From Idea to Interaction: How I Built a Facilitation Coach in Two Hours | Fred Deichler

Last week I wrote about stopping the generation cycle and starting to build. I talked about the three modes of AI: Generation, Automation, and Partnership. I challenged you to stop being a copy-paste API.

This week I took my own medicine. It worked validatingly fast.

The Distraction

I got distracted in a meeting today. Not the normal kind - the kind where a thought grabs you by the collar and won't let go.

The thought? Am I actually any good at facilitating?

Not just this meeting. Any meeting. I've been a Scrum Master for years, but that question sat there, nagging, while the conversation continued around me.

In my previous post, the "Generative Trap" was described - using AI to get point-in-time answers that never compound. The trap was recognized immediately: I could tab over to O365 Copilot, ask it to analyze my facilitation style, maybe get some tips. Friction.

The answer would be fleeting. Point-in-time. Tomorrow I'm back to zero.

So instead of generating, I decided to build.

The Low-Fi Capture

First, I had to survive the meeting.

A large sticky note was grabbed. The core problem got written down: "analyze transcripts for improvements." Then components were listed:

Calendar integration
Meeting roles
Transcript ingestion
Local database
Prompts with history
Outcome tracking
Weekly meta-feedback
Meeting types

Two minutes. The meeting got my attention back.

That's the human hack. When something steals your focus, exorcise it onto paper. Come back later when you have the ability to focus.

Voice Mode: Removing the Internal Editor

The meeting ended five minutes early. (Small wins.) I pulled up Idea Coach - a cool agent inside O365 Copilot - and started talking. "Hey, I want you to help me build an app that helps me improve my facilitation." Why voice? Typing activates my internal editor. I wordsmith instead of dump. As I wrote in The Vomit Prompt, voice lets you bypass the perfectionist filter. Dump first, refine later. Idea Coach took that messy brain-dump and helped me shape it into something useful. "That's a great idea!" I was told, because that's what AI does. Makes you feel good about things.

Once that good feeling wore off, it was asked to help flesh out and generate a PRD - a Product Requirements Document. Not because generating a PRD is the goal, but because the PRD becomes the handoff to the next step.

Building.

Into Cursor

I opened Cursor - the AI code editor I mentioned in my previous post - and pasted in the PRD. But first, a specific request. "Hey, can you make sure this uses Markdown formatting?" As I wrote in GitHub for Everyone, Markdown is the universal language of AI and humans. Readable by people. Parseable by machines. It bridges the gap between my messy thinking and the structured output I needed.

The PRD was taken by Cursor and building started. A Python app. A SQLite database. A simple analysis framework.

I'm not a developer. I've said this before. But as I wrote last week: if you can write a user story, you can now build software. I can read code well enough to know when something's wrong.

And something was wrong. The batch file didn't run.

The Edit Loop

Mode 3 in action: Partnership. I'm not just asking AI for answers - I'm building with AI. The result? A persistent tool, not a point-in-time response. Something I can hand to another Scrum Master and say "run this."

Here's the loop:

Describe what you want.
AI writes the first draft.
Try to run it. It breaks.
Copy the error message.
Paste the error back: "I got this error. Fix it."
Repeat until it works.

I didn't give up when it broke. I didn't go ask a different question somewhere else. That's the difference between generating and building - when you're building, you iterate. You partner.

Three or four rounds of back-and-forth were completed, and suddenly: a working app.

An app that ingests meeting transcripts and summaries from Teams. Meta-analysis gets run on my facilitation patterns. My role as an observer versus participant can be distinguished.

Two hours. Total.

What I Mapped

Let me break down what I actually did, using the framework from my last post.

The Repetitive Action: The meeting-after-meeting question of "how am I doing as a facilitator?" - something that nags but never gets addressed.

The Generative Action I Could Have Taken: Asking Copilot for tips every time. Point-in-time answers. Nothing compounds. Classic Generation mode.

What I Did Instead: The repetitive action was mapped, the generative trap was identified, and AI was partnered with to build something persistent.

In a couple hours, I went from an idea - a fleeting thought about self-improvement - to interaction. A tool I can run again. Data I can accumulate. Patterns I can spot over weeks, not moments.

The Shift Is Real

Most practitioners think AI proficiency means better prompting. Getting smarter about what you ask. Crafting the perfect query.

That's still Generation mode.

The shift is moving from consumer to creator. From transactional to persistent. From point-in-time answers to tools that compound.

You don't need to be a developer. I'm not. You need to be willing to describe what you want, iterate when it breaks, and partner with AI instead of just chatting with it.

I've got something rough I'm trialing for a couple weeks. I'll be honest - I don't know if this tool will survive a month. But that's not the point. The point is I stopped asking for answers and started building solutions.

If it sticks? I'm gonna share it with the Scrum Masters I work with.

Your Turn

What fleeting thought is stealing your focus today?

Write it down. Voice-dump it. Paste the PRD into Cursor or Windsurf.

Two hours from now, you might have something that lasts.

Stop generating. Start building.

Continue Your Journey

Stop Generating, Start Building: The full framework for moving from AI consumer to AI builder.

The AI-Assisted Scrum Master: A practical guide to partnering with AI in your daily Agile practice.