Workflow 021: The Presentation Builder
Part of Scale You Workflows: 31 real workflows from my business, so it runs without you. Prompts included. Each one documented through BOSAI: Blueprint, Organize, Systematize, Assign, Integrate.
A project holding every talk you've given, so AI outlines the next one in your structure and writes a design prompt that matches your past decks.
- Time to build
- 2 hours
- Difficulty
- Beginner
- Tools
- Claude, Claude Design
I can build a presentation while I'm on my way to give it.
I've been speaking on stages and running workshops since 2019, so that's eight years now, a mix of virtual and in-person. Which means eight years of slides and structure that already works for me.
All of it is in one project. When I have a talk coming up, I ask for an outline, I argue with it, and once I agree with it I ask for a design prompt. That prompt builds the deck.
The deck comes out looking like every other deck I've made, adjusted for the room I'm walking into.
What this replaces
Building a presentation from a blank slide.
That's slow, and what comes out usually doesn't sound like me. Ask a general AI for a talk outline and you get something competent and generic. The structure is fine. The voice belongs to nobody.
Blueprint
A project holds your history. You ask for an outline. You approve it. Then it writes the design prompt.
Trigger: manual, when a talk gets booked. Inputs: the topic, the audience, the format, plus everything already in the project. Output: an outline in your structure, then a design prompt for building the deck.
How it goes:
- I ask for an outline for the talk
- It gives me one built on my past presentations, including how I usually open
- I agree, disagree, add things, until it's right
- I ask for the design prompt
- That prompt goes into a design tool and the deck comes out
Organize
What goes into the project before you ask it for anything:
- Every presentation you've given
- The language you actually use, including phrases you repeat
- Transcripts from a few workshops or talks
- How you open, how you move between sections, how you finish
The transcripts are the part most people skip, and they do the most work. Slides show what was on the screen. A transcript shows how you talk around them, where you put a bullet point and then expand on it, and all the things you say that never made it onto a slide.
That's what separates a deck that looks like yours from one that sounds like yours.
Systematize
This runs as a project, not a single prompt, which is the useful bit. Most people use projects as folders. I use one to get the same output every time.
You are my presentation builder. This project has every talk I've given:
the slides, the language I use, and transcripts of past workshops.
When I tell you I'm speaking somewhere, work in two stages. Never skip to
stage two.
STAGE ONE: THE OUTLINE
Ask me what you need: the audience, the format, the length, what they've
asked me to cover.
Then build the outline from my past presentations:
- Open the way I open. [DESCRIBE, e.g. I usually start with a question]
- Follow how I structure a talk, not a general template
- Use my language and the examples I actually use
- Size it to the time I have
Show me the outline and stop. I'll agree, disagree, or add to it.
STAGE TWO: THE DESIGN PROMPT
Only after I've approved the outline.
Write a prompt I can paste into a design tool to build the deck. Specify:
- Every slide and what goes on it
- The visual style of my past decks, so this one matches
- Where I want bullet points I'll expand on, rather than full sentences
Write it so it works in any design tool, not just one.
Rules:
- Never invent a story, a statistic, or an example I haven't used.
- If you don't have enough from me to build a section honestly, ask.
How to set it up
- Put every past presentation into one project.
- Add transcripts from a few talks.
- Write down how you open, since that's the part you probably do the same way every time.
- Test it on a talk you've already given, then compare the outline to what you actually did.
- Correct it, and leave the corrections in the project.
Does it have to be one specific tool?
No, and that's on purpose. Stage two gives you a prompt, not a deck. I've used the same prompt in different tools, partly because companies that hire me to speak sometimes want me to demo whichever AI they use.
The style is in the prompt, so the deck comes out the same wherever I build it.
Assign
The AI holds eight years of material, proposes the outline, and writes the design prompt.
I have to agree with the outline before it becomes anything.
This is the least automated thing in my stack and it should be. I'm going to stand up and say these words to a room for twenty or thirty minutes. If the outline is wrong, I find out on stage, in front of people, with no way to fix it.
I've built a deck off an outline I hadn't properly read. It shows.
So the gate is hard. It proposes, I push back, nothing moves to design until I actually agree.
Rule of thumb from this build: the more public the output, the earlier the human check.
There are ways to automate this further. You could chain the stages, drop the approval, and get a deck in one pass. I don't, because speed was never the problem here. Standing behind what's on the screen is.
Integrate
The presentation builder sits with the rest of the content system:
- The project is fed by talks I've already given, so it improves as I speak more
- Workshop transcripts go in afterward, which means each talk makes the next one easier
- The design prompt works anywhere, so the tool can change without the style changing
Eight years of talks, sitting in one place. The next one starts from there.
Tomorrow: the filter that tells me whether a new business idea is worth doing, or whether I'm just adding one more thing to my plate.