I used to build all my presentations in Quarto. It was great for creating self-contained HTML files that anyone could open, on any machine. No dependencies, special software, and works with version control. Just a single, portable file.
My process has changed. Now, I code my presentations directly with AI, and it’s made me more productive by letting me focus more on what actually matters: the message.
Focus on Message, Not Mechanics
I don’t have to fiddle with layouts or syntax — I spend my time on higher-level questions.
What is the story I’m trying to tell?
What is the core value for the audience?
Which features are worth mentioning and which aren't?
How should my communication be structured.
AI handling all that lets me focus exclusively on content and narrative. The quality of most presentations is limited by the clarity of the message, not the aesthetics of the slides anyway.
My Presentation Workflow
My process is about getting a good first draft as quickly as possible. The faster I can get a tangible output, the faster I can iterate and be creative. Seeing an idea implemented gives me 5 more ideas.
Here’s what I do:
Voice Transcription: I start by talking. I transcribe a stream of thought for 5-10 minutes, describing everything I want the presentation to be. This includes the intent, the target audience, the story, the value proposition, and the feeling I want to create. Talking gets me to say more than just facts and descriptions and into describing why this topic matters and why and more opinions.
AI-Generated HTML: I feed this transcript to an AI and ask it to code a self-contained HTML presentation with embedded CSS and JavaScript. If I have more info I give it - maybe that’s a python file, or a screen recording of a feature, or other artifacts.
Rapid Iteration: I review the generated file in a browser. Seeing the result of my ideas is the best way for me to know if it’s a good direction. I can immediately see if it worked out like I hoped. If it doesn’t, I can re-order, re-phrase, and re-generate in seconds. I can do many more iterations in the same amount of time this way.
This process helps me explore more branches and see their impact instantly.
The End Result
This method produces a nice HTML file you can host on github pages or package up into a single file to share. Anyone with a browser can open an html file, not just devs.
For example, my workshop presentation on AI-assisted coding for Hamel and Shreya’s AI Evals course was created in this way: https://ai-evals-course.github.io/isaac-ai-coding-fasthtml-annotation-workshop/
Limitations and Workarounds
Of course, there are trade-offs. The generated code is verbose and not something you want to manually edit much. You have less direct control than if you were hand-coding everything from scratch, but get the results infinitely faster. Direct editing of text requires searching through the HTML file rather than clicking a text box.
However, I’ve found the benefits of rapid iteration and focusing on the core message far outweigh these limitations. For most presentations, this workflow is a more productive and creative way to build.
If you're an engineer looking to integrate custom, reliable AI workflows into your development lifecycle, you might be interested in our course, Elite AI Assisted Coding. We cover how to build universal AI setups, design agentic workflows, and apply enterprise best practices.