Gbiz Ai Work Log

---
name: gbiz-ai-work-log
description: Keeps a running AI Work Log while you work, producing dated entries of what you asked, what came back, what you checked and what you changed, plus a weekly look-back with one change to how you brief or check. Use for "run gbiz-ai-work-log", "log how I used AI", "keep a record of my AI use", "show how I used Claude on this project", "AI usage log for my portfolio", "prove I checked the AI output", "start my work log", part of the Claude for Business Graduates Pack by Polar Bear.
---

# AI Work Log

## When To Use
You want to be able to show how you used AI, not just say it. Three weeks from now an interviewer will ask "what did it get wrong?" and you will not remember. This log answers that question while the work is fresh, one short entry at a time.

## When Not To Use
If the work is finished and you need a disclosure line, use the AI Use Statement on whatever records you have. If you want to draw lessons from the whole project, use the Project Reflection; the log is the raw record, not the story.

## Inputs
- The chat or prompt you just used, or a one-line summary of it
- What came back, and what you did with it
- The file or version name where the result lives
If you have none of this, I start from today's task in one sentence and mark the entry as reconstructed, not contemporaneous.

## Approach
A lab notebook habit, structured by the four competencies of the AI Fluency framework (Rick Dakan and Joseph Feller with Anthropic), taught in Anthropic's AI Fluency for students course: Delegation (why AI here, what you kept), Description (how you briefed it), Discernment (how you judged what came back) and Diligence (what you own, what data went in, what needs disclosing). The weekly look-back borrows the action-plan stage of Gibbs' Reflective Cycle (University of Edinburgh Reflection Toolkit). The failure it prevents: writing the log at the end from memory, which turns it into a flattering story nobody can check.

## Workflow
1. Ask three questions: what you just did with Claude, which file or version holds the result, and whether the work falls under your employer's AI policy or your university's rules (if unknown, the entry flags it to find out).
2. Write the entry the same day, two to five lines: date; task; Delegation (why AI, what you kept for yourself); Description (the brief, or where it is saved).
3. Record one line on what came back, then Discernment: what you checked, how (source traced, number recomputed, logic tested) and what was wrong. "Nothing checked" is a valid entry; leaving it blank is not.
4. Record Diligence: what you changed and why, what data went in (confirm no personal or confidential data, or under which permission; check with a qualified adviser on data protection), and whether this use needs disclosing.
5. Name the evidence: the file version or screenshot you keep. I never create evidence; you save it.
6. Once a week, run the look-back: read the entries, name the pattern in what Claude got wrong, and set one change to how you brief or check next week. One change, written as an action.
7. Skip entries for trivial uses (a synonym, a spelling). One entry per meaningful interaction; a log too long to read later is busywork.

## Output Format
```markdown
# AI Work Log
Project: [name] | Rules that apply: [employer AI policy / university rules / to find out]
## Entries
| Date | Task | Delegation | Description | What came back | Discernment | Diligence | Evidence |
|---|---|---|---|---|---|---|---|
| [date] | [task] | [why AI, what you kept] | [brief or link] | [one line] | [checked how, what was wrong] | [changed, data in, disclose?] | [file or screenshot] |
## Weekly look-back
| Week | Pattern in what went wrong | One change next week |
|---|---|---|
| [week] | [pattern] | [action] |
## Decision
[You decide which entries need disclosing and run the look-back by [day each week].]
```

## Done When
- Every entry is dated the day the work happened, or marked reconstructed
- Every entry has a Discernment line, even if it says nothing was checked
- Each entry names a piece of evidence you keep
- The look-back ends in one action, not a list

## Quality Bar
- Two to five lines per entry; clarity beats completeness.
- Other people appear by role only, never by name.
- No confidential employer, client or personal data goes in the log or the chat.
- The log never grades you; it records what happened to the work.
- The log records what really happened; nothing is added after the fact.

## Next
Run gbiz-portfolio-case-study (Portfolio Case Study) to turn the finished work into a page employers read.

## About the makers

This pack is made by Polar Bear, a consultancy built by ex-McKinsey founders with a dream to make AI work for People, not instead of them. We help our clients build people systems and AI-first ways of working, and we run our own company on Claude. If your team has outgrown the self-serve version, message Pauline (linkedin.com/in/paulinebertry).