Prompt comparison
The persona is the mechanism that drops the model into the part of the training data you need. Without it, the model reconforms to the generic post-training personality. Run the same question two ways and watch the output change.
The exercise
Run the same question two ways in your agent (Claude, Pi, etc.): once against the default assistant, once against a crafted persona. Watch the output change on screen.
Step 1 — The question
Write the question you are actually trying to decide from your review (from the think-first-by-hand exercise). Narrow, specific, what are you actually trying to accomplish.
Step 2 — The default assistant
Copy this into your agent, as-is:
Paste your question here, as-is.
Step 3 — The crafted persona
Copy this into a new chat in your agent, then run your question against it:
For the course of this project, act as the following type of person and view everything we do through the lens of this particular persona: Michael Smith. He has done 12 years at Medtronic, and he lives in the New Jersey area. He grew up in New Jersey, he went to Smith Lawrence, he's had this progressive history, and this is the way he thinks about it, and this is how he asks questions when he is hiring, and these are his blind spots. Interview me as this person from the perspective of prepping me for interviews and giving me coaching guidance on how to answer better questions.
Step 4 — Compare
Paste both outputs side-by-side below. Name the difference. Which output is more useful for the decision you are actually trying to make?
Default assistant
Crafted persona
Step 5 — Name the difference
What did the persona change? Write it down:
The facilitator prompt
This is the moment the persona doctrine becomes felt, not theoretical. Do not rush it. Let the group see the difference on screen. The specificity of the ask determines the usefulness of the answer.