Case Study 02 · Speculative

Designing calibrated trust into an agentic copilot

An AI copilot embedded in a complex workflow tool, designed around trust the agent has to earn incrementally — not assumed on day one.

Role

Sole designer

Duration

2–4 weeks

Status

Scheduled — Weeks 3–4

Not yet built

Per the project timeline, this case study is scheduled for weeks 3–4, after Case Study 1's real pipeline capture. It will be clearly labeled speculative throughout, anchored to published patterns (Microsoft HAX, Google PAIR), grounded in a real competitor teardown (Claude Cowork/Computer Use, OpenAI Operator, Notion Custom Agents, Microsoft Copilot Studio), and include a small moderated usability test with directional — not fabricated — findings. Any synthetic research will be explicitly disclosed as synthetic.

Planned scope

Trust primitives to design: an Intent Preview with Proceed / Edit / Handle-it-myself, an Autonomy Dial, confidence signals, a transparency/reasoning feed, and undo/override. Target: a working interactive prototype slice embedded in the page, with a clickable-prototype-plus-narrated-video fallback if the working slice proves too heavy for the timeline.