What this toolkit is: A practical companion to Day 8 of Learning AI Out Loud. Two exercises to move you from understanding the centaur model to living it — one for everyone, one for people building AI systems.
Yesterday we talked about unlearning — and why the strongest performers resist AI the hardest. Today we move from the problem to the possibility. The centaur model reframes the entire human vs AI conversation — not replacement, not competition, but combination. These two exercises help you move from understanding the concept to living it.
In freestyle chess tournaments, the winners weren't the best computers or the best human players. They were average humans who were exceptionally good at working with computers — knowing when to trust the machine and when to override it. Chess players called them centaurs. Half human, half machine. Better than either alone. The centaur model is the most useful mental model for working with AI — and these exercises help you find your own centaur opportunities.
The winner isn't the best human or the best AI. It's the best human working with AI.
Most people either use AI as a full replacement or don't use it at all. The centaur model lives in the middle — where you deliberately combine what you bring with what AI brings. Complete these four fields for one specific task you do regularly.
The Task
One specific thing you do regularly. Not a category — a specific task.
Not: "communication" · Yes: "Write the weekly status update for my team"
What You Bring
Your specific human contribution. Think about: context only you have, relationships and history, judgment calls, accountability, tone, nuance — the things AI cannot replicate for your version of this task.
What AI Brings
The specific capability AI adds. Think about: speed, structure, options you wouldn't have thought of, consistency, breadth, pattern recognition, drafting without fatigue.
The Experiment
One specific way to combine both — this week. Not a vague intention. A concrete action with a clear sequence.
"I will draft the status update myself in bullet points capturing the key context and judgment calls, ask AI to structure and expand it into a full update, then edit the output for voice, accuracy, and anything AI missed."
What to notice when you run it: Where did the combination produce something better than you would have alone? Where did your judgment change or improve the AI output? Where did AI surprise you? Those observations are your centaur data — they tell you where to experiment next.
🎯
Your Challenge For This Week
Run the experiment you just designed. Don't wait for the perfect task or the perfect moment. Pick something small, try the combination, and observe what happens. The centaur model is learned by doing, not by reading about it.
Most AI workflows are designed at the extremes — either fully automated or fully reviewed. Neither is the centaur model. Pick one AI workflow you currently own or are building. Answer these three questions.
Question 01
Where is AI making decisions that should involve human judgment?
Look for moments where: context matters significantly, accountability needs a human face, ethical considerations are in play, or nuance genuinely changes the right answer. These are the places where full automation is a risk, not a feature.
Question 02
Where are humans reviewing AI output that doesn't need review?
Look for: high-volume, low-stakes, consistent outputs where AI accuracy is reliably high and the cost of an occasional error is low. These are the places where human review adds friction without adding value — and where you're underusing AI.
Question 03
What does the centaur version look like?
Redesign the workflow so AI handles what it's genuinely better at, humans handle what they're genuinely better at, and the handoff points are deliberate rather than defaulted. Define three specific handoff points:
2
Human hands back to AI when:
3
Final human review before:
What to notice: The three handoff points you identified are your centaur architecture. They're the difference between an AI system that replaces human judgment and one that amplifies it. Build those handoffs deliberately into your next AI deployment.
🎯
Your Challenge For This Week
Take the redesigned workflow you just sketched and share it with one person who works on that system. Ask them: does this feel right? Where would you draw the handoffs differently? That conversation is worth more than any framework.
What you leave with
- One centaur experiment designed and ready to run this week
- One AI workflow redesigned with deliberate human-AI handoff points
- A practical understanding of where you specifically add value that AI cannot replicate