From mocks to shipped

I’m proud of this one. I designed Omaru’s launch site, then directed an agent through the build, from the first rough mock to the production frontend. I made the design and engineering calls; the agent handled most of the implementation. This is how the session went.

Start by stealing from the product

I didn’t want to invent a second visual language for the website. I started by asking the agent to study Omaru’s design system: the tokens, the components, and the way we combine them in the app. That gave us a real starting point instead of a mood board and kept the landing page recognisably Omaru.

Nine mocks, then three

I asked for nine concepts, each built as a working page. Nine sounds excessive, but being able to scroll them mattered. A page can look convincing as a still frame and fall apart the moment you move through it. I cut the weak directions, then my cofounder and I picked pieces from three of the remaining ones. The final direction came from that mix rather than from one immaculate first idea.

Mocks into a pixel-perfect frontend

Once the direction was set, the work became much less glamorous: compare, correct, repeat. I’d already told the agent to build the scene with components from the product, but saying that once wasn’t enough. Most of my time went into checking the implementation against the concept and pulling it back whenever it improvised.

The mistakes were small but obvious once they sat together. One window came from the wrong part of the product. The taskbar tabs were inaccurate. The code streaming in the agent-work panel didn’t match the trace beside it. Agents are much better at frontend work than they were in the Sonnet 3.5 days, but they still need someone to notice when the details stop agreeing with each other.

The last twenty percent

The page follows one long day, from sunrise to two in the morning. As you scroll, the background moves from day to night to suggest the progress someone makes with Omaru. The agent produced the basic effect quickly. Making it feel right took longer: where exactly should the crossfade begin, which element should move first, and how much space should there be between each beat? “Make it smoother” wasn’t useful direction. I had to be specific about timing, easing, and stagger.

The part I love

Watching an idea leave my head and start working on screen is still my favourite part of the job. It’s why I learnt engineering in the first place. Agents shorten that loop considerably, but they haven’t changed what I enjoy about it: imagining something, making it real, and immediately seeing three things I want to try next.