Enterprise AI

I created structure where there was none. All design tickets were produced in joint sessions with developers and the Tech Lead, with objectives, constraints, and decision criteria documented before opening Figma. I sent weekly reports to the business with the status of the design, progress, and dependencies that needed external decisions. It wasn't bureaucracy, it was the only way to make design work visible in a high-noise environment.

I built a local component library. The corporate design system provided the foundations. But a conversational AI application has components that do not exist in a conventional enterprise design system. The prompt input, the token counter, the inline governance alerts — each was designed to answer two questions simultaneously: what does it do functionally, and what does it communicate in terms of trust and accountability.

I treated governance as a feature, not as a warning. I distinguished two types of governance messages: contextual guidance before action, and informative alerts afterwards. The image generation disclaimer, for example, appears at the right moment, with microcopy that communicates responsibility without interrupting the flow of a user in a high-pressure context.

I adapted the point of intervention when the process changed. When formal ceremonies disappeared, I started reviewing the solutions that developers were building. Instead of specifying beforehand, I evaluated suitability during the process. This adaptation was not a loss of ownership, it was a decision about where design still had real impact. I made the decision not to proceed. The business voice pitch adjustment feature was identified in the discovery phase as a high-value differentiator. However, the product architecture was not clear enough to implement it without creating technical debt. I documented the concept and deliberately deferred it, rather than forcing a solution that would need to be redone.