A 65-slide AI-agent lecture, from one conversation

Selected slides

Cover

Part 1 — how LLMs work

From chatbot to agent

Part 2 — the model landscape

Token cost, made tangible

Part 3 — a self-hosted assistant

The workspace is a folder

AGENTS.md — the onboarding note

The task

The goal was a ~60-minute lecture on AI agents for a general audience — three parts: how LLMs and agents work, the current landscape of models and products, and a hands-on walkthrough of a self-hosted assistant.

Over a single working session, Xiaohe interviewed for audience and length, researched the 2026-10 model landscape, planned a 65-page outline, generated 49 consistent watercolor illustrations, wrote every page as HTML, and exported an editable PPTX plus a PDF preview.

The deck is self-referential: a worked example inside it uses the deck itself to make token cost tangible — roughly 20k tokens, about one mao — so the audience can point at the slides and feel the economics.

How Xiaohe did it

• 1. Interview

  Confirm audience, length, visual style, and deliverables.

• 2. Research

  Verify the 2026-10 model landscape and pricing from live sources.

• 3. Outline

  Structure three parts into a 65-page narrative with a clear through-line.

• 4. Illustrate

  Generate 49 watercolor illustrations in one consistent style.

• 5. Build

  Write each page as HTML, then export to editable vector slides.

• 6. Deliver

  Export the editable PPTX and a PDF preview back to the project folder.

Skills used

• ppt-agent
• gen-image
• HTML → SVG → PPTX

Deliverables