Materials for a 30–40 minute seminar (plus ~20 minutes of discussion) on agent-driven software development, given at the University of Oxford. The audience is PhD students, postdoctoral researchers, research software engineers, and faculty in computational geoscience.
slides/— the presentation, built with Slidev. Entry point:slides/slides.md.docs/seminar_notes.md— the full seminar notes: talking points, tool details, Oxford-specific access information, workflows, failure modes, demo script, and discussion questions.docs/d8_prompts.md— the staged prompts used in the live demo.
cd slides
npm install
npm run dev # dev server at http://localhost:3030
npm run build # static build
npm run export # export to PDFThe seminar includes an optional live demo that builds
d8demo — a tiny educational Python
package for D8 flow routing and flow accumulation on synthetic digital
elevation models — from a clean scaffold, using the staged prompts in
docs/d8_prompts.md. The demo starts from that
repository's demo-start branch, with the finished package on
demo-solution as a safety net. d8demo is intentionally minimal and
educational; it is not a production hydrology package.
In the spirit of the talk, these materials were themselves built with the tools they describe: Claude Code, ChatGPT, OpenAI Codex, and GitHub Copilot. AI generated substantial portions of the design and text; the author directed and supervised the work throughout, curated and verified all outputs, and is responsible for the final content. See the disclosure slide at the end of the deck for details.
Licensed under the Apache License 2.0. You are welcome to reuse and adapt these materials for your own teaching, with attribution.