Shipped with
Building AI decision-support across MixShift’s Amazon analytics platform: onboarding ops, eval infrastructure, and the operator-grade interfaces that turn the model output into something seller and advertising teams open every day.The case study lands when the engagement wraps.
See the rest of the workLong engagements, inside your team, on your tools. Not a vendor who emails PDFs. I join your Slack, your standups, your codebase, and your on-call rotation if it matters. I stay long enough to make the systems ship, and long enough to hand them off cleanly.
Every AI system your team uses has to be operable by humans. A decade of design engineering at Disney, Apple, and a long list of startups before that taught me that information architecture matters as much as model selection. Dashboards, trace viewers, eval tools your team will actually open.
Six-month engagements end with your team owning the system. Code, prompts, infrastructure, runbooks. All yours. No lock-in, no retainer-for-life. If the engagement extends, it extends because it's delivering. If it doesn't, you're not stuck.
Execution got cheap, so building the wrong thing faster is the new failure mode. The job moved upstream, to what to build and for whom, before any code.
Founder LessonsFigma's on-canvas agent now details every state for you. That moves the value to your design system and your judgment about what belongs on the screen.
Founder LessonsMost UI bugs are design system bugs in disguise. The same source-of-truth discipline applies to prompts living in twelve agents. Fix the layer, not the leaf.