AI workflow

How I use AI in design work

AI helps me move faster through work that is mostly sorting, drafting, and documenting. The parts that depend on people, such as observing how work happens, deciding what to build, and taking responsibility for the result, remain my job.

Rules I keep

Four rules, every time

  1. Treat every AI output as a first draft that needs review.
  2. Verify facts and figures against their sources before sharing them.
  3. Be open with my team about where AI was used.
  4. Keep a person accountable for every decision that ships.

Where it fits

Step by step, AI drafts and I decide

Every one of these produces a first draft. Each one gets checked against the source, the people who do the work, or the build.

  1. Step 01Synthesis After shadowing sessions and interviews, I use Claude to sort notes into first-draft themes, then check each theme against the original notes and drop any the notes don't support. Themes
  2. Step 02Mapping the current state Ticket exports, SOPs, and side spreadsheets become first-draft journey maps and service blueprints, which I review with the people who do the work. Journey maps
  3. Step 03Exploring options Tools like Google Stitch help me explore a wider range of layouts early. I select and refine in Figma, where the design system lives. Layouts
  4. Step 04Prototyping with real logic When an interaction depends on calculations, real data, or many states, I build a working prototype in code with Cursor, so people test real behavior. Working prototypes
  5. Step 05Writing Drafts of microcopy, error messages, guides, and release notes, rewritten in plain language and checked against the product's own vocabulary. Plain-language copy
  6. Step 06QA Comparing builds against approved designs, checking contrast and accessibility, and drafting test cases for edge cases that are easy to overlook. Test cases

Where it doesn't

Some work stays human

Designing AI features

AI features people trust

At Comcast I designed Planning, which used AI to predict construction completion dates for a finance team that was understandably cautious about AI. The principles that helped there guide how I approach AI features.

Show what drives the answer
People trust a prediction when they can see which inputs move it.
Make uncertainty visible
A forecast without a sense of confidence can read as a promise.
Design for the wrong answer
Make it easy to question, correct, or override the output, and to report when it seems wrong.
Connect it to real work
Adoption grew when the predictions made resource planning easier.