Walmart AI: How It Works Across Shopping and Operations

Walmart is not using one all-purpose artificial intelligence system. Its public record shows a collection of narrower tools tied to specific retail jobs: helping customers find products, handling parts of customer service, analyzing merchandising data, guiding associate tasks, positioning inventory, choosing fulfillment nodes, and coordinating delivery. The clearest customer-facing example is Sparky across Walmart's app, web, and in-store experiences. Behind it sit less visible systems such as the Wally merchant assistant and Walmart's fulfillment and supply-chain models.

Retail case studies often collapse three questions into one: does the tool exist, who can actually use it, and did it produce the advertised result? This article keeps them separate. A Walmart statement can establish what Walmart says is deployed or planned, but it cannot serve as independent proof of accuracy, savings, coverage, or job impact. No customer account, associate interview, internal system access, or reproduced performance result is part of the evidence here.

The status map matters more than the AI label

Walmart uses words such as AI, generative AI, predictive models, agents, automation, and digital twins across its disclosures. Those terms describe different kinds of software and different levels of autonomy. For this article, in use means Walmart currently describes a capability as live or operating. Pilot means the company limits it to selected shifts, stores, or trials. Announced means a future experience has been presented without a verified general-availability date.

The table is intentionally conservative. A tool can be real without being available to every customer or associate. A named pilot can be promising without becoming a permanent deployment. Walmart also updates capabilities without always publishing a feature-by-feature change log.

Sparky is live, while its larger promise is still forming

Walmart introduced Sparky in June 2025 through an Ask Sparky control in its app. At launch, Walmart said customers could use it across product categories to search, compare items, synthesize reviews, and prepare for an occasion. Walmart's current technology overview now describes Sparky as integrated into the app, where it can help shoppers find and compare products, build lists, receive recommendations, and plan occasions from product data and reviews.

Walmart has since confirmed part of that roadmap. In its May 2026 earnings-call transcript, the company said Sparky was live across its app, web, and in-store experiences, could speak Spanish, and could reorder items a customer buys repeatedly. Service booking and broad support for image, audio, and video input were not confirmed in the same update. The current overview calls voice and camera capabilities emerging, so those broader functions should not be labeled universally available.

Sparky is best understood as a retail interface over Walmart information, not as an independent product laboratory. It can compress listings and reviews into an answer, but Walmart does not publish a complete evaluation set for recommendation quality or a guarantee that every summary captures every material detail. Before buying, a shopper still needs to check the final item page, seller, size, price, stock state, delivery promise, and return conditions.

Customer care is moving from answers toward actions

Walmart's customer-service AI predates the current agent branding. In its October 2024 platform announcement, the company described a Customer Support Assistant that could recognize a signed-in customer, find an order, and manage returns. The current technology page says agents now route inquiries, resolve common issues, and increasingly handle tasks from start to finish while associates take more complex cases.

The phrase "increasingly" is important. It signals a changing boundary rather than a published catalog of fully autonomous service actions. Walmart does not disclose which return exceptions always require a person, what confidence threshold triggers escalation, or how often an automated action is reversed. A useful description is therefore simple: the assistant can act on some routine service needs, but the public record does not justify calling customer care fully automated.

There is also a data boundary. Walmart's U.S. customer privacy notice, updated in June 2026, includes interactions with AI assistant chatbots within communications data. It also lists categories such as purchase history, device activity, location, sensory information, and inferred preferences, depending on the service and settings. That notice is broader than Sparky alone, but it is the relevant starting point before sharing information in a retail chat.

Wally gives merchants a retail-specific analysis layer

Walmart merchants decide what products to source and how to manage assortments. Their questions combine sales by store, channel, item, market, and brand, which makes a generic chatbot a poor fit. Walmart says Wally is built on its proprietary merchandising data and a semantic layer designed to understand that internal context.

The documented functions are concrete: data entry and analysis, possible root-cause identification for product performance, process guidance, ticket creation for unresolved issues, and advanced calculations. That makes Wally an internal decision-support surface. It is not the same system as Sparky, and it does not answer a customer's shopping question.

Walmart's Wally announcement also describes a future in which the tool could execute tactical actions within configurable guardrails. The future tense sets the limit. The page does not establish that Wally already makes autonomous buying or inventory decisions. Nor does it publish independent accuracy tests for its diagnoses. A merchant can receive a suggested cause without that suggestion becoming a verified explanation.

A second merchandising tool, Trend-to-Product, supports Walmart fashion designers and merchants. Walmart says it analyzes trend inputs and can generate mood boards, collection concepts, and a technical pack. The same page says designers and merchants refine the material and use their own experience before final products are made. That human step is part of the disclosed workflow, not decorative language.

Associate AI is a patchwork of access and pilots

There is no honest one-line claim that every Walmart associate uses the same AI. Corporate-office work, store operations, merchandising, benefits, and software development have different tools and access rules.

For campus associates, Walmart documented My Assistant deployments for drafting, summarizing documents, and generating thought starters. The named rollout page is from 2024. Newer Walmart materials discuss a broader associate-agent framework, but they do not provide a current country-by-country My Assistant inventory. The safe conclusion is that Walmart deployed the named tool and continued building associate agents, not that every past rollout detail remains unchanged.

Frontline store tools have a clearer mix of states. In June 2025, Walmart said an AI-directed workflow for overnight stocking was already available, while its use for other shifts remained a pilot in selected locations. The same announcement described text and speech translation in the associate app. It also said a generative AI upgrade to an existing conversational help tool would launch in the following months. Without a later named release record, that last feature stays in the announced column here.

This distinction prevents two common exaggerations. A pilot on daytime shifts is not a chain-wide deployment. A plan to convert long process guides into step-by-step instructions is not proof that the upgrade reached all associates on schedule. Walmart's current technology page confirms that agentic functions exist inside associate tools, but it does not erase the narrower status attached to each named feature.

Inventory and fulfillment carry much of the operational load

The least visible Walmart AI may be the most operationally important. In July 2025, Walmart described predictive warehouse and transportation systems, self-healing inventory, enterprise inventory, and associate-facing agents as live in named international markets. The disclosed jobs include aligning orders with store demand, flagging anomalies, redirecting excess stock, and giving teams a shared inventory view across stores, fulfillment centers, and online channels.

A later Walmart Global Tech account explains the order flow in more detail. Forecasting models help position products before an order arrives. Walmart's Fulfillment Engine then uses multiple agents and real-time decision logic to select a store or fulfillment center while balancing availability, speed, capacity, and other constraints. After that choice, an end-to-end agentic workflow coordinates routing, driver availability, assignment, order density, and timing.

Walmart also documents AI-assisted planning for disruption. A June 2026 Global Tech article describes teams using weather intelligence, network models, and digital twins of logistics networks to examine facility closures, delays, demand changes, and alternative fulfillment paths. The software supports planners; the page does not say a model independently commands the entire supply chain.

These examples also show why "AI-powered store" is too vague to be useful. Forecasting, optimization, computer vision, generative interfaces, and physical automation may interact, but they are not interchangeable. A conveyor, camera, or robot should not be called AI merely because it appears inside a facility that also runs predictive software.

Shopping through ChatGPT and Gemini remains a separate track

Walmart wants its product and fulfillment systems to meet shoppers outside Walmart's own search box. In October 2025, Walmart and OpenAI announced a ChatGPT shopping experience using Instant Checkout and said customers would be able to shop Walmart through ChatGPT soon. In January 2026, Walmart and Google shared plans for a Gemini experience that could connect discovery, account-linked recommendations, carts, and Walmart fulfillment.

Both are evidence of direction and integration work. Neither announcement, by itself, proves general availability in every account, region, product category, or membership state. Walmart's current technology overview calls them announced integrations. Until an official availability page gives a firmer boundary, they belong in the announced column rather than the live Sparky column.

Public disclosure leaves important blanks

Walmart names many systems and explains what business step each one supports. It publishes far less about model selection, training-data lineage, evaluation sets, error rates, override frequency, incident reporting, and the exact human-approval rule for each action. It also does not provide one current deployment matrix covering every store, market, shift, associate role, and customer account.

The company says its AI work follows a Responsible AI Pledge centered on transparency, security, privacy, fairness, accountability, and customer focus. Walmart also describes an internal governance process in its digital citizenship overview. Those are public principles. They are not a substitute for product-level documentation, external audit findings, or a published evaluation record for Sparky, Wally, or a routing model.

Job impact is another blank. Walmart repeatedly frames these tools as support for associates, yet the cited pages do not establish a complete causal account of staffing, workload, surveillance, or role changes. Walmart's own Jobs Spotlight report from July 16, 2026 says it is too early to predict AI's effect on job elimination. The same caution applies to business outcomes. This article omits the company's promotional time and savings figures because they are not needed to understand what the systems are designed to do.

Read the next Walmart AI claim in five passes

A new Walmart AI announcement becomes easier to evaluate when the product name is less important than its operational boundary. Five questions keep the reading grounded:

• Who uses it? A customer, merchant, store associate, support agent, driver, supplier, or developer faces a different data and decision context.
• What action can it take? Summarizing reviews is not the same as changing an order, moving inventory, assigning a driver, or issuing a supplier instruction.
• What is the stated status? Look for direct terms such as live, available, pilot, selected locations, plans, soon, and emerging.
• Where does a person remain involved? Find the review, override, escalation, refinement, or approval step instead of assuming either total autonomy or total human control.
• What evidence is missing? A named capability without scope, evaluation, or update history should stay a bounded claim, not become a chain-wide conclusion.

The durable picture of Walmart AI is not a futuristic store run by one machine. It is a retail stack in which specialized models and agents sit beside product data, associate apps, inventory systems, fulfillment logic, and human decisions. Some parts are plainly in operation. Others are limited trials or public plans.

Keeping those states separate gives a more useful answer than repeating Walmart's marketing language or dismissing the whole program as hype. It shows where software is already doing defined work, where people still shape the outcome, and where the public evidence stops.