At the VB Transform 2026 conference on July 29, Siobhán Mc Feeney, Senior Vice President (SVP) of American retail giant Target, shared a pragmatic perspective on the current AI wave. According to her, AI models are not Target's core competitive advantage; rather, it is the entire architecture, governance, and operational workflows built around them. This assertion reflects the grounded view of a retail giant amid the global AI boom.
Background & Drivers
The generative AI rush has pushed many enterprises to hastily deploy autonomous agents without thorough systemic preparation. At Target, this process is approached with far greater caution and practicality. Before deciding to build any agent, the development team must answer a series of core questions to determine whether the problem truly requires an AI agent, and if so, of what type. A strict 'agent registration and certification' process has been established to avoid duplicating resources and efforts. Target also focuses on building a lineage system from the moment an agent is created until its live operation. This ensures that when an incident occurs in the middle of the night, engineers can quickly understand the entire chain of behavior that took place.
Technical Analysis & Technology
From a technical perspective, Target does not rely on a single model but applies them flexibly depending on cost-optimization challenges. The most advanced frontier models are reserved only for highly complex tasks, such as large-scale supply chains, which require processing billions of data points. To manage this effectively, Target constructed a four-level 'autonomy ladder' for its agents:
* Level 1: Passive Observation - The AI is only allowed to observe without taking action. * Level 2: Suggested Action - The AI proposes actions that await human approval. * Level 3: Guardrailed Autonomy - The AI operates autonomously within predefined safety guardrails. * Level 4: Full Autonomy with Human-in-the-Loop - Agents execute end-to-end processes independently but remain under human supervision.
Notably, agents must 'prove their capability' to be promoted up the autonomy ladder and will be stripped of their privileges or downgraded immediately if performance degrades or drift occurs.
Expert Insights & Perspectives
Sharing real-world results, Mc Feeney pointed to a digital-twin simulation that predicted inventory levels for men's shorts at stores in Long Beach. The AI system suggested increasing stock by six to seven times at a specific store, making analysts initially skeptical of the data. However, deeper analysis revealed that the store was less than two miles from the beach, whereas other stores were located further inland. Target accepted the AI's recommendation, and the entire inventory subsequently sold out. According to Mc Feeney: 'This is science. This is rigorous math, and it brings much greater confidence than previous manual forecasting methods.'
Impact & Future Outlook
Target's AI success demonstrates that long-term technological advantage does not lie in owning the most expensive AI model, but in system integration and data governance capabilities. This trend demands a new workforce with specialized skills, where engineers do not just write code but act as orchestrators, training both humans and machines to work in tandem. For Vietnamese businesses undergoing digital transformation, Target's journey highlights the critical importance of building robust infrastructure and safety guardrails before chasing transient AI trends.