I've shipped two products solo — ChartPilot, an AI chart-analysis tool for traders, and Yaptım mı?, an ADHD task tracker on the iOS App Store. Before that I spent 8+ years as a product designer on trading platforms and consumer apps. This is the exact workflow that lets one designer do the work of a small product team.
Stage 1 — Design like a designer
AI hasn't changed this part as much as people think. I still start in Figma: flows, wireframes, a small design system with real tokens. Two habits matter more when you'll build it yourself:
- Design with real data early. Trading screens with fake round numbers hide layout problems that real market data exposes immediately.
- Decide the data model while designing. Every card and list on the screen is secretly a database table. Naming them during design makes the build stage dramatically faster.
Stage 2 — Build with AI pair-programmers
I use Claude for architecture and complex features, Cursor for fast in-editor iteration, and Codex for parallel, well-scoped tasks. The skill that transfers from design is briefing: I prompt the way I'd brief a senior engineer — context, constraints, edge cases, acceptance criteria. Vague prompts produce vague products.
- Prompt like a brief, review like a design crit. I read every diff. AI writes the code; I stay responsible for it.
- Small loops win. Feature → deploy → click through it → fix. Never five features between deploys.
- Own your schema. Letting AI improvise the database design is the most expensive mistake a non-engineer can make. Design it deliberately, then have AI implement it.
Stage 3 — Ship on a boring, proven stack
- Vercel for hosting and deployment — push to deploy, preview URLs for every change.
- Supabase when I want auth + PostgreSQL + storage in one box; Neon when I just need serverless PostgreSQL.
- RunPod for GPU workloads when a product needs its own model inference.
- Analytics from day one — GA4 plus Vercel Analytics, with funnel events (sign-up, activation, purchase). You can't design a funnel you can't see.
What actually changes as a designer
Owning the code changes your design judgment. You stop drawing states that are expensive to build and start noticing cheap wins — an empty state, a loading skeleton, a better error message — because you'll implement them yourself in minutes. The gap between "designed" and "live" disappears, and with it most of the politics of product development.
FAQ
Which AI coding tool should a designer start with?
Start with Claude or Cursor. Claude is strongest at architecture, reasoning and multi-file changes; Cursor is excellent for fast in-editor iteration. Most AI-native designers end up using both, plus Codex for parallel scoped tasks.
Can you ship production apps without an engineering background?
Yes, with discipline: design the data model deliberately, review every change, deploy in small loops and use a managed stack (Vercel, Supabase/Neon) that removes infrastructure risk. My live products were shipped exactly this way.
How long does an MVP take with an AI-native workflow?
A focused MVP typically takes two to six weeks solo — including design, build, deployment and analytics — versus a quarter or more with a traditional handoff process.
Do you still write code by hand?
Constantly, but mostly as editing rather than authoring: AI produces the first draft, I review, adjust and refine — the same relationship an art director has with production work.
