Seizing the means of machine listening/learning and agentic AI

creative practices with conTorchionist and vibe composing

Abstract

This article addresses the challenges posed by the hegemony of commercial, “black-boxed” Artificial Intelligence (AI) tools in musical creation, aiming to counteract the alienation inherent in Software as a Service (SaaS) platforms. Drawing on the concepts of mechanology and technodiversity, we advocate for a critical and inventive engagement with technology through the appropriation of machine listening and learning infrastructures. To this end, we present conTorchionist, an open-source, nomadic, and cross-platform library built upon libtorch, designed to bridge the gap between non-real-time research (Python) and real-time performance environments (Pure Data, Max, SuperCollider). Furthermore, we introduce the concept of ‘vibe composing’ as a methodology within research-development-creation and practice-based research in music, where the coding of algorithms assisted by agentic AI constitutes a fundamental poetic gesture. This approach is exemplified through two musical works, second nature (2025) and Eutrópia III (2025), which demonstrate how the development of idiosyncratic technical objects can engender unique aesthetic outcomes, ranging from spectral spatialization to neural control of latent spaces. Finally, we reflect on how seizing these technical means allows artists, particularly in the context of the Global South, to assert epistemic and technical sovereignty, fostering an ecology of invention that resists the standardization of the algorithmic age.