An agent is useful when it has a narrow job, clear tools, and a human checkpoint. It is a liability when it is asked to “run the business.”
The word “agent” is doing too much work in vendor conversations. In practice, an agent is a system that can take a defined action: look something up, fill a structure, route a ticket, assemble a first draft, or flag an anomaly for review.
The design questions are ordinary. What data is it allowed to see? What can it change? Who reviews the output? How is the action logged? Those questions matter as much in a commercial operations team as they do in a research environment.
Ruisra treats agents as workflow components, not as a new department. They sit next to existing systems — documents, data stores, ticketing, clinical review queues — and they fail closed when the task is outside their scope.