Agential is a series of generative real-time works in which thousands of gaussian splats inhabit an environment produced by a custom machine learning algorithm.
The works present as large-scale abstract compositions in which thousands of discrete marks accumulate into dense, pressurized fields of color. Individual units, elongated, seed shaped, occasionally filamental, are legible at the edges where the population disperses, but dissolve into continuous chromatic mass at the center. The palette draws on earthy tones producing surfaces that carry the visual logic of paint without its permanence.
Each splat is a minimal information unit, carrying position, opacity, covariance, and color as intrinsic properties. It expresses these into the space it occupies, unaware of the units around it. The machine learned environment provides the morphogenetic field, the conditions within which each splat expresses its properties. The form that emerges is a consequence of that navigation.
Michael Levin's research into cellular cognition demonstrates that goal-directed behavior exists at the level of individual cells, prior to nervous systems and without central coordination. The body in this account is a community of semi-autonomous agents whose pooled behavior produces structure no individual agent planned. Agential runs the same organizational logic computationally.
Agency here is a property of matter under the right conditions, prior to coordination, prior to intelligence. The drive of an agent to express what it is into space, and the world that builds itself from that.
Real-time generative and machine learning based computational work, single channel. Variable dimensions. PC, custom software.
PROCESS
These images show some of the computer vision algorithms behind Agential, revealing the perceptual pre-processes that shape how agents respond to their environment. By making visible how machines see, they open a new way of interacting with the agents that occupy the canvas. This moves away from prompt-based interaction toward building a shared visual language that both humans and machines can understand, yielding new ways of manipulating behavior.
