Target SVP Siobhán Mc Feeney told VB Transform 2026 that the company's competitive advantage in AI is not the models it runs. It is the architecture, taxonomy, data governance layer, autonomy framework, and observability stack built around those models. Her term for it: the moat. The models are necessary but not sufficient.
The operational detail here is worth reading in full. Target runs a four-level autonomy ladder for agents: observe only, suggest and wait, act within guardrails, run end-to-end with a human in the loop. Agents start at level one and earn promotion through measurable performance, calibration scores, and drift monitoring. They can also be demoted. A digital-twin simulation flagged that one Long Beach store needed six to seven times more men's shorts inventory than two nearby locations. Analysts pushed back. The system had detected the store was less than two miles from the beach. The stock sold through. That result is what earns an agent more autonomy. Before any agent is built, Mc Feeney's team forces three questions: does this problem actually need an agent, what type of agent, and is what you're calling an agent just a tool.
The piece also covers how Target selects model types by cost-benefit gradient, using frontier models for complex merchandising and supply chain tasks but not defaulting to them everywhere. The workforce implications are direct: engineers now coach humans who observe agents building, a skill set that did not exist two years ago. Mc Feeney calls the career evolution exciting. Whether you believe that or not, the governance scaffolding she describes is the most detailed public account of enterprise agent deployment at retail scale.
[READ ORIGINAL →]