AI-driven prototyping: A/Bing with synthetic users at SketchForge

AI · 5 min read

SketchForge, a prototyping toolmaker, experimented with synthetic users powered by behavior-modeling AI to run rapid A/B tests on early UI concepts. Designers fed common user personas and heuristics into the system and used the AI agents to simulate thousands of sessions across variant prototypes. The goal was to identify dead-ends, usability regressions, and accessibility gaps before investing in full development or recruiting participants.

The synthetic runs flagged issues like unreachable CTA states, keyboard focus traps, and poor color-contrast combinations. In one case, an agent-based simulation showed a three-step account recovery flow that 18% of synthetic users couldn't complete due to an unexpected validation rule. Engineers patched the rule and reran tests, delivering faster confidence than waiting for scheduled usability sessions.

However, SketchForge's team also learned the limits of synthetic agents. While AI caught structural problems, it missed emotion-driven behaviors and cultural context that real users brought to interviews. The company now uses synthetic testing as a preliminary filter to prioritize which flows need human testing, reducing the human research load by about 40% while keeping final validation grounded in real-world sessions.