From Prompts to Portfolios: AI Agents as Agentic Multimedia Firms

Agentic AI is moving from producing one-shot outputs to operating persistent multimedia portfolios. While multimodal foundation models and agentic systems are rapidly advancing in their ability to generate content and execute tasks across environments, the dominant abstraction in multimedia research still treats these systems as mere tools or workflows. We argue that this abstraction fails once a system produces assets rather than transient answers. Assets persist, branch into derivative families, accumulate provenance and rights obligations, and interact reflexively with platform ecosystems. To address this, we introduce the media-firm hypothesis and the agentic multimedia firm (AMF): a bounded computational organization that pursues persistent objectives over an asset portfolio through compiled contracts, make-or-buy decisions, provenance-aware ledgers, and liability-bearing governance. By treating the governed asset portfolio as the primary unit of analysis, we establish the organizational primitives that separate a firm from a workflow and derive boundary rules for internalization, oversight, and market structure. Ultimately, we demonstrate that frontier multimedia systems face a critical capability–commitment gap: model generation quality is improving faster than the institutional mechanisms needed to govern assets at scale. For the next generation of multimedia AI, organizational design will therefore be as consequential as algorithmic capability.