Practice Tools for Product Work
Exercises, copyable prompts and a reusable skill for practising product judgement with AI.
The handbook describes how to investigate a problem, test a product and operate it. These exercises let you try parts of that work with a real task or a fictional example. Choose one that fits the decision in front of you; there is no required sequence.
Find an exercise
| If you want to... | Start here | You will need |
|---|---|---|
| Examine customer evidence | Inspect customer evidence | A question and a small set of notes you may share, or a fictional example |
| Challenge a proposed feature | Challenge a product bet | The problem, evidence, proposed solution and constraints |
| Make a portfolio trade-off | Choose with limited capacity | Two opportunities, an existing obligation and a capacity constraint |
| Understand information in an AI task | Follow the information | A familiar task and, if available, a knowledgeable colleague |
| Practise defining quality | Draft and calibrate an eval | A task, its permissions and someone who can judge the work |
| Investigate an AI failure | Inspect an incomplete trace | The fictional trace provided, or an appropriately shared real one |
| Develop a competency | Make a practice plan | One capability, a task and a colleague who can give feedback |
The customer, discovery, evaluation and practice-plan exercises include text you can copy into an assistant. Replace the bracketed sections with your own context. The planning and trace exercises can be done without AI; that is often a useful way to form your own view before asking for a challenge.
Use the material with care
Share only material you are authorised to use with the tool and account you have chosen. Remove identifying or sensitive details where appropriate, and use fictional material if you cannot share the real work. A prompt cannot grant permission to access information or make a consequential decision.
Treat the output as a draft for inspection. Check claims and quotations against their sources, separate observations from inferences, and look for missing or contradictory evidence. A generated test case remains synthetic even after it is checked against real work; a generated threshold is a suggestion, not an approved release criterion.
A reusable product-decision skill
If you regularly assess product bets, a reusable skill can give an assistant a consistent workflow: frame the decision, distinguish evidence from assumptions, compare alternatives, examine the operating consequences and propose a test. The product-decision skill file contains those instructions.
Read the file before using it. You can copy its instructions into a conversation, or save the file in a product-decision folder and follow your tool's instructions for adding a skill if it supports that format. Different tools discover and apply skills differently; a file alone does not guarantee that a tool will load it. No external connections or scripts are required by the file itself.
Start with a small decision and check whether the result reflects the evidence you supplied, names a credible alternative and proposes a test that could change the decision. Ask the assistant what it assumed. Revise the instructions for your own context and review them again whenever the workflow or responsibilities change.
For consequential work, use the output to prepare a discussion with the accountable people. It does not replace customer research, domain judgement, security controls or a release decision.
v3.1 · Updated Oct 2026