Cue Match Agent: Designing an AI Teammate for High-Stakes Metadata Work

This feature adds an AI matching agent to the Soundmouse Partnerships flow - proposing cue-to-rights matches automatically, but never applying them without review. The core bet: a wrong suggestion needs to be cheap and obvious to correct, since trust in high-stakes data work is earned through a visible, checkable track record, not granted through a permissions toggle switched on once. Confidence indicators, an exception queue for uncertain matches, and visibility into the agent’s own accuracy over time are the pieces that make that possible.
Soundmouse is a cue sheet reporting platform for film and broadcast. The Partnerships feature I designed lets music users share cue sheet data with rights owners automatically, instead of over email. But sharing the data doesn’t solve the next problem: someone still has to match each reported cue to the correct composition and rights holder, by hand, across thousands of rows. This concept asks what happens when an AI agent takes the first pass at that matching - and what it takes to make that trustworthy rather than reckless.