The five capabilities every builder should know
A hands-on tour of the five core AI capabilities everyone should understand — creation, transformation, insight, interplay, and action — with live demos in Gemini, Google AI Studio, NotebookLM, and Claude in Excel. Mary and Sebastian walked through each capability with exercises attendees could run from their own laptops, and closed with a voice-interview prompt that turns the AI from answer-machine into thought partner.
Key takeaways
Five categories map the full landscape of AI work. Creation (something from nothing), transformation (one format to another), insight (meaning from data), interplay (AI as thought partner), and action (autonomous agents). Mary and Sebastian walked through each with a live exercise — vibe-coding a map in Google AI Studio, querying Hannibal's Alps crossing in NotebookLM, finding three altered words in Romeo and Juliet, and running an Interview Me prompt.
Non-engineers can vibe-code production-grade tools overnight. Sebastian reverse-engineered a Palantir-style map from a tweet using Claude Code and the Compound Engineering skill — live plane tracking, satellite data, traffic feeds — then pivoted it to a globe of live animal cams. A friend with zero coding experience built a customer-experience dashboard with churn prediction and email campaigns in a few days.
Interview Me surfaces blind spots before long-running tasks. One prompt — "Interview me in depth about what I'm working on, ask one non-obvious question at a time, then write a structured synthesis" — forces you to articulate assumptions before an agent runs autonomously. Sebastian uses it before vibe-coding projects, workshop curriculums, and long briefs. The short version works in any chat; the long version adds setup questions that narrow scope.
Tools covered
- Claude — Anthropic's AI assistant. Desktop, mobile, and web apps.
- ChatGPT — OpenAI's AI assistant.
- Gemini — Google's AI assistant.
- NotebookLM — upload documents, books, PDFs, YouTube videos, or podcasts; turn them into queryable sources with citations. Best for knowledge that isn't on the public internet.
- Google AI Studio — Google's playground for raw Gemini model access, including Build mode for one-shot generation of small apps and pages.
- Claude in Excel — Excel add-in; knows how to use Excel and how to do financial modelling.
Referenced
- Compound Engineering — Claude Code skill used for deep research during the Palantir reverse-engineering demo.
- Lovable — vibe-coding platform mentioned alongside Google AI Studio as an accessible code-generation tool.
- v0 — Vercel's AI code-generation tool, referenced as another option for non-engineers.
- Raycast — Mac productivity launcher. Sebastian uses its snippets feature to store and quick-launch frequently used prompts.
- Project Gutenberg — source of the Romeo and Juliet text used in the needle-in-a-haystack insight exercise.