The tools I actually use.

The ownership ladder

Use it whenYou need to level designers, or work out what to grow next.

1. L1, the solution. Given a problem, design the right solution.
2. L2, the workflow. Define how the whole workflow works, across screens and states.
3. L3, the problem. Find the right problem instead of solving what was handed over.
4. L4, the business outcome. Connect the work to adoption, retention or activation, and defend the choice.

Each rung contains the one below it. It measures the size of the question you own, which you can observe, not craft quality, which you can always argue about.

The three baseline questions

Use it whenAny significant project is about to start.

1. What metric will move?
2. What is the baseline today?
3. How will we measure it?

Ask after launch and you get an argument. Ask during scoping and you get a number.

The relief filter

Use it whenYou're choosing which UX fixes to ship first.

Will a customer or a support teammate notice this changed, and feel relief?

It rules out a lot of work that is genuinely better and genuinely invisible.

The conflict order

Use it whenTwo good designers disagree and neither is wrong.

Safety and accessibility › usability › consistency › brand coherence › local preference

When brand expression weakens usability, translate the brand's intent instead of copying its visual treatment. Read the note →

The nine stages of an AI action

Use it whenYou're designing anything where AI acts on someone's behalf.

1. Signal
2. →
3. Analysis
4. →
5. Recommendation
6. →
7. Action
8. →
9. Approval
10. →
11. Execution
12. →
13. Verification
14. →
15. Evidence
16. →
17. Recovery

Most AI design stops at Recommendation. The last four stages are where trust is earned or lost. Read the note →

Five automation classes

Use it whenYou need to decide how much an AI feature may do alone.

Four promotion gates

Use it whenPrototypes keep sneaking into your design system.

1. Explore. Anything goes, flagged experimental.
2. Validate. Evidence of a recurring problem, plus a working build.
3. Standardise. Becomes a documented pattern, with states, accessibility and metrics.
4. Componentise. Only if a truly reusable primitive is needed.

The default at every gate is no. Quality isn't the criterion; frequency is. Read the note →

Four files for AI memory

Use it whenYou use AI on anything that has to stay true for months.

1. PROJECT: what this owns and doesn't own.
2. MEMORY: what's true right now, including what's unresolved.
3. DECISIONS: ratified decisions, with reason and impact.
4. BACKLOG: what's next and what's blocking it.

End every session with one rule: do not record proposals as accepted decisions.

The ten-point screen review

Use it whenA screen comes in for critique.

1. Job to be done
2. Pattern match
3. What is working
4. Friction
5. Recommended pattern
6. Design-system dependencies
7. AI and trust considerations
8. Edge states
9. Validation hypothesis
10. Decision candidate

The case-study template

Use it whenYou're writing up your own work for a portfolio.

1. One-line result, with a number
2. Context and your exact role, four lines at most
3. The real problem: what people thought it was, and what it was
4. Three decisions: the options, the trade-off, your call
5. What changed, with how each number was verified
6. How you used AI, and where your judgment stayed in charge
7. What didn't work yet

Every case study on this site follows it.