Nishita Nirmal
Back to all work
AI Fluency

How I Use AI in My Design Process

AI shows up at every stage of my design process, not as a single tool bolted onto one step. The clearest live proof is the Resource Library App, an in-progress build where the "How I used AI" log is being written stage-by-stage, in public, as it happens.

AI across the process, stage by stage

These are the same seven stages already tracked on the Resource Library App page. Each line below is a real, specific note on how AI actually fits that stage in my workflow, not a generic "I use AI throughout" statement.

  • Research & problem framing: Before designing Mercury Ask, I used ChatGPT to compile a list of the interface patterns that document-grounded tools use to build trust. I then decided which of them fitted the brief. Inline citations and a reasoning drawer went in. Verbatim quote blocks did not. [[PLACEHOLDER: drafted from the Court Documents Manager case study; confirm or replace]]
  • Information architecture: The structure of the case studies on this site was worked out with Claude from my M.Des reports, and then reorganised around decisions instead of project phases. [[PLACEHOLDER: drafted; add how AI fits IA in client work, if it does]]
  • Wireframes & flows: [[PLACEHOLDER: no evidence on the site yet. Say whether AI touches your flows at all, e.g. pasting a flow in to find the missing edge cases before hand-off, or whether this stage stays fully by hand in Figma]]
  • Visual design & prototyping: The three D.E. Shaw portals on this site were recreated from my original exports in Claude Design, so that confidential products could be shown with placeholder data. Their clickable prototypes were built the same way. [[PLACEHOLDER: drafted; add what Claude Design generated versus what you refined by hand]]
  • Build & execution: This site was built with Claude Code from my Claude Design export and my design guidelines. That includes the encrypted password gate, the case-study components and the preview tooling. Brometheus, the household app, uses Claude for book lookup, medicine lookup and free-text parsing. [[PLACEHOLDER: drafted; confirm]]
  • User testing: [[PLACEHOLDER: no evidence on the site yet. Your tests so far were think-aloud sessions with real users. Say whether AI plays any part, e.g. synthesising notes or drafting scenarios, or whether it deliberately does not]]
  • Iteration & outcome: I asked Claude to review how I used it across one working session, and published the result unedited as the self-audit. [[PLACEHOLDER: drafted; add how you iterate with AI on client work]]

Tools

Claude Strategy, writing, research.
Claude Code Implementation.
Claude Design Prototyping. See Comp Planner, Margin Simulator, Profit and Loss Portal, Compass.
Figma
Framer Two custom components built for a personal-website case study.
Wispr Voice dictation for faster AI collaboration.

A real example, self-audited

Self-Audit: How I Use Claude is a real example of how I evaluate my own AI usage: I asked Claude to analyze one actual working session and report back unedited, strengths and gaps included. A designer auditing their own tool use honestly is itself part of the AI-fluency pitch.

The thesis, argued in full

The FAQ's free-range exploration essay makes this same case at more length; points 5 and 7 in particular are direct statements of the AI-fluency thesis: that AI closes the gap between ambition and execution, and that the more AI absorbs the execution layer, the more valuable qualitative judgment becomes.