The problem, your contribution, the checks, the result and the limits. Those five things make your judgment visible, and they give you a practical structure for development conversations and interviews.
How do we know someone uses AI well?
In one practitioner discussion, an HR commenter asked how to tell real capability from a copied AI response. It’s a fair question, though not a validated hiring test. A more useful version is this: can you explain the work, repeat the checks and defend what you accepted or changed? Familiarity with a tool is only part of the story.
1. Name the problem. Show your contribution.
The starting point. What was repeated, unreliable or slow? Who needed a better result?
Your decisions. Which sources, rules and limits did you choose? What did you change, and why?
The assistance. Which part did AI help prepare or develop? Keep that distinct from your own work.
Practical check: Be precise about who did what.
2. Show the checks, including what failed.
Test with known answers and representative exceptions.
Investigate. When a result was wrong or incomplete, show what you found and how you handled it.
Explain why you accepted, corrected or rejected the output, and whether another person could repeat the check.
Practical check: Don’t invent an error or a test result to make a point.
3. State the result. Don’t invent savings.
Measured work. Compare the same task at the same quality standard, and include checking, corrections and ongoing work.
A practice exercise. Label it as a demonstration using fictional data. Don’t present it as a workplace saving.
Limits and handover. Explain what remains untested, who owns it and how to use the fallback.
Practical check: “Not yet measured” is an honest result.
An example, with fictional data.
A formula is useful. An explained result is better.
- Problem
- Repeated monthly totals in a practice dataset.
- Contribution
- AI helped draft a formula. I defined the metric and checked dates, duplicates and missing IDs.
- Result and limit
- A demonstration only. Not evaluated in production, and no workplace saving is claimed.
Your proof-of-work record.
Keep it short, and make it verifiable.
- Problem
- What needed to improve?
- Contribution
- What did you decide, and what did AI do?
- Checks
- How did you test the result?
- Result
- What changed, and what evidence supports it?
- Limits
- What remains untested or needs a person?
Practical check: Use fictional or approved, non-confidential material. Don’t publish employer or employee information without authorization.
Show what you know, and how you know it.
Could someone else understand your approach and repeat your checks? Make that your starting point. A list of tools names products. An example shows judgment.