Direct
Give the agent the context, purpose, audience, constraints, and immediate next step.
Guided workshop
How to collaborate with your AI agent while something is being built
“Today isn’t about finding the perfect prompt. It’s about learning how to work with your agent while something is being built.”
The focus
The focus is not learning how to code.
The focus is learning how to direct, review, react to, and improve what your agent is building.
Begin the workshopCore promise
The focus is not learning how to code.
The focus is learning how to direct, review, react to, and improve what your agent is building.
Part One: Set the context
“Today isn’t about finding the perfect prompt. It’s about learning how to work with your agent while something is being built.”
“Your agent is not a vending machine where you enter a prompt and hope the right thing comes out. It’s more like a fast junior builder who needs direction, feedback, and testing.”
The point isn’t that the agent is incapable. The point is that a good result comes from collaboration, not one giant request.
“I’m not going to build something while you watch. We’re going to move through the process together. I’ll explain each step, you’ll use that step with your own agent, and then we’ll come back together before moving on.”
“Give the agent enough direction to take the next useful step, not enough instruction to disappear for an hour.”
Explain that the goal is not to give the agent less information. The goal is to give it the right information at the right stage.
Part Two: Learn the full Build Loop
Show the complete process before participants begin:
Explain that this is a loop, not a straight line. You may move through it more than once.
Give the agent the context, purpose, audience, constraints, and immediate next step.
Ask the agent to create one small, visible piece.
Look at what actually exists instead of assuming it matches what you meant.
Describe what feels unclear, incorrect, missing, unnecessary, or off-brand.
Ask for a focused change while protecting what is already working.
Use the project as a real user and see what happens.
Make a decision: use it, share it privately, revise it once, or pause because of a real blocker.
“We’re going to walk through this loop together. I’ll teach one step, you’ll apply that step, and then we’ll regroup before moving on.”
Part Three: Guided workshop
The first direction should give the agent enough context to take the next useful step.
It should clarify:
The goal is not to explain the entire future project. It is to create a strong starting point.
Participants choose a small project to work on.
They can build:
Give them this starting prompt:
I want to build a small tool for [specific user]. The problem it should help with is [specific problem]. The user should be able to [one clear outcome]. For the first version, I want to keep this small and focused. Before building anything, ask me the most important questions you need answered.
Ask participants:
Part Three: Guided workshop
Once the agent understands the direction, don’t ask for the entire finished project.
Ask it to build one small, visible piece.
This might be:
“We want something we can review quickly. If the direction is wrong, we want to find that out before building five more layers on top of it.”
Participants ask their agent to propose the smallest useful version:
Based on what we’ve decided, propose the smallest useful version. Show me: - The core user - The main action - The inputs - The output - The basic steps or screens - What is intentionally out of scope Do not build the full project yet.
Then they choose one piece:
Build only [specific section, screen, or function]. Do not add extra features or polish yet. After building, tell me: - What changed - What assumptions you made - What I should test
Ask:
Normalize that the first build may not be right yet. That is the purpose of reviewing early.
Part Three: Guided workshop
Review means looking at the actual result before immediately telling the agent what to change.
Have participants review from the user’s perspective:
“Reviewing isn’t approving or rejecting. It’s noticing what is actually happening.”
Have participants use the project without revising it yet.
Review this as if you were the intended user. Write down: - What is clear - What is confusing - What feels unnecessary - What is missing - What does not match the original purpose
Invite a few quick shares:
Part Three: Guided workshop
This is where participants give the agent useful feedback without needing to diagnose the code.
Introduce the feedback structure:
Here’s what I’m noticing: [describe what feels wrong or unclear] Here’s why it matters: [explain the impact on the user or result] Please change: [name the specific section] Keep these parts unchanged: [list what is already working]
Example:
The opening section feels too abstract. The user doesn’t immediately understand what this tool will help them decide. Rewrite the opening so the audience, problem, and expected result are clear within a few seconds. Keep the current layout and first question unchanged.
Ask:
Part Three: Guided workshop
Revision should be focused, not a complete restart.
“A revision is not an invitation to redesign everything. It’s a chance to make the one change that improves the result most.”
Make the focused revision we discussed. Do not redesign or rebuild unrelated parts. Afterward, summarize: - What you changed - What stayed the same - Any new assumptions - What I should review next
Participants review the updated version.
Ask:
For the workshop, introduce a one-revision rule. Everyone gets one focused improvement before testing.
Part Three: Guided workshop
Testing means using a realistic example and seeing what happens.
It is not waiting until the end to discover that the project doesn’t work.
Participants should test:
Test this with the following realistic example: [user scenario] Walk through the experience and tell me: - What works - What is confusing - What breaks - Whether the result solves the original problem - The one highest-value improvement, if there is one
Participants then use the project themselves and compare the result with their original intention.
Ask:
Part Three: Guided workshop
Shipping does not necessarily mean launching publicly.
For this workshop, shipping means the idea has become a working version that can be used, shared privately, or evaluated honestly.
Give participants four possible decisions:
The tool works well enough for its intended low-risk use.
The tool is useful, but they want feedback before public use.
There is one clear improvement, but it does not prevent the current version from being useful.
A real missing decision, asset, or technical issue prevents responsible completion.
“Do not keep building simply because more building is possible.”
Evaluate this version against the original purpose. Tell me whether it is: - Ready to use - Ready to share privately - Worth one focused revision - Blocked by a real issue Explain the decision briefly. Do not suggest a feature backlog. Recommend only the next most useful action.
Each participant chooses their own shipping status.
No shipping status selected.
Part Four: Closing and share
Have everyone share:
The key reflection question:
“What did you learn about how to work with your agent that you can use on your next build?”
You could also ask:
“Which part of the Build Loop do you usually skip?”
That helps them identify their personal pattern.
“Before our next training, use the Build Loop on one small thing you actually need. Don’t ask your agent to build the whole project at once. Direct it, review early, give focused feedback, test it, and make a clear shipping decision.”
Their challenge is not to build something bigger.
It is to repeat the collaboration rhythm.
Final positioning
“Today isn’t about learning how to make AI code for you. It’s about learning how to direct, review, test, and improve what your AI is building so you stay the owner of the result.”
The workshop rhythm is:
That gives them the full map first, but they experience every part of the map themselves instead of watching you complete a separate demo.