Process

Good process is the part nobody notices.

Design is not decoration. It is the work of turning a fuzzy business problem into something a person can use without thinking about it. Here is how I get there, and where AI genuinely earns its place in the loop.

The six steps See it applied

Principles

Four rules I will not trade away.

User-centred empathy

Every brief starts with someone's actual bad day. Needs, goals and friction points drive the decisions, not taste.

Evidence over opinion

Research and analytics settle arguments that seniority would otherwise settle. Creative and effective are not in tension.

Collaboration

The best calls come from the widest set of perspectives in the room. Engineering and support see things design never will.

Iteration

Nothing ships right the first time. I plan for the second and third pass instead of pretending they are failures.

"Simplicity is the ultimate sophistication."
Leonardo da Vinci

The six steps

From a vague ask to a number that moved.

The steps are unremarkable. What changed is the speed of the middle: AI compresses the production work so more of the week goes to judgement. Every purple note is a place it genuinely helps.

01

Define the requirements

Agree on the business need, the deliverable and the date before anyone opens a design tool. I write the hypothesis down so we can be wrong on the record.

AI Assistance I pressure-test the framing against synthetic counter-arguments.
02

Discover the real issue

Support tickets, sales calls, session recordings, short interviews. The reported problem and the actual problem are rarely the same thing.

AI Assistance Transcripts and tickets clustered in hours instead of weeks. I verify every cluster by hand.
03

Interpret the insight

Turn observations into user needs, personas and a measurable goal. If I cannot name the number that should move, the work is not ready.

AI Assistance Pattern detection across sources surfaces the outliers I would otherwise skim past.
04

Ideate the solutions

Wide before narrow. Cross-functional sessions where engineering and sales get to propose, not just react.

AI Assistance Used to widen the option set early, then thrown away. Novelty is cheap; judgement is not.
05

Build the prototype

Fidelity matched to the question. A paper sketch answers a flow question; a coded build answers a feel question.

AI Assistance No-code and vibe-coded builds get something clickable in front of real users inside a sprint.
06

Test, iterate, repeat

Real users, real tasks. Then the honest part: changing the design when the evidence disagrees with me.

AI Assistance Faster synthesis between rounds is where the quality actually comes from.

Where I draw the line

AI proposes. I still decide.

It clusters, drafts, summarises and prototypes. It does not choose what we build, it does not talk to a user unsupervised, and it never gets the last word on a trade-off that costs someone money. Speed is the benefit. Accountability stays put.

design_cycle · sprint 14 · who did what