orchids is an AI infrastructure company building the next generation of intelligent agents. the team behind Stack AI pivoted to focus on orchids.app, a platform for deploying and managing AI agents, and bud.app, a general-purpose AI agent that can use a computer autonomously.
Bud
leading the design engineering of Bud, a general-purpose agent that uses a computer the way a person does. it browses the web, navigates applications, completes multi-step tasks on someone's behalf. my work covers the full product surface: the interaction model, agent feedback loops, and the visual language that makes autonomous computer use feel transparent and trustworthy.
the core design problem is trust. an agent that can act on someone's behalf only earns its job if the person watching it can read what it's doing in real time, and step in without breaking the task. i shaped three layers around that: an observation layer that surfaces the agent's reasoning at the speed people actually read, an intervention model that lets users redirect mid-task without forcing a restart, and a session model that scopes "what the agent is currently doing" cleanly enough to pause, hand off, or resume.
Bud also runs through iMessage and Telegram. designing the same agent across a full UI and a chat thread forced sharper choices about which signals were essential to the experience and which were artifacts of the canvas. that constraint ended up making the web product better in return.