AI shopping agents promise to remove friction between deciding and buying. For retailers, however, the most convenient interface may also become the most dangerous intermediary. If customers ask a third-party assistant what to purchase and complete the transaction without visiting a retailer, control over discovery, recommendations, loyalty, and data moves elsewhere.
Kroger, Lowe’s, and Papa Johns are testing a different model. They are using Google-backed agent technology inside their own apps, websites, and customer journeys while keeping the retailer’s identity and systems in front.
The experiments show that agentic commerce is not simply a race to automate checkout. It is a contest over who represents the customer.
First-Party Agents Preserve Context and Control
Kroger’s agent is designed to compare items, complete purchases, and adapt suggestions to meal plans, time constraints, price sensitivity, and brand preferences. Because it sits inside Kroger’s app, the retailer can combine the conversation with information it already holds.
That creates a richer service, but it also protects commercial advantages. Kroger can decide how alternatives are presented, preserve opportunities for relevant additional purchases, and maintain the direct relationship that supports loyalty and retail media.
Lowe’s takes a similar approach with Mylow. Google’s shopping agent operates behind the retailer’s own assistant, which Lowe’s says more than doubles conversion rates among online users. The customer experiences Lowe’s, even though external technology helps power the interaction.
This model is not full ownership. Retailers still depend on model providers and partners, but they control the visible journey, define business rules, and retain a clearer view of what customers asked before buying.
Multiple Vendors Reduce a Fast-Moving Dependency
Retail technology leaders cannot assume today’s best model or interface will remain competitive. Lowe’s chief digital and information officer, Seemantini Godbole, says technology can become outdated within two weeks. The company therefore works with several vendors, including OpenAI, rather than betting everything on one supplier.
Kroger also combines Google with partners such as Instacart. A multi-vendor approach can preserve negotiating power and let retailers match different models to search, support, recommendations, or fulfilment. It also makes replacement possible when a provider changes its terms or performance.
The trade-off is integration complexity. Customer context, permissions, product data, and evaluation must remain consistent while underlying services change. Retailers need an orchestration layer that separates their commercial logic from any single model.
They should also monitor whether recommendations reflect inventory, pricing, dietary needs, and customer intent accurately. An agent that confidently suggests unavailable or unsuitable products damages the retailer’s reputation, regardless of which provider generated the answer.
Not Every Retailer Needs to Build the Agent
Papa Johns offers the pragmatic counterpoint. It is testing Google’s food-ordering agent rather than developing models itself. One proposed use estimates how many pizzas a group may need from an uploaded photograph, with access through phone, website, or app.
Chief digital and technology officer Kevin Vasconi wants the company to become expert in using agents, not building them. That distinction is sensible for businesses whose advantage lies in products, service, and operations rather than model engineering.
The strategic choice is not build versus buy in absolute terms. Retailers must decide which layers differentiate them. Customer rules, catalogue quality, loyalty, fulfilment, and measurement may deserve direct control even when the conversational engine comes from a partner.
Detailed trial results remain limited, and agentic shopping is not yet the dominant way people buy everyday goods. Customers still call stores and use familiar channels. That gives retailers time to test without pretending the transition is complete.
The strongest approach combines openness with caution: experiment where customers already interact, keep data and business rules portable, and avoid surrendering the brand interface by default. AI agents may simplify shopping, but retailers should remember that the assistant making the decision can become more influential than the store fulfilling it.
