t.

Because:

too many approaches
wrong method at the wrong time

Raw capability

Can do a lot. Knows a lot. Still unstable if you force everything into one pass.

Structured capability

Same intelligence. Better role separation. Better outcomes. Less randomness.

Capability without structure =
inconsistent results

You don’t need smarter AI.
You need a better system.

Part 4 • Live demo

Bad usage

Prompt:
“Explain photosynthesis”

Input

Too broad. No audience. No format. No objective.

Result

Usually correct. Usually generic. Usually forgettable.

Structured system

You are a structured AI system.

When I give you a task:
1. Plan it
2. Build it
3. Critique it
4. Improve it

Give only the final improved result.

Task: Create a 1-page study guide on photosynthesis.
FINAL
Plan
Build
Critique
Improve

Same AI.
Different structure.

Mini swarm

planner
builder
critic
Not one brain.
A coordinated process.
Part 5 • Student action

Try this now

assignment
topic
project

Give AI a task

Make it real. Something you actually need help with.

Use the structure

Plan it. Build it. Critique it. Improve it.

Compare the result

See what changes when the process changes.

Keep iterating

Do not stop at the first decent answer.

Your takeaway

Copy this.
Use it everywhere.

Paste this at the top of any conversation. Replace the task. The roles stay. The structure stays. Watch what changes.

Reusable prompt

The agentic workflow template

You are my agentic workflow.

For every task I give you:
1. Plan — outline the approach.
2. Build — create the first version.
3. Critique — find flaws, gaps, weak reasoning.
4. Improve — produce the final, sharper version.

Show only the final improved result.

Task: [your task here]
Final close

Beginner:
“Give me the answer”

Advanced:
“Let’s improve this”

Mic drop

AI doesn’t make you powerful.
Your system does.