AI Amplifies. Humans Decide. ← aiamplifies.org
Start here · Free

The Care Package.

Everything I hand out at sessions, in one place. A paste-ready configuration for your own AI thinking partner, the full workshop handout, sample data to practice on, and the research behind the talk. No signup, no email gate. Built for Total Rewards and HR people who want to start Monday.

60Seconds to set up
3Steps
0Code required

Downloads

Take what you need

Start with the care package. If you only want the walkthrough, grab the handout on its own.

Full Care Package↓
.zip · handout, starter configuration, exercise kit, sample data
Handout only↓
.docx · the Thinking Partner walkthrough
Student edition↓
.docx · the version I use with graduate students

What’s inside

  • Getting_Started_Four_Doors.docx — where the configuration pastes in Copilot, ChatGPT, Claude, and Gemini, plus the judgment upgrade and five tests
  • AI_Thinking_Partner_Handout.docx — the full walkthrough, personality options, test prompts, next steps
  • AI_Thinking_Partner_Starter.md — paste-ready configuration with bracketed fields to fill in
  • Geozone_Map_Exercise_Kit.zip — build an interactive geographic pay map from raw data
  • greenleaf_market_survey_data.csv — market survey data across three sources, fictional company
  • greenleaf_retail_census.csv — 50-person census with grades, salaries, compa-ratios, ratings
  • Lab_Care_Package_README.txt — what each file is and where to start

How to use it

  1. Open the starter file and copy the text between the cut lines.
  2. Paste it into your AI tool’s custom instructions. In Copilot that is Agents → New agent (works on a work or school account); in Claude, a Project; in ChatGPT, Custom Instructions or a Project; in Gemini, a Gem. The Getting Started guide in the package walks each one. If your tool has none of those, paste it at the top of a conversation instead. It still works.
  3. Fill in the bracketed fields with your role, your team, your systems, your priorities. That context is the whole thing.
  4. Ask it something you are actually working on this week. Then tell it what it got wrong.

Use fictional or sample data until your organization says otherwise. The exercise data here is entirely made up for exactly that reason. If you can see it, the AI can see it.

Receipts

The research behind the talk

Every study cited on screen, so you can read the source rather than take my word for it. Titles are searchable; I have deliberately left off anything I have not read myself.

Dell’Acqua et al. — Harvard / BCG, 2023Field experiment with consultants. Bottom performers gained 43%, top performers 17%.
KPMG & UT Austin, 20261.4 million real workplace AI interactions. Roughly 5% show sophisticated use.
BCG, AI at Work, June 202642% of frontline users save a full workday a week; 66% get little or no guidance on the freed time.
Doshi & Hauser — Science Advances, 2024AI-assisted work gets individually better and collectively more similar.
Wan & Kalman — Computers in Human Behavior: Artificial Humans, 2026Ten deliberately different AI personas restore the diversity that a single one flattens. Open access.
BCG / HBR, “Brain Fry,” March 2026n=1,488. Constant AI monitoring: +33% decision fatigue, +39% serious mistakes. Task replacement goes the other way.
Shaw & Nave — Wharton, 2026 (working paper)“Cognitive surrender.” When the AI was right people beat baseline; when it was wrong they did worse than people with no AI at all.
Budzyn et al. — The Lancet Gastroenterology & Hepatology, 2025Clinicians’ unaided detection skill measurably declined after routine AI exposure.
Cheng et al. — Science, 2026Americans rated a chatbot’s ethical advice more moral and more trustworthy than a New York Times ethicist’s. Sycophantic AI increases dependence.
PwC, Global AI Jobs Barometer, 2026Roughly one billion job ads across 27 countries. Judgment-amplifying roles grow about twice as fast, with materially faster wage growth.
Microsoft Work Trend Index, 2026Organizational factors leaders control explain more than twice as much of AI’s real-world impact as individual mindset.
Korn Ferry, 2026A majority of Total Rewards teams have not begun experimenting with AI.
Gallup, 202652% of US employees now use AI at work, roughly double two years earlier.
Anthropic Economic Index, 2026Over 90% of agentic usage is everyday knowledge work, not software development.
Kasparov, 2005 — freestyle chessHis own account: amateurs with ordinary computers and a better process beat grandmasters with supercomputers.

The one rule

AI amplifies. Humans decide.

Treat it like a sharp new hire whose work you still review. Verify the confident-sounding output, and ask it to argue against you on purpose. Keep sensitive employee data out of anything your organization has not approved. Then go build the thing you have been describing to people for years.