As businesses evolve from experimenting with AI to deploying agents in production or customer-facing applications, the legal and operational structure around decision-making must evolve to match the software’s capabilities.
Defining the Scope
David Greenberg, chief marketing officer at BlueRock Security, argues that once an agent can update a product catalog, issue refunds, or change a campaign, it effectively acts as operational software with delegated authority. This shift means businesses should manage these systems with the same rigor they apply to employees.
Consider a merchant instructing an AI to boost sales for underperforming products. The agent might analyze inventory, identify slow-moving items, reduce prices, and shift advertising spending toward those items. While each individual step seems logical, the combined effect can erode profit margins or exhaust inventory levels. Greenberg offered a hypothetical example of an advertising agent maximizing revenue within a fixed budget. The agent targets a product with a high conversion rate but thin margins, expensive fulfillment, and a high return rate.
The Massachusetts Institute of Technology’s Center for Information Systems Research identified ambiguity and risk as key differentiators in determining which decisions AI should make. Greenberg suggested additional considerations like monetary value, reversibility, customer impact, and cumulative exposure. Routine tasks with predictable outcomes and limited consequences are good candidates for automation.
A small refund repeated thousands of times can become a significant problem. Greenberg advised that if a business lacks visibility beyond the initial prompt, it should keep authority narrow. Merchants should establish guardrails that define what an agent can access, change, or spend, rather than relying on instructions alone.
Managing Multiple Units
Greenberg notes that when multiple AI agents operate simultaneously, a pricing agent cutting costs, an advertising agent increasing spending, and a purchasing agent restocking inventory, they can function reasonably within their own assignments while creating an unwanted feedback loop when their actions combine.
Harvard Business Review research highlights that companies like Walmart are already developing frameworks to coordinate these specialized units, requiring shared rules to resolve conflicting outcomes.
Accountability and Oversight
Responsibility for outcomes does not transfer simply because an agent performs the work. The department accountable for the process remains liable, regardless of whether a human or a machine executed the task. Greenberg emphasizes that an internal explanation about which vendor generated a poor customer experience is insufficient.
