A workshop at MAZ Journalistenschule in Luzern, part of their Recherche Tag, introducing journalists to terminal-based AI agents for investigative work.
What we covered
The session moved journalists beyond chatbot interfaces into the command line, where AI agents can run overnight research, access local files, chain tools together, and work in parallel.
I demonstrated real investigation use cases including:
- Document analysis — leaked files, FOIAs, financial records
- Transcription & translation of foreign-language sources
- Data extraction & structuring from messy PDFs into publishable datasets
- Background research and pattern detection across records
- Geolocation, verification, and network mapping
Live demo
I walked through our Assam investigation (a collaboration with The New Humanitarian) to show how these tools work on a real story — building a custom investigative agent with OSINT skills, running parallel research across land records, social media, and contracts, and mapping evictions against development patterns.
Hands-on
Participants installed Gemini CLI, Claude Code, and OpenAI Codex, then worked through exercises in data extraction, document analysis, and scrollytelling — building and deploying projects to the web by the end of the session.