Nana Muazin
As businesses, particularly those in the e-commerce sector, transition from the experimental phase of Artificial Intelligence to its widespread deployment in production and customer-facing applications, a critical question emerges: how should merchants manage AI agents that operate with delegated business authority? David Greenberg, Chief Marketing Officer at BlueRock Security, a cybersecurity firm, stresses that before granting AI agents the ability to autonomously update product catalogs, issue refunds, or modify marketing campaigns, companies must establish clear lines of responsibility for any errors or extreme risks incurred. "Once an agent can update a product catalog, issue a refund, or change a campaign, it becomes operational software with delegated business authority. The company needs to treat it that way," Greenberg articulated in a recent interview.
The evolving landscape of e-commerce sees AI agents moving beyond theoretical applications to becoming active participants in business operations. This shift necessitates a robust framework for governance and oversight. Consider a scenario where an e-commerce merchant instructs an AI agent to enhance conversion rates for underperforming products. The agent, in its pursuit of this objective, might analyze inventory levels, identify slow-moving items, implement price reductions, and redirect advertising expenditure towards these products. While each individual action might appear logical and beneficial in isolation, the cumulative effect of these automated decisions could inadvertently erode profit margins, deplete stock prematurely, or generate unforeseen costs. This underscores the fundamental question of how much autonomy merchants should entrust to these sophisticated algorithms.
Defining Delegated Authority: The Ambiguity and Risk Matrix
The Massachusetts Institute of Technology’s Center for Information Systems Research (CISR) has identified two pivotal factors—ambiguity and risk—as crucial determinants for deciding which decisions are suitable for AI execution. Routine tasks characterized by predictable outcomes and limited potential consequences are prime candidates for automation. Conversely, decisions involving a degree of uncertainty, substantial financial exposure, or conflicting objectives warrant a greater degree of human involvement and oversight.
Greenberg further elaborated on these considerations in his correspondence with Practical Ecommerce, introducing additional critical factors for evaluating the delegation of authority to AI agents. These include the monetary value of the action, its reversibility, the potential impact on customers, the sensitivity of the data involved, and the cumulative exposure to risk over time. He cautioned, "A small refund repeated thousands of times can become a significant problem." This highlights the need for merchants to implement stringent guardrails that clearly define the scope of an agent’s access, modification capabilities, and spending limits. These restrictions are not merely advisory; they must be technically enforceable as the agent operates, rather than relying solely on initial instructions. Greenberg’s advice is stark: "If your visibility ends at the prompt, keep authority narrow." This emphasizes the importance of ensuring that human oversight and control are maintained at every stage of an AI agent’s operation, especially when its actions have tangible business consequences.
Establishing Accountability: Who Owns the Outcome?
A fundamental principle in the delegation of tasks, whether to human employees or AI agents, is that responsibility for the consequences remains with the delegator. Greenberg asserts, "A department accountable for the process should remain accountable when an agent performs it." This means that the marketing department remains responsible for campaign outcomes, customer service for resolving customer issues, and procurement for purchasing decisions, even when AI agents are executing these functions. This principle is universally applicable, regardless of the specific AI agent or vendor involved.
"A shopper experiences the company’s promise, price, or refund. An internal explanation about which AI vendor generated it will not repair a poor experience," Greenberg noted. This underscores the direct customer-facing impact of AI-driven decisions. To address this, he recommends that every operating AI agent should have a designated "owner"—an individual responsible for its configuration, access management, ongoing monitoring, and incident response. In smaller organizations, these responsibilities might be consolidated within a single individual. Furthermore, merchants must possess the capability to reconstruct the sequence of events following an AI agent’s action. For instance, in the case of a refund, this involves identifying the specific agent, the customer involved, the item or service in question, the refund amount, the timestamp of the transaction, and any associated notes or justifications. Such detailed records are crucial for differentiating between a flawed decision, an execution error, or a customer-initiated issue, ultimately enabling performance improvements.
The Nuance of Commercial Judgment: Beyond Optimized Metrics
Perhaps the most significant risk associated with AI agents is when they perform precisely as instructed but yield undesirable business outcomes. Greenberg illustrated this with a hypothetical example: a merchant authorizes an advertising agent to maximize revenue within a fixed weekly budget. The agent identifies a product with a relatively high conversion rate and strategically reallocates spending towards it. This leads to an increase in sales, with the agent adhering to the budget and advertising only approved products. On the surface, this AI automation appears successful.
However, Greenberg pointed out a critical flaw in this scenario: "But the product has thin margins, expensive fulfillment, and a high return rate. The agent has optimized the metric it was given while reducing the business’s profitability." This highlights a crucial distinction between optimizing a specific, narrowly defined metric and achieving broader business objectives. "This is where marketing judgment becomes more valuable," Greenberg emphasized. "Teams need to translate business strategy into measurable objectives and enforceable constraints." This implies that human expertise is indispensable in defining the parameters within which AI operates, ensuring that the AI’s pursuit of efficiency aligns with the company’s overarching profitability and strategic goals. The ability to discern the potential downstream consequences of an AI’s actions, even when those actions are technically correct according to its programming, is a uniquely human capability.
The Complexity of Connected Agents: Orchestrating Interdependent Systems
The operational complexity escalates significantly when multiple AI agents function concurrently within the same business environment. Consider a scenario where a merchandising agent is tasked with reducing prices, an advertising agent is instructed to increase spending, and a purchasing agent is programmed to replenish inventory. While each agent might operate reasonably within its own defined parameters, their combined actions could inadvertently create a detrimental feedback loop. For example, a price reduction by the merchandising agent might trigger increased advertising by the advertising agent, leading to a surge in demand that the purchasing agent attempts to fulfill, potentially exacerbating issues like stockouts or increased fulfillment costs if not carefully managed.
A September 2026 Harvard Business Review article explored how companies, including retail giant Walmart, are actively developing sophisticated systems to coordinate AI agents across different functional silos. The research indicated that while specialized agents can generate valuable independent recommendations, their uncoordinated actions can lead to conflicting outcomes across departments. The potential to automate entire business processes by effectively orchestrating these agents is substantial, but it necessitates the establishment of shared rules, robust management oversight, and effective mechanisms for reconciling competing objectives. This requires a holistic view of how different AI agents interact and influence each other, moving beyond individual task optimization to system-wide efficiency and strategic alignment. The development of such coordination frameworks is becoming increasingly vital as businesses integrate more autonomous AI systems into their operations.
Operating AI Agents: From Building to Governing
The increasing accessibility of AI empowers owners, managers, and contributors to build custom tools and automate processes that historically required specialized developers, complex software, or extensive manual labor. However, the act of building an AI agent is distinct from the ongoing process of operating one effectively and responsibly. The ultimate objective is not to mandate human approval for every routine action, thereby negating the benefits of automation. Instead, the goal is to implement sufficient control mechanisms to delegate valuable work without relinquishing ultimate responsibility.
For e-commerce companies, the pertinent questions are not solely about what AI agents can do, but rather, what they are allowed to do, under what specific conditions, and who ultimately bears accountability for the resulting outcomes. This strategic framing is essential for harnessing the power of AI while mitigating its inherent risks, ensuring that these advanced tools serve as reliable extensions of business strategy rather than unpredictable agents of chaos. The journey from experimentation to operationalization of AI agents demands a proactive and comprehensive approach to governance, risk management, and accountability, ensuring that the pursuit of technological advancement is intrinsically linked to sound business principles and ethical considerations. The future of e-commerce will undoubtedly be shaped by the intelligent integration of AI, but its success will hinge on the ability of businesses to master the art of delegating authority while maintaining unwavering control and accountability.
Muslim
Azzam Bilal Chamdy
