Move from instructing AI to produce output, to using AI to think through a problem — while the human still owns the decision.
This track moves a team from instructing AI to produce output, to using AI to understand a problem, test assumptions, generate options, and learn — with the human still fully responsible for the decision. By the end, a learner frames a problem clearly, separates fact from conclusion from uncertainty, uses dialogue with AI to stress-test their own thinking, and produces a reasoned, evidenced recommendation.
The approach P1 uses throughout: from asking for an answer → to framing the problem → to testing assumptions → to comparing options → to deciding, with an owner and evidence.
Every session follows the same arc: connect to real work → teach the concept → demonstrate it with and without context/evidence → practise it → apply it to one real, approved task → reflect and set the risk-checked homework.
Define what AI may help think through and what stays the human's responsibility.
Turn the symptoms of a problem into a solvable question — statement, stakeholder, outcome, constraint, non-goal.
Separate what you know, what you believe, and what still needs proof. Confidence is not evidence.
Use AI as a constructive challenger: counterarguments, alternative explanations, disconfirming evidence, a pre-mortem.
Generate and compare an option set against shared criteria, and mark which decisions are reversible.
Use AI to find gaps in information — source hierarchy, triangulation, recency — not to decorate false confidence.
Design context hygiene — versioned, owned, with a decision log and an expiry — so thinking continues without stale or wrong information creeping in.
Run an after-action review — expected, actual, cause, learning, next experiment — using AI to review work without offloading responsibility.
Create a decision log and disagreement log the team can review, with uncertainty stated explicitly.
Decide what AI or source to use under real data constraints: least disclosure, permission, anonymisation, an audit trail.
Turn an idea into an experiment with a hypothesis, a leading indicator, a safeguard, and a stop condition.
Bring problem framing, evidence, options, and reflection together into one proposal — a recommendation is not the same thing as an AI answer; it needs a decision-maker and a next action.
The difference between a tool and a thinking partner is what you ask for. Ask for an answer and you get an answer. Ask it to separate fact from assumption and question you back, and you get shared thinking. The prompt below puts weeks 3 to 6 on a single page.
This prompt is vendor-neutral and works in Claude, Gemini or ChatGPT. What makes it work is the assigned role and the ordered sequence of thinking, not the brand.
What differs is making the conversation persistent. Week 7 calls this an AI Brain; in practice that is Gems if you are in Gemini, and Projects if you are in Claude or ChatGPT. Same idea, different names and different plan-dependent limits.
Check the feature name against the screen in front of you before you start, so you do not hunt for a menu that product does not have.
Context: I am about to make a decision, and I want you to help me think, not help me write. Task: before offering any conclusion, work through these four steps in order 1) separate what is a fact I gave you, what is an assumption of mine, and what nobody yet knows 2) ask me three questions back — choose the ones whose answers would change your answer most 3) offer three options, each with the trade-off it requires me to accept 4) tell me what evidence, if I found it, would make your recommendation wrong Constraints: do not agree with me out of politeness. If my framing is wrong from the start, say plainly where it is wrong. Format: four headings matching the steps above, no more than five lines each. My situation: [write your situation and the decision you face, 5-10 lines]
This prompt lets the AI push back on you; it does not make the AI accountable for you. The conclusion remains yours, and personal, customer or financial data still does not go into an unapproved tool.
Facilitation rule throughout: use only anonymised or constructed P1 examples; pair learners as task-owner and checker; measure decision quality and reusability, never the number of prompts or apps tried; and if a use case touches sensitive data, finance, personnel, or system access, stop and escalate through P1's approval channel.
Instructor note: open the official references below before teaching the related session, and match button names and plan tiers to whatever P1 has approved on the teaching day.
This track assumes the working fluency built in Track 01, and prepares a team for the tool-selection judgement in Track 03.
This is the curriculum P1 Thailand built and uses to train its own team, before teaching it to anyone else.
Twelve weeks, one shared foundation, applied to each person's real work.
Request a consultation