Core program
Foundations
Become AI-fluent. Stay human.
Fifteen modules of general AI fluency: how models think, prompting, evaluation, agents, grounded research, and a one-page playbook you leave with.
The modules
- 01 beginner 28 min
How AI actually thinks
A large language model predicts the next chunk of text from probabilities. It does not look things up. Get that one fact and most of the magic and the failures make sense.
mental-modelfundamentals - 02 beginner 34 min
The anatomy of a real prompt
A real prompt has five parts: role, task, constraints, examples, and an output spec. Miss one and quality drops. Stop chasing magic phrases and learn the anatomy.
promptinghands-on - 03 beginner 26 min
Choosing the right model
Claude, Gemini, ChatGPT, and NotebookLM are different tools, not brands of the same thing. Picking by habit costs you quality. Match the tool to the job.
tool-selectionfundamentals - 04 beginner 40 min
Reading the output critically
Hallucination, confabulation, and sycophancy are the three ways AI output misleads you. Learn to run a 90-second checklist and treat fluency as a red flag.
verificationcritical-thinking - 05 beginner 32 min
The first ten prompts you'll reuse
Most of your AI work is a handful of repeated moves. Keep ten workhorses built on four cores: summarize, sharpen, draft-reply, and audit-doc. Save what works.
promptinghands-on - 06 beginner 38 min
Where humans must stay loud
Some calls are yours by law and ethics: hiring, reviews, sensitive comms, anything binding. Apply a three-question gate and keep AI as drafting support, never the decider.
judgmenthuman-in-the-loop - 07 intermediate 42 min
Working with your own documents
Stop re-pasting. Ground AI in your own documents with NotebookLM, Claude Projects, or Gemini Gems. Grounding cuts hallucination, but it is not a force field.
groundingyour documents - 08 intermediate 46 min
Agents, in plain language
An agent is a model that uses tools in a loop toward a goal: plan, act, observe, repeat. Useful when a task needs lookup plus drafting, but keep a human checkpoint in the loop.
agentsworkflows - 09 intermediate 36 min
Building an evaluation harness
"It feels better" is not proof. Set up a cheap eval: reference outputs plus an A/B scorecard in a Google Sheet. Change one thing at a time and let evidence decide.
evaluationhands-on - 10 intermediate 44 min
Designing a multi-step workflow
One mega-prompt rarely wins. Decompose a big task into focused steps, route each to the right tool, pick a model per step, and put checkpoints where an error compounds.
workflowsorchestration - 11 intermediate 38 min
Cost, latency, and the engineering reality
You pay by the token, both ways. Reason about budgets, caching, streaming, and smaller-model tradeoffs so you can make a sane cost, speed, and quality call and spot cheap wins.
costengineering - 12 intermediate 50 min
Your AI playbook: write it now
The playbook is the point, not a certificate. In session you build a one-page playbook: what to delegate, which models, and which checkpoints stay human. You leave with something you would pin above your desk.
capstoneplaybook - 13 intermediate 25 min
The Prompt Architecture: T.E.C.R.E.I.
Stop hunting for clever wording. A reliable prompt has six parts. Fill all six and the quality stops being luck.
promptingframeworkfoundations - 14 advanced 30 min
Advancing Prompting
Once the architecture is solid, these six techniques raise the ceiling: reverse, recursive, self-critique, deeper reasoning, debate, and creative temperature. Optional depth: be comfortable with the T.E.C.R.E.I. basics first.
promptingadvanced - 15 intermediate 25 min
Mastering AI Research
AI invents facts when it has none. Research mode fixes that by grounding every answer in sources you trust. Includes NotebookLM fundamentals.
researchnotebooklm
Build once, reuse everywhere
Your AI Context Document
Build a one-page document you can paste at the top of any AI conversation so you never have to re-explain yourself. Answer these in your own words, keep it to a page, and leave out anything private (no personal data about other people, no signed dollar figures tied to a named party). Update it when your role or work really changes, not constantly.
Open the Context Document builder →