By admin 
Siobhán Mc Feeney, Senior Vice President of Technology at Target, articulated a compelling vision for the role of artificial intelligence within the retail giant during her presentation at VB Transform 2026. She firmly stated that while AI models are indeed powerful and essential components, they are not the sole or even primary drivers of Target’s competitive advantage. Instead, Mc Feeney emphasized that the true "moat" – the sustainable differentiation that protects Target from rivals – lies in the comprehensive infrastructure, disciplined processes, and strategic frameworks meticulously constructed around these AI models. This nuanced perspective challenges the prevailing narrative that simply adopting advanced AI algorithms guarantees market leadership, highlighting instead the critical importance of holistic implementation and operational excellence.
Mc Feeney’s approach to AI deployment is characterized by a profound sense of deliberation, a discipline that begins with the fundamental question of whether an AI agent is necessary at all. She candidly acknowledged what she termed the "controversial" reality of the current AI landscape: while every enterprise clamors for AI agents, not every problem warrants such a solution. This discerning stance underscores Target’s commitment to strategic investment, ensuring that AI resources are directed towards applications that genuinely enhance value for both the business and its customers. The principle of "earning autonomy" is central to this philosophy. AI agents at Target do not automatically receive expansive permissions; rather, they progressively gain more control and agency as they demonstrate reliability, accuracy, and consistent value delivery. This incremental approach to autonomy is a cornerstone of Target’s AI strategy, permeating every facet of their agent development and deployment.
The ultimate objective, as articulated by Mc Feeney, is to ensure that AI agents are precisely aligned with the challenges that yield the most significant positive impact for Target’s guests. "We want to make sure we’re investing in the right places," she reiterated, underscoring a commitment to data-driven decision-making and a keen focus on customer-centric innovation. This strategic alignment is paramount in an era where the retail landscape is constantly evolving, demanding agility and foresight.
A Deliberate and Structured Approach to Agent Development
Within Target’s operational framework, AI agents are increasingly becoming integral to the company’s underlying architecture. They are meticulously designed to forge robust connections between disparate signals, complex systems, and critical decision-making processes that span across vital areas such as supply chain management, inventory replenishment, and demand forecasting. This integration is not merely about adopting new technology; it represents a sophisticated evolution of retail’s foundational promise: delivering the right product, to the right place, at the right time, and at an unprecedented scale.
Mc Feeney detailed a highly structured and deliberate process for building AI agents. This systematic approach commences with the most elementary yet crucial question: "What is the problem we are trying to solve?" This initial inquiry naturally branches into a series of essential follow-up questions that guide the entire development lifecycle. "You define that upfront, and this may sound a little process-heavy, then you have to register and certify your agent," Mc Feeney explained. This rigorous registration and certification process serves a dual purpose: it prevents the duplication of effort by ensuring that a solution doesn’t already exist within the organization, and it establishes a baseline for accountability and oversight.
The agent design phase itself triggers another cascade of critical considerations. A fundamental question revolves around the triggers that will initiate an agent’s action. Will it be an automated process, a direct command from an engineer, or a scheduled timer? Equally important is the establishment of robust mechanisms for tracking the agent’s performance and actions. "We’re trying to make sure we have lineage from the very beginning – the birthing of this agent, all the way through – because at 2 a.m. one morning, when something goes sideways, we want to make sure we understand everything that happened," Mc Feeney emphasized. This commitment to end-to-end lineage is vital for effective troubleshooting, auditing, and continuous improvement, especially in a 24/7 retail environment where issues can arise unexpectedly.
The level of autonomy granted to an agent is another crucial design parameter. New agents typically begin with a foundational level of autonomy and gradually earn greater permissions based on their demonstrated performance and reliability. This principle of earned autonomy is consistently applied across all aspects of agent functionality. Furthermore, the scope of an agent’s access is meticulously defined. What data sources, systems, tables, or databases will the agent be permitted to interact with? This granular control is essential for maintaining data integrity, security, and compliance.
Finally, the critical functions of monitoring and observability are woven into the fabric of agent development. Mc Feeney stressed that agents will not inherently solve problems or improve over time if they are not subjected to continuous evaluation. "We measure everything: What it was intended to do, its calibration, its trajectory, not just runtime and latency," she stated. This comprehensive measurement framework provides full transparency into an agent’s behavior, enabling the necessary adjustments and optimizations to be made over time. The result is a dynamic system where agents can be continuously refined to meet evolving business needs and customer expectations.
Mc Feeney elaborated that this deliberate approach necessitates the establishment of a robust foundation encompassing "architecture and taxonomy and a data governance layer that absolutely had to be established." These foundational elements are the bedrock upon which Target builds its "layers of autonomy," providing the scalability and strategic investment capabilities required to deploy the most appropriate models for specific challenges. The recognition that different AI models possess varying "gradients" – making them better suited for different tasks – is also a key consideration. For instance, while frontier models might excel at complex, data-intensive operations like those found in heavy merchandising supply chains, their computational cost can be prohibitive in certain scenarios. Therefore, Mc Feeney highlighted the constant imperative to "make sure there’s always a cost benefit." This pragmatic approach ensures that AI investments are not only technologically advanced but also economically sound.
Autonomy as an Earned Privilege, Not an Automatic Right
The concept of "earned autonomy" was vividly illustrated by a real-world example involving a digital-twin simulation designed to predict men’s shorts inventory across three Target stores in Long Beach. The simulation revealed a striking anomaly: one store required six to seven times more stock than the other two. The initial reaction from inventory analysts was skepticism – "That can’t be right." However, the AI system had identified a critical factor that had been overlooked: proximity to the beach. The outlier store was located less than two miles from the coastline, while the other two were situated 10 to 12 miles inland. Trusting the AI’s data-driven insight, analysts allowed the recommendation to stand, and the increased stock sold through as predicted. Mc Feeney presented this outcome as a testament to the power of AI: "This is science. This is mathematically more significant and more confidence-filling than humans doing it." It is precisely these kinds of demonstrably successful outcomes that allow Target’s agentic systems to progressively earn greater autonomy.
Target conceptualizes AI agent autonomy as a four-level ladder, where agents begin by merely observing and making recommendations, progressing through stages of suggesting actions with human approval, to operating within defined guardrails, and ultimately, at the highest current level, running end-to-end processes with a human still remaining "in the loop" for critical oversight. "The autonomy levels for the agents are super important," Mc Feeney emphasized. "They earn them, and they can lose them if they don’t perform as expected." This dynamic reassessment means that models exhibiting drift or underperformance are promptly taken out of service, ensuring that the AI ecosystem remains robust and reliable.
Mc Feeney drew a compelling parallel between human professional development and AI agent autonomy. Just as humans earn greater responsibilities through demonstrated competence and a track record of success, AI agents must prove their capabilities through quantifiable metrics. This scientific measurement and quantification – including accuracy, degree of drift, and proximity to intended goals – establish clear guardrails. These guardrails, in turn, empower builders to operate with greater speed and confidence, as they are not constantly second-guessing the boundaries of acceptable operation. "If you follow these guardrails, you [follow] security guidelines, you register the agent, and something still goes wrong, we have full lineage all the way through from the start," Mc Feeney assured. "Our ability to recover is much better." This comprehensive approach to risk management and error mitigation is a direct consequence of the disciplined, layered approach to AI implementation.
Ultimately, Mc Feeney concluded, agent success is not the result of a single factor but a confluence of critical elements: "It’s about your architecture. It’s about your taxonomy. It’s about the autonomy levels your agents have, and it’s about security and observability." This holistic view underscores that technology alone is insufficient; it must be embedded within a well-defined strategic and operational framework.
Cultivating a New Skill Set for Evolving Workflows
Even as AI agents achieve higher levels of autonomy, the accountability of human builders remains paramount. Mc Feeney acknowledged that the unprecedented pace at which teams are now operating necessitates the establishment of robust evaluation harnesses and meticulous tracking mechanisms for registered agents. This acceleration in work velocity demands a parallel evolution in organizational culture and workforce capabilities.
A significant aspect of this evolution is cultural. The traditional workforce is being reshaped, requiring builders and engineers to acquire new skills that enable them to effectively manage and collaborate with both human workers and AI systems concurrently. These are distinct but complementary contexts, and Mc Feeney described this career evolution as "super exciting." The role of the builder is transforming from a sole executor of tasks to a more nuanced facilitator and coach. "You’re a builder. You’re observing agents building, and you’re also coaching humans observing agents building," she explained. "The level of nuance is pretty special." This signifies a move towards a more symbiotic relationship between humans and AI, where human expertise is amplified by intelligent systems, and AI performance is guided by human insight and strategic direction. This complex interplay of human and artificial intelligence represents the future of work, and Target is actively positioning itself at the forefront of this transformative shift.
