# Pmjob AI Work Sample

> Write up one small real task you did with AI as a one page work sample, with the brief you gave, what came back, what you corrected and the result, plus an optional prototype labelled as a prototype and a confidentiality check. Use for "run pmjob-ai-work-sample", "show how I work with AI", "AI work sample for an interview", "they want to see it not hear about it", "write up my AI task", "label my prototype", "can I show this in an interview", "proof of AI work", part of the Claude Guide for Product Interviews Pack by 90n.

- Skill: `polar-bear-org/pmjob-ai-work-sample` (Agent Skill)
- Install (CLI): `npx get-skill add polar-bear-org/pmjob-ai-work-sample`
- Raw SKILL.md: https://api.skillmd.com/api/skills/polar-bear-org/pmjob-ai-work-sample/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Productivity
- Author: polar-bear-org (https://skillmd.com/u/polar-bear-org)
- Updated: 2026-11-14
- Page: https://skillmd.com/skills/polar-bear-org/pmjob-ai-work-sample

---


# AI Work Sample

## When To Use
They will ask you to show it, not describe it. You have a spoken answer on how you use AI, and you want one small piece of work an interviewer can read in a few minutes and question you on. The question it answers: what is the one thing I can put on the table, and is it safe to show?

## When Not To Use
Not for the spoken answer across several examples: AI Fluency Answer builds that. A shipped project belongs in Portfolio and Project Walkthrough, and work the employer set belongs in Take-Home Assignment Plan. If you have not done a real task with AI yet, do one first: this skill writes up, it does not manufacture.

## Inputs
- One real task, small enough to explain in three minutes, and the decision it informed.
- The brief you actually gave, what came back, what you changed, and anything you plan to show (screenshots, data, documents) with who owns each.
If you have none of this, I start from your description of one task and mark the output as a first draft.

## Approach
The method is a work sample with its limits stated: one task shown end to end, with what AI did and what you did kept apart. It is a plain working practice with no named originator. The judgment: the corrections are the evidence, so they get the most space, and a prototype is a demonstration, never production. The failure it prevents is the polished demo an interviewer takes for a shipped product, until the second question shows it is not.

## Workflow
1. I ask up to three questions: which real task, and what decision did it inform; is any of the material owned by a current or former employer or a client; where will you show it, and what does that employer's AI rule say.
2. If the real task is confidential, you redo the same kind of task on public or made-up material, and the sample carries the label "recreated on public material". The task stays one you really did; only the material changes.
3. I arrange the one page in a fixed order: the task and why it mattered; the brief you gave, trimmed; what came back; what you corrected or rejected, and how you could tell; the result; the limits (what this does not show, what real use would need); what AI did and what you did. I work only from what you paste. I do not write a brief you did not give or a correction you did not make.
4. Optional artifact: a small prototype made in Claude Design, or a short analysis. Its first screen or first line carries a visible label: "Prototype for discussion. Not production. Built with Claude on [date]." The write-up can live as a document in Claude Docs, or in any file you choose.
5. Confidentiality check as a table: every item (data, screenshots, names, numbers, internal documents) is marked cleared, recreated or removed. Anything owned by a current or former employer stays out unless cleared. No customer data, no real names in screenshots, no colleague's work shown as yours. Contract and confidentiality questions: check with a qualified adviser.
6. Presenting order for three minutes: the decision it informed first, then the brief, then the corrections, then the limits before anyone asks. You show it when asked, or offer it where that employer's AI rule allows, and the label and the "what AI did" line travel with it.

## Output Format
```markdown
# AI Work Sample Write-Up
Label: [real task, or recreated on public material] | Date: [date] | Made with: Claude | Status: [first draft or ready to show]
## The task and why it mattered
[Two or three sentences: the task and the decision it informed.]
## The brief I gave
[My actual prompt or instructions, trimmed.]
## What came back, and what I corrected or rejected
| What came back | What I did | How I could tell |
|---|---|---|
| [output] | [corrected, rejected or kept] | [the check I ran] |
## The result and its limits
[What changed, a number only with a source I could show. Then what this does not show and what real use would need.]
## What AI did and what I did
AI did: [parts]. I did: [parts].
## Prototype or analysis (optional)
First screen or line reads "Prototype for discussion. Not production. Built with Claude on [date]." Link or file: [where]
## Confidentiality check
| Item | Kind (data, screenshot, name, number, internal document) | Status: cleared, recreated or removed | Note |
|---|---|---|---|
| [item] | [kind] | [status] | [who cleared it, or what replaced it] |
## Decision
You decide whether this sample is ready to show, and to which employers, by [date].
```

## Done When
- The write-up fits one page in the fixed order, with at least one correction or rejection and how you could tell.
- Every item in the confidentiality check is cleared, recreated or removed; none is left open.
- Any prototype or analysis carries the label on its first screen or line.
- You can present it in three minutes, decision first.

## Quality Bar
- One task, one page. A second task is a second sample.
- The brief shown is the brief you gave, trimmed and never improved after the fact.
- Limits are written by you before an interviewer finds them, and a result with no source stays in words.
- A recreated sample says so at the top, in the label.
- The sample shows a task you really did, labelled with what AI did and what you did; a prototype is labelled a prototype.

## Next
Run pmjob-target-company-list (Target Company List) to decide where to take all this.

## About the makers

This pack is made by 90n ([90n.ai](https://90n.ai)): AI-native ways of working and tools for Product, Marketing and HR teams, built by ex-McKinsey founders. We map where time and money leak, redesign the work around AI and stay until your people run it themselves, and we run our own company on Claude. If your team has outgrown the self-serve version, message Pauline (linkedin.com/in/paulinebertry).

