All AI decks I ever created were shit

I tried all the tools. Gamma, Claude Code, Google Slides. They all look sloppy. Way too much text. Images that don't fit the context of the slides. Titles that don't say anything. Until I had an idea on how to build awesome slides. And the results blew my brains out.

I was preparing a pitch for a project that I presented at OpenAI in September and I hate building slide decks. I am not a great designer. I can tell if something looks good. But I am incapable of doing it myself. So, what did I do? I procrastinated (obviously).

Then I had a session with my co-founder where we talked vision. And he had the idea of our product creating slide decks for startups based on best practices. My first reaction was: "Dude, these AI generated decks are so shitty. I don't know."

But then my brain started working. What if we don't ask AI to create slides? What if we ask it to design websites in slide format? I tried that in Claude Code and holy shit did it work.

Last week I showed how to do this in a free session. Not in Claude Code, but in Claude Design, because it's easier and takes most of the work away. It gets you 80 to 90% of the way. It's in every Claude plan, even the free one. And it turns out the tool is not the trick. What you give it is.

Building slides is the last step

Before you ask for a single slide, you prepare three things.

An outline. What you actually want to say, slide by slide. Two or three sentences per slide: the idea, what the room should take away. Not a teleprompter.

Your principles. Standing rules for every deck. One of mine: "A slide must be absorbable in about five seconds. If it takes 30 seconds, the audience reads it instead of listening, and finishes reading after you have moved on, so you lose them twice."

Your template. Your company's fonts and colours. Otherwise AI makes them up, and AI is not that good at colour schemes.

Then a short prompt pulls it all together. I always say 90% is planning, 10% is execution.

In the session I gave Claude Design the same outline and the same prompt twice. Once without my principles, once with them.

Without them, I still found it very ugly. With them, it was already much better than anything I ever created with Gamma. And the reason is not that the prompt was better. The principles were there.

Every failure becomes a principle

What I see a lot: a slide comes out wrong and people type "change this, change that". That fixes one slide. On the next one, AI makes the same mistake again.

So don't tell it what to do. Tell it why. Not "put the title at the top", but: "I expect titles to be at the top every time. Make this a standing rule." Every failure becomes a principle, and that principle carries forward. Your principles and your template you build once. The outline you write for every talk.

Everything from the session is in one folder: the prompt, my slide principles, an example outline, and my slide skill for Claude Code if you'd rather work the way I do. And you can watch the recording here.

Try it on your next deck and tell me how it went. Just hit reply.

Best

Myles

PS: Join the Slack to ask me anything.