Progress with meaning
Replace vague loading states with clear operational stages users could understand.
Designing and building a multi-sided AI concierge that could contact 20+ vendors at once—with traceable states, human intervention, and verification when automation became uncertain.
01 / Context
The product needed to translate a customer’s open-ended request into coordinated outreach across many independent vendors. Availability changes, replies arrive asynchronously, and an AI agent can misunderstand context. The safety principle was straightforward: keep every consequential state traceable, surface uncertainty, and give operators explicit paths to inspect, override, verify, or recover the workflow. That same interaction model is relevant wherever AI operates in regulated or high-stakes environments.
02 / Decisions
The product made the agent’s work visible enough to build confidence while keeping the customer’s journey calm. The operator experience focused on intervention, context, and recovery rather than simply exposing raw agent logs.
Replace vague loading states with clear operational stages users could understand.
Design intervention, override, and verification patterns as primary product behavior.
Use one shared status language across customer, vendor, and internal surfaces.
03 / Outcome
The redesigned live-status experience reduced the booking journey to four steps and made the parallel vendor workflow legible to both customers and operators. The current draft uses résumé-verified outcomes; interface evidence and deeper iteration history will be added later.
Next case study
ContentOS