The Generic Training Problem
Many schools and hospitals respond to AI by booking a one-hour general presentation about it. Staff leave without knowing how to apply AI safely to their own daily work, because what AI is good for depends on the specific workflow.
If you are an administrator bringing AI training to your organization, here is how I structure a workshop.
1. Separate the Lanes Immediately
In a healthcare or clinical education setting, the very first thing an AI workshop must teach is the Two-Layer Architecture.
Staff must understand the difference between Lane 1 (administrative, educational, non-PHI) and Lane 2 (clinical, PHI, secure). If you do not establish these boundaries in the first 15 minutes, staff will either be too terrified of compliance to use the tools at all, or they will accidentally paste patient data into a public LLM. (This is general information, not legal advice.)
2. Teach Prompt Engineering as Clinical Reasoning
Do not teach "prompt hacks." Teach structured reasoning.
Healthcare professionals are already trained to use structured communication (like SBAR: Situation, Background, Assessment, Recommendation). Teach them to prompt AI the exact same way. Give the AI the Situation, provide the Background constraints, ask for an Assessment, and demand a Recommendation format.
Staff already know the structure, so there is less to learn.
3. Search Your Own Documents
For most organizations, the most useful demonstration is searching their own documents, not writing emails.
A workshop should include a live demonstration of tools like NotebookLM or secure enterprise instances of Gemini. Show the staff how to upload a 200-page institutional policy manual and query it for specific procedural answers. Staff see the time it saves on their own material.
4. Address the Hallucination Risk Head-On
A workshop must spend real time on the hallucination problem (AI confidently inventing facts).
Teach staff how to use citation-based engines like Perplexity for research, rather than generative engines like ChatGPT. Teach them the "Trust but Verify" workflow: use AI for synthesis, but demand primary source citations for any clinical or policy claim.
Hiring the Right Speaker
When hiring someone to lead this workshop, look for someone who has worked in clinical practice as well as with the technology. A developer can explain how a model works. A PharmD or clinical educator can explain how to fit it into a busy practice within HIPAA constraints.