Early experiments and enterprise scenarios
We’re already seeing glimpses of this future in experimental work. In one project, Reuven Cohen of the Agentics Foundation demonstrated how outcome-driven prompting could orchestrate swarms of agents to handle research, design, coding, and testing, all without a fixed workflow defined in advance. The system self-organized, spinning up and shutting down agents as needed.
It wasn’t perfect. Workflows and tools were needed to handle deployment decisions and data access, and often took several attempts to get working correctly, if at all. But it illustrated what becomes possible when agents can self-organize around outcomes rather than follow rigid workflows.
For enterprises, the stakes are higher. Imagine customer service agents spinning up around the world to handle localized requests. Each must access the right customer data in compliance with regional regulations and retire cleanly once the interaction ends. Without an abstraction between capacity and consumption, every new agent risks becoming its own operational headache. With it, ephemeral swarms can become manageable.
