A presentation for researchers and lecturers at Utrecht School of Economics: 40 minutes of presentation, 10 minutes of live demonstration, and 10 minutes of Q&A.
Live slides: https://usefinance.github.io/agents/
The opening slide has a QR code linking to the live deck, with matching frames for the QR at the top left and the Finance Utrecht logo at the top right.
- How an agent extends a familiar chat workflow with tools and feedback.
- The model, harness and working loop, explained with one task.
- Research and teaching examples with outputs people can inspect.
- Two kinds of failure: convincing wrong results, and consequential actions through terminal or file access.
- Student thesis use and misuse, with Utrecht University's assignment-specific AI Index as the policy frame.
The core sequence has 13 visual slides. Slide 14 is the live demo and slide 15 is Q&A. Five reference slides follow, with sources, current setup links, prompt templates, demo resources and additional use cases. Product documentation was reviewed on 23 September 2026; model rankings, fixed prices and performance promises are intentionally omitted from the teaching material.
Run a local web server from the repository root:
python -m http.server 8000
Use python3 if that is the Python command on your system, then open http://localhost:8000.
- Arrow keys or Space: navigate.
- O: overview.
- S or the footer's Speaker notes button: open the current slide's notes on the page.
- Slide 2's notes include a Copy prompt button for the live comparison.
- F: fullscreen.
- The footer links directly to Q&A and references.
The deck uses the existing Reveal.js 4.6.1 runtime, now included locally with its MIT license. Fonts use the system font stack. Slides, images and reference demo outputs load without a CDN or font service; external resource links still need an internet connection. Speaker notes open on the same page, so they work in browsers that block popups.
The demo uses real public data from FRED to chart US annual CPI inflation and the effective federal funds rate since 2019. A coding agent fetches the data, then writes and runs a Stata analysis. The comparison is descriptive and makes no causal claim.
- Presenter guide: the ten-minute run sheet and fallback.
- Demo prompt: a copyable task brief.
- FRED CSV snapshot: saved on 23 September 2026.
- Download script: refreshes the public CSV without an API key.
- Stata do-file: creates the prepared outputs.
- Prepared chart, analysis CSV, data audit and methods note.
Stata's built-in FRED API command requires a personal key. None is configured on this workstation, so the live route uses FRED's public CSV download and imports it into Stata. The saved snapshot makes the prepared result reproducible if the network is unavailable. The earlier synthetic-firm example remains in the repository as an optional offline exercise.
index.html Slide deck, styles and presenter notes
assets/compustat_sample.csv Earlier synthetic teaching data
assets/images/Finance.png Original Utrecht Finance logo
assets/images/slides-qr.svg QR for the public presentation URL
assets/vendor/reveal/ Pinned runtime and license
demo/README.md Presenter instructions
demo/prompt.txt Live task brief
demo/fred/ Public FRED snapshot, Stata analysis and outputs
demo/reference/ Earlier synthetic analysis
Before the update, the original repository was forked to https://github.com/tchew86/agents.
The original main commit is 1787a07f94df9b6109a3f1aad057efbb18887c5f. The fork and that commit preserve the earlier Claude Code workshop.
Earlier workshop inspiration: Scott Cunningham's mixtapetools, Mihail Velikov's master class, and Jukka Sihvonen's strategic-revision.