Software as Content: Dynamic Applications as the Human-Agent Interaction Layer

Researchers Mulong Xie and Yang Xie have published a paper introducing 'Software as Content' (SaC), a radical new paradigm for human-agent interaction. The core argument is that chat-based interfaces, while dominant, are fundamentally limiting for complex tasks. They identify three critical flaws: the mismatch between structured data and linear text, the high entropy (or unpredictability) of unconstrained natural language input, and the lack of a persistent, evolving interaction state. SaC proposes that AI agents should generate dynamic, interactive applications as their primary output, not text.

These agentic applications are task-specific interfaces that present structured information and expose actionable controls—buttons, sliders, filters—allowing users to guide the AI iteratively without relying solely on typing. Crucially, these interfaces persist and evolve across interaction cycles, forming a shared, stateful layer that progressively converges into personalized software for the task at hand. The researchers formalize this with a human-agent-environment model, derive design principles, and present a system architecture. Their evaluation across tasks of selection, exploration, and execution demonstrates technical viability while acknowledging that natural language chat remains preferable for certain boundary conditions. This reframes the interface itself as a dynamically generated software artifact, opening a new design space for human-AI collaboration.