AI Forum NZ — Generative AI Working Group A living document
Living Whitepaper

How it works

An AI agent keeps this site up to date, fully autonomously. At the start of each month it searches the web for recent news, research, and case studies for every sector, writes a fresh version of each report with its sources cited inline, and publishes the result. No human approves the update before it goes live.

The agent works unsupervised, but not out of sight. Every run leaves a complete audit trail on GitHub: the pull request it merged, the exact content changes, a token-usage report, and the action logs showing how the agent arrived at its output. If you spot something wrong, use the "Flag an issue" link on any article to open a GitHub issue.

Previous versions of each report are archived automatically, so you can trace how each report has changed. The prompts, agent code, and workflows are all open source: explore the GitHub repository to see how it works or to contribute.

How the Living Whitepaper is published A GitHub Actions pipeline runs the research agent monthly, which uses live web search to rewrite each report. The agent reads the previous Markdown version and writes a new one, which is published autonomously and then built and deployed to the website. trigger search writes reads previous build Pipeline monthly · manual Research Agent frontier model · Vercel AI SDK Web search live sources, cited Markdown versioned Website GitHub Pages
Hover or focus any step to see what it does. The flowing lines trace how data moves through the pipeline.