How do you actually use AI in the design process?

Most designers use AI to make screens faster. That is the smallest part of what it is good for. Here is where it actually sits in my work, phase by phase.

Research. FullStory shows me what users do. Session recordings, rage clicks, the exact step where people leave. AI mines that data plus analytics and support tickets for patterns no human has time to find. I read the patterns and decide which ones are the money leaks. Machines find signals. Experience ranks them.

Synthesis. This is where Claude earns its place. Raw research goes in. I push it to argue with me. Stress-test the problem statements, attack my favourite direction, find the case I missed. The output is a ranked backlog of hypotheses, each scored by effort against expected impact. AI is a brutal sparring partner if you ask it to be one.

Design. In Figma, AI drafts let me explore 20 directions where I used to afford 3. Most die fast. That is the point. The winning direction earns its place by surviving comparison, not by being first. The final screens are mine, built into a system that scales.

Build. Cursor and Claude in the terminal turn approved designs into working code. This is why my AI engagements ship a running application, not a folder of mockups. One person, design through deployment, no handoff loss.

Testing. Real users try the prototype. AI transcribes every session and themes the patterns in hours instead of days. I watch the moments that matter and turn them into fixes before any build spend.

Notice the shape. In every phase the machine widens and accelerates. In every phase a human narrows and decides. Founders sometimes ask if AI will replace this process. It already is this process. The question is who is steering it.