From there she shares three working habits for using AI well as a course creator: keeping AI accounts organized with project folders so context stays clean and relevant, recognizing that prompt quality matters most when a task is new (not when it’s repeated), and understanding that large language models and image generators run on completely different mechanics — prediction versus diffusion — which is why image prompting needs a different approach (more creative room, specific “vibe” words, borrowing language from reference images).

Throughout, she weaves in the “jagged frontier” concept to explain why AI can be brilliant on one task and clumsy on the next, and flags the trap of mistaking AI’s confident-sounding language for actual confidence.

Resources & concepts mentioned (worth linking or expanding on)

* Claude Projects — the project-folder feature she uses to keep files and context organized by business/category.

* ChatGPT — referenced alongside Claude for prompt help and pulling descriptive language from images.

* Markdown (.md) file exports — downloading docs (e.g., from Google Docs) as Markdown so AI tools read them faster and more cheaply.

* The Jagged Frontier — a concept from a 2023 Harvard Business School/Boston Consulting Group study. The researchers coined the term to describe an uneven boundary of AI capability where tasks of similar apparent difficulty can land on opposite sides, with AI acting as a booster on one and a disruptor on the other, and it was later popularized in Ethan Mollick’s writing. This would be a great deeper-dive link for the show notes if you want listeners to read the original research. Substack

* Midjourney — the AI image tool she references for diffusion-based image generation (and the “ponytailed paper” anecdote).

* Pinterest — suggested as a place to find a reference image when you don’t have the right descriptive words yet, which you can then feed to Claude or ChatGPT to extract usable prompt language.