Skip to content

README.md

Latest commit

 

History

History
71 lines (48 loc) · 4.2 KB

README.md

File metadata and controls

71 lines (48 loc) · 4.2 KB

Coding with AI Agents: A Hands-On Workshop for Scientists

Workshop materials for the 2026 Interdisciplinary Science Summit, hosted by Schmidt Sciences and the University of Washington Scientific Software Engineering Center (UW SSEC).

This repository contains the slides, demos, hands-on notebooks, and instructor notes for a ~2-hour workshop teaching scientists how AI coding agents work and how to use them well in research software engineering.

Quick start (participants)

  1. Make sure your GitHub handle has been added to the team 2026-viss-ai-workshop-participants.
  2. Click Code → Create codespace on main at the top of this repo. The first build takes ~3 minutes.
  3. When VSCode opens in your browser, wait for postCreate.sh to finish (you'll see a green "Done" line in the terminal).
  4. Open blocks/01-landscape/demo/notebook.ipynb and follow along.

Detailed setup, troubleshooting, and pre-workshop checks are in docs/setup.md.

Workshop structure

The workshop is organized into four 30-minute-ish blocks. See docs/workshop-outline.md for the full prose outline.

# Block Folder Format
1 The AI Coding Agent Landscape blocks/01-landscape/ Slides + demo notebook
2 How It Actually Works (post-training & tool calling) blocks/02-how-it-works/ Slides + model-swap notebook
3 Agent-Driven Research Software Engineering blocks/03-research-loop/ Slides + live Claude Code demo
4 Build Your Own Skill (capstone) blocks/04-build-a-skill/ Slides + hands-on chat-mode build

Each block folder follows the same layout:

blocks/0N-name/
  README.md            # what this block teaches, learning goals, timing
  slides.md            # Marp-flavored slides
  instructor-notes.md  # speaker notes, demo script, fallbacks
  resources.md         # curated further reading
  demo/                # runnable code, starter files, notebooks

Tools used

  • GitHub Copilot in VSCode / Codespaces: the deep-dive tool through Blocks 1, 2, and 4.
    • Block 4's capstone uses custom chat modes (.github/chatmodes/*.chatmode.md); the workshop ships two worked examples in .github/chatmodes/ for participants to read and remix.
  • Claude Code VS Code extension + the rse-plugins plugin, used in Block 3's live demo to show the structured research-plan-implement workflow. Same Codespace, different chat panel.
  • Claude (via the workshop's LLM proxy server): the model backend. Copilot, the notebooks, and Claude Code all hit the same proxy. The notebooks use the litellm Python SDK so the same agent loop also works against any other model the proxy fronts (GPT, Gemini, ...), change one constant.
  • Python 3.12 + uv: environment and package management.
  • Marp: slides as markdown.

You don't need anything installed locally: everything runs in a GitHub Codespace.

Repo conventions

  • One folder per block, reusable layout.
  • All Python work goes through uv (uv run pytest, uv add <pkg>, etc.).
  • Notebooks are written to be read top-to-bottom as a teaching document, not just executed.
  • Slides are written for the Marp for VS Code extension (preinstalled in the Codespace). Open any slides.md and click the preview icon, or run Marp: Export Slide Deck... from the command palette to produce HTML/PDF/PPTX.

Following day: office hours

VISS members will be available the day after the workshop to help you try the agent on your own data, write your own skills, and explore agentic research workflows. Bring a snippet of code or data you'd like to work on.

Instructors

Anant Mittal, Carlos García Jurado Suarez, Anshul Tambay, Vani Mandava, Robert Bates, Ryan Hausen, Eric Liu, Tina Dang, Arfon Smith.

Contact

License

MIT, see LICENSE (to be added).