Prompt as Set: Compositional Literacy, Cultural Memory, and the Long-Form AI Music Performance

Prompt as Set: Compositional Literacy, Cultural Memory, and the Long-Form AI Music Performance This paper documents the design, compositional logic, and prompt engineering methodology underlying a feature-length AI-generated musical performance intended for live presentation at Burning Man 2026, Black Rock City, Nevada. The work — provisionally titled صابرين / Sabirin (Be Patient) — is a sixty-to-ninety-minute continuous set developed by Craig Ellenwood (Squaawke) in collaboration with Claude (Anthropic), generated using ACE-Step, an open-source music generation architecture. The set traverses a structural and cultural arc: from Western Arabic Bedouin oral tradition addressing fundamental human conditions — hunger, displacement, the search for shelter — through the genre of playa techno as defined by the authors' direct participation in the founding of the Burning Man music program in 1992, to comic and satirical interludes functioning as oral history, and finally to an Italian-language operatic closing and an acoustic benediction. We present this work as a case study in what we term long-form prompt architecture: the use of language prompts not merely to elicit discrete musical outputs, but to engineer a sustained aesthetic and emotional arc across an extended composition. We identify structural parallels between the AI-directed music session and traditional oral performance traditions — specifically the maqam-based suite, the storytelling djemaa, and the Bedouin bardic convention — in which a single performer holds a community's attention through modulation of theme, tempo, humor, grief, and catharsis. We argue that the prompt functions in this context as set list, score, libretto, and cultural frame simultaneously, and that mastery of this multi-layered prompt function constitutes an emerging compositional discipline with direct lineage to pre-literate musical traditions.