Starting in August 2026 Investigating with AI Agents
A five-session bootcamp for journalists who want to use AI agents for research and investigations—without losing evidence, provenance, verification, or editorial judgment.
Format 5 live sessions × 60 minutes
Cadence Two weeks
Outcome A real investigation lead, grounded in evidence, with prioritized next checks
Before class: setup instructions arrive by email. Bring one rough lead, question, or claim from your beat.
Five sessions
S1
Understand
Know what AI agents are for
- See the difference between chat in a browser and an agent that can work with files.
- Understand what a model is, and why local models matter for sensitive work.
- Learn how tools and MCP let an agent use services like search or monitoring.
- Use simple skill files so good reporting habits become reusable instructions.
Deliverable A plain-English map of the AI agent toolkit
S2
Check
Give every claim a confidence level
- Learn grounding: the agent must save and cite the files behind its answer.
- Use SIFT as a simple habit: stop, inspect the source, find better coverage, trace the claim.
- Ask the AI to produce source files, direct quotes, and links.
- Label each claim as high, medium, or low confidence based on the evidence.
Deliverable A source-backed claim check with confidence labels
S3
Monitor
Turn repeated checks into routines
- Learn Automations or Routines in the desktop app: jobs that run again later and surface alerts.
- Choose a page, social profile, council source, or beat you should not check by hand.
- Walk through Scoutpost: Page, Beat, Social, and Civic Scouts.
- Treat Scoutpost alerts as leads with source links and verification status.
Deliverable One routine and one Scoutpost alert pattern
S4
Remember
Build a reporting memory you can reuse
- Understand Markdown: simple files that keep notes, sources, and findings portable.
- Use OpenKnowledge to separate captured sources, provisional research, and stable knowledge.
- Connect the workspace to your agent through MCP while keeping the reporting files local and inspectable.
- Preview and lint the knowledge base so gaps, provenance, and unfinished work remain visible.
Deliverable An OpenKnowledge reporting memory
S5
Investigate
Use Spotlight to bring the system together
- See Spotlight as the wrapper around evidence, monitoring, and knowledge-base work.
- Bring the lead you refined during training.
- Turn it into questions, source targets, findings, uncertainties, and next checks.
- Leave with a real lead advanced, not a fake classroom exercise.
Deliverable A real lead with evidence and next steps