Responsible Design/Human in the loop /safeguarding

# Designing AI you can actually trust: the Responsible Design Notes inside Artist Engine Room

*A blog post about the small column on the right-hand side of every workflow map — and why it might be the most important part of the whole system.*

---

## The thing on the right of every diagram

If you open any of the workflow maps in the Artist Engine Room project — the `.drawio` (XML) files for things like *Log a Note*, *Assign Work to a Reader*, *Track Artist Outcome*, or *Equity Reporting Snapshot* — you'll notice the diagrams are split into vertical lanes:

- **Literary Manager** (the human doing the work)
- **AI Assist** (where the AI helps)
- **Artist Engine System** (validation, saving, audit)
- **Database** (where things actually get written)
- **Responsible Design Notes** (the rightmost lane)

The first four lanes show *what happens*. The last lane — Responsible Design Notes — explains *why it's safe to happen that way*. That's the part I want to talk about, because it's where the ethics of the tool stop being a slogan and start being a design decision.

Each map carries the same four kinds of notes, in the same place, every time:

1. **Data Handling** — what data is touched, the lawful basis, who can see it, how long it's kept.
2. **Ethical Design** — the specific design choice that protects the human in this workflow.
3. **Error Recovery** — what happens when the AI gets it wrong (because it will).
4. **Human Oversight Checkpoint** — exactly what a person is reviewing, with what info, and roughly how long it should take.

Think of it as a safety label that travels with the workflow. If a diagram ever moves into code, the label moves with it.

---

## Why put this directly on the diagram (and not in a separate policy doc)?

Most organisations keep their AI ethics in a PDF somewhere and their workflows in another system entirely. The two never meet. So the engineer building the feature ends up making ethics decisions by accident, and the policy person writes rules about a system they can't see.

Putting Responsible Design Notes *inside the diagram itself* fixes that, in a small but stubborn way:

- The person designing the workflow has to fill the column in. **You can't ship the map without it.**
- The person reviewing the workflow sees the safeguards next to the steps they protect — not three clicks away.
- When the workflow changes, the safeguards have to change with it. They can't quietly fall out of sync.

It's a tiny piece of friction, on purpose.