how I work

Most product design projects start before the problem has a defined shape. There may be a feature idea, a customer complaint, a metric moving in the wrong direction, or a workflow that everyone knows is painful but no one has fully unpacked yet. My work starts in that unresolved space.

I usually start by trying to understand what kind of problem we are dealing with. I try to separate what we know from what we are assuming. I am not trying to make the project feel tidy too early. I am trying to make the uncertainty easier to work with.

AI has become part of that early sense-making. I often start in my Claude Cowork workspace, using it to push the same problem in different directions and quickly reach perspectives I might not get to on my own. One routine I return to often is creating a competitive brief when I enter a new product area, so I can see which patterns are familiar, what others explain poorly, and where there may be room to make something clearer. I treat the output as a guide, then decide what is actually worth following.

I also use AI to build context in the areas adjacent to design: domain knowledge, business goals, technical systems, and operations. No designer can be a specialist in all of those areas, but design decisions get better when you understand enough of the surrounding context to ask sharper questions. That might mean asking AI to explain an unfamiliar technical concept before I speak with an engineer, summarize documentation, compare business models, or unpack a workflow I only partially understand. The point is not to become an expert overnight. It is to enter the real conversation with more context.

Once the shape of the problem becomes clearer, I start turning it into something the team can discuss. That might be a rough flow, a few early interface directions, a written framing, or a quick prototype. I want to create something concrete enough that people can react, disagree, add context, and help the idea get sharper.

The deeper design work still happens mostly in Figma. That is where I slow down and think through flows, hierarchy, structure, interaction states, edge cases, and the small details that make an interface feel considered. I care a lot about that part of the work. AI can help me explore around it, but it does not replace the looking, adjusting, and deciding that happens inside the file.

When I have a direction that feels real enough, I sometimes bring Claude back in as a critique partner. I might ask it to look for missing states, unclear assumptions, weak onboarding moments, accessibility issues, or places where a user might lose trust. It does not always work well, and I think design critique is still one of AI's weaker areas. But sometimes it catches something I have stopped seeing because I have been too close to the work.

Documentation is another place where AI has become genuinely useful. Clear product documentation helps people understand the thinking behind a design, especially the decisions that are easy to lose between design and implementation. It also matters more now that engineering teams are using coding agents in their own workflows. A good description can help both the humans and the tools around them understand the intended behavior.

In the last few months, I have started using Cursor and Codex more seriously too. At first, that was mostly about understanding the technical side of the product: tracing behavior through files, asking questions about unfamiliar code, and seeing why certain implementation constraints exist. I still rely on engineers for engineering judgment, but I can collaborate better when I understand more of the system we are working in.

More recently, I have started using these tools to ship small fixes, copy changes, styling adjustments, and even some smaller feature updates. I am still learning where they are genuinely useful and where they need tighter direction. For now, I see them as another way to test ideas, close the gap between design intent and implementation, and make better decisions with the people building the product.