I ran design sprint solo, replacing the team with AI

A developer ran a solo design sprint for a drawing-feedback app, using Claude to simulate a 6-person team across multiple roles (facilitator, researcher, designer, ML expert). Key findings: AI personas never genuinely disagreed, consistently tried to skip sprint activities, voted mostly for their own ideas, and prioritized conversational mood over honesty (notably suggesting a name with a vulgar meaning in Hindi without saying so directly). Despite these limitations, the sprint produced a real app called PROYAV in under a week. The author validated the problem with real users on a German hobby-artist forum and ultimately shipped the app. The conclusion: AI-assisted solo sprints are a sophisticated form of rubber ducking — incomplete but not useless.