Inside Suno, the AI Music App You Won't Be Able to Stop Listening To

Shulman and his co-founders moved from financial-services AI into audio after working on speech transcription and recognizing how much room there was to explore sound. Though they first considered speech, their shared experience as musicians drew them toward music. He says music depends on taste and subjective judgment, so scaling a model alone does not solve the product challenge.

Suno uses Transformers to predict small pieces of sound, represented by audio tokens that compress the continuous signal. Shulman describes two common user groups: power users who spend hours refining songs, and casual users who make and share songs about family or everyday events. He says 25 million people had made songs on Suno, most for the first time, and that sharing supports organic growth.

Shulman says Suno v4 improved audio fidelity, songs, and lyrics without changing the product interface. He highlights Covers as a feature shaped around the appeal of keeping a song familiar while changing its style, which required work on both the model and the interface. The team gathers feedback from its 400,000-person Discord community and tracks signals such as likes, listening past 30 seconds, and shares.

Shulman frames AI music as part of a longer history of technologies that made music easier to record, copy, and access. He warns that AI could enable uncredited imitation of artists or solitary hyper-personalized listening, and argues for a future where more people participate in making and sharing music. He says Suno aims to make creation more social and to work with more professional musicians.

Shulman says teams should build features around what users will love, then work out how to make the model deliver them. Suno charges users and uses payment behavior to learn what people value beyond novelty. His advice to product builders is to focus on user problems and comparative strengths rather than treating model internals as the product.