Post by Melissa Milloway

Learning Leader & Strategist | ATD Author | Speaker | LinkedIn Top Voice in Education | 115K+ Community

Building a 13 slide presentation used 25% of my weekly $20 plan Claude tokens on Sonnet 4.6 across Claude and Claude Code. The talk is for ATD 2026. It's a TED style talk about how my team at Amazon built something from a scrappy workaround that eventually scaled to the company. Here's how I planned with Claude. I started in regular Claude and planned out the copy, slide layout templates, and image placeholders. All of this was just text to tell Claude Code what to do. Then I moved over to Claude Code. In Claude Code I pulled my personal brand design system directly from Claude Design using the Huashu design skill. Huashu takes a prompt and produces interactive prototypes, animated slide decks, and motion design exports. It read my font pairings, color palette, spacing, and motion easing and adapted everything to a dark-background conference format. So the deck isn't a generic slide template. It's built on my brand, the same typefaces and accent colors I use everywhere. From there we built out all 13 slides from the spec I'd described. Headlines, supporting text, eyebrow verb labels that run through the whole story arc, reflection questions for the audience on every story slide. Then I started iterating: ➡️ I had big italic punchline lines that felt cluttered. Using the UI/UX Pro Max skill, which runs on 161 industry-specific reasoning rules and 67 UI styles, I switched to smaller sans-serif subtext instead to give the slides more breathing room. ➡️ I used Emil Kowalski's animation skill, which covers animations, design, and performance, to add entrance animations so content fades in sequence when each slide appears, with stagger effects on the multi-card slides. ➡️ I integrated photos, including a two-column layout and a title slide with a gradient overlay so the text stays readable over the image. ➡️ I asked Claude to remove widow lines in the headlines to keep what minimal text I have on slides readable. That's 25% of my weekly tokens, which I think is fair given that with planning the slides it got me pretty close to what I wanted. I believe planning and using skills are key in saving tokens, along with choosing a model that works well for the use case but isn't a token eater. Next up, I'm going to build the same deck using Claude Design to compare the two outputs and token usage using the same inputs. I'll document the differences. Do you plan before you build with AI? I'm curious whether you see a difference in your results and token usage when you do. #eLearning #InstructionalDesign #LearningandDevelopment #AIinLearning

Post content