How Walmart Uses AI in 2026: A Real-World Case Study

In 2025, Walmart employs more than 2.1 million people and serves roughly 255 million customers every week across its stores and digital platforms. At that scale, even a small technology decision ripples across millions of daily interactions. Since 2023, the company has pursued a methodical rollout of generative AI, documented publicly by its own technical leadership. This case study relies exclusively on primary sources: Walmart's official announcements, Walmart Global Tech engineering posts, and financial data shared during quarterly earnings calls.

An architecture built around four agents

Rather than deploying a single, general-purpose AI assistant, Walmart chose to build what Global CTO Suresh Kumar calls “super agents,” each dedicated to a specific audience. This approach, unveiled in July 2025 on the company's engineering blog, outlines four agents: Sparky for shoppers, an agent for in-store associates, Marty for suppliers, third-party sellers and advertisers, and an agent for internal developers. Each agent runs on a shared architecture while staying specialized in its own domain, which limits errors and makes it easier to measure performance area by area.

This segmentation isn't just an internal organizational choice: it reflects a philosophy voiced by Hari Vasudev, Walmart's Chief Technology Officer, who describes a strategy of “surgical” agents, built for precise tasks rather than trying to do everything at once. A language model tailored to retail is combined with other models depending on the need, instead of betting everything on a single provider.

Two assistants, two very different audiences

Sparky, the customer-facing conversational agent, helps with product search, recommendations and order tracking directly inside the Walmart app. Wally, on the other hand, is built for internal merchandising teams: launched in March 2025, this assistant taps into a semantic layer connected to internal data to automate data entry, analysis, root-cause investigation and calculations once handled manually by merchandising staff.

This split between a customer-facing agent and a business-facing one illustrates a simple but often overlooked lesson: the same technology foundation can serve very different use cases, as long as the interface and functional scope are adapted to each audience.

Investing in tools for developers

Walmart Global Tech has documented, across several engineering posts, a suite of tools built for its own developers. The DX AI Assistant reportedly saves around five minutes per technical question asked by a developer. ArchiText, a tool that generates UML diagrams from natural language, claims a 69% faster output, or roughly 31 minutes saved per diagram. Other tools such as Code Buddy and the Pipeline Visualizer round out the toolkit, all built on a proprietary machine learning platform called Element.

Walmart has also formalized a “responsible AI” commitment built on six internal principles governing how its technical teams use these tools. This kind of explicit governance is rarely made public by large enterprises, which makes it a useful reference point for any organization rolling out AI internally.

Measured results before scaling further

Before extending its tools, Walmart measured their impact. The internal “Trend-to-Product” process, which turns a spotted trend into a product on the shelf, reportedly saw its timeline cut by 18 weeks thanks to generative AI, according to Hari Vasudev. On the customer service side, Walmart reports a 40% reduction in request resolution time thanks to its dedicated conversational assistant, even before the ChatGPT integration announced in October 2025.

These figures, shared directly by the company, have since been complemented by independent financial data. When reporting fourth-quarter results for fiscal year 2026, John Furner (President and CEO of Walmart US) and David Guggina (President of Walmart US) noted that roughly 50% of app users had already tried Sparky, that weekly active users of the assistant had more than doubled in a single quarter, and that Sparky users' average basket size was about 35% higher than that of other customers, according to reporting by trade outlet Retail Dive.

Building and buying at the same time: the OpenAI partnership

In October 2025, Walmart announced a partnership with OpenAI allowing customers to complete a purchase directly from ChatGPT through an instant checkout feature. Walmart said it wants to let customers shop wherever they already are, including inside conversational interfaces like ChatGPT. The partnership also includes an internal rollout of ChatGPT Enterprise and OpenAI certifications offered to Walmart associates.

It illustrates a deliberate hybrid approach: building specialized in-house agents (Sparky, Wally, Marty) while also partnering with outside providers for specific use cases, rather than treating the two strategies as mutually exclusive.

An outside view: what analysts are saying

Not every source in this case study comes from Walmart itself. Neil Saunders, an analyst at GlobalData Retail, has offered a more tempered take, noting that a share of consumers remain skeptical about shopping through conversational assistants, and that trust in these tools builds gradually, particularly through narrow, well-defined use cases rather than broad promises. This outside perspective, echoed by trade press, is a reminder that real adoption is best measured by actual usage over time, not internal announcements alone.

What smaller businesses can take away

A smaller company obviously doesn't have Walmart's resources, but several principles still apply. Segmenting use cases by audience, rather than searching for one universal tool, makes it easier to measure the impact of each rollout. Testing within a narrow scope before expanding, the way Walmart did with customer service before the OpenAI partnership, reduces the risk of a rushed deployment. Combining in-house tools with third-party solutions isn't a contradiction, but a deliberate strategy used by one of the world's largest retailers. Finally, user trust is earned through precise, measured use cases, not broad claims about artificial intelligence.

For companies trying to identify the right AI tools for each of these use cases, a category-by-category comparison is a useful starting point before making a decision.