Take a real or hypothetical AI feature and redesign transparency, opt-in and human override — as if you had to present it to the team tomorrow.
Choose an AI feature: a chatbot, personalised suggestions, automatic summary, "smart" filter. It can be from your product or an app you use (e.g. e-commerce, bank, healthcare).
Document the current state: does the user know there is AI? Can they disable it? Is there a human path? Note every gap with an annotated screenshot.
Redesign three key moments: (1) Disclosure on first use — what it says, where it appears, tone; (2) Opt-in/opt-out — granular toggle, clear language; (3) Human override — visible button, indicated response times, receipt confirmation.
Add an error state: what does the user see when AI gets it wrong or hallucinates? Include ability to report and correct.
Write 3 bullets to present the redesign to the PM: current problem, proposed solution, impact on trust/compliance.
Before/after sketch (even low-fi) with disclosure, control, override and error state — plus 3 bullets for stakeholders.
Which redesign element was hardest? What does the team sacrifice if it implements everything? Which single improvement would you ask for as a quick win?