The AI landscape is quickly evolving, fueled by learnings, new models, robust infrastructure and more. Specifically, Agentic AI will become integral to modern enterprises, not just enhancing productivity and efficiency, but unlocking new levels of value through intelligent decision-making and personalized customer experiences.
While the technology is still in its early days, Walmart is moving rapidly and intentionally to integrate agentic capabilities into existing workflows across the business, creating powerful new capabilities as we build the future of retail.
Built-to-purpose technology: Agentic AI
Our approach to agentic AI at Walmart is surgical. Extensive early testing proved that, for us, agents work best when deployed for highly specific tasks, to produce outputs that can then be stitched together to orchestrate and solve complex workflows. As a result, we are hyper-focused on solving for specific use cases tailored to the unique needs of our business, versus providers that are likely building for multiple potential use cases. In parallel, we are also focused on defining and refining the set of core agentic AI capabilities that are foundational to how we build cohesive, best-in-class tools and experiences that can evolve and scale globally. For example, we leverage our retail-specific LLM to build agents for tasks such as item comparison, deep personalization, shopping journey completion, etc. within our GenAI-powered shopping assistant. It is trained on Walmart data, enabling us to combine it with other LLMs to create responses and complete tasks that are highly contextual and tailored to the customer’s specific needs.
Early success and meaningful impact of agentic AI
AI has long been highly pervasive throughout our business, and the path to agentic AI has been paved — many of our GenAI-powered copilot tools are well on their way to becoming assistive agents to fully autonomous agents.
From merchant tools that automate a range of time-intensive tasks, particularly entry and analysis, to Trend-to-Product, which shortens the traditional production timeline for Walmart fashion by as much as 18 weeks, we are exploring agentic systems to further optimize activities across our ecosystem such as associate tasks in our stores, customer shopping journeys online and merchandize planning in our home office. In our Customer Support Assistant, agents are already routing, resolving and, increasingly, acting autonomously to automate the mundane and free associates to focus on more complex tasks. Our GenAI-powered shopping assistant uses multi-agent orchestration, fallback handling and evolving voice/camera capabilities to support customers from discovery to purchase.
In-store optimization is driven by the integration of agentic capabilities into a range of associate tools, automating day-to-day tasks which creates more time for associates to focus on more complex and fulfilling aspects of their roles, delivering excellent customer experiences. Underpinning it all, developer productivity is accelerating, with agents handling tasks within our CICD pipelines such as test generation, error resolution and environment setup, freeing up our developers to spend more time on strategic initiatives and innovative features because we believe that the best way to unlock transformation is through the creativity of our associates.
Preparing for personal shopping agents
The rise of personal shopping agents will require a collaborative approach between retailers, providers and customers. It’s not just about how we prepare our tools and experiences for personal agents, it’s also about how we build infrastructure with providers and how customers learn to effectively utilize these tools.
Currently, the experience of using an agent to shop is similar to a sophisticated web search with automated purchasing capabilities. To truly unlock the potential, two key pieces need to fall into place:
- Customers need to effectively train their agents. This involves providing specific query parameters – think budgetary limits, brand preferences, sizes, colors and even preferred store locations. Crucially, providing feedback to the agent over time will allow it to learn and adapt to individual needs. Consistent use will be key to the agent truly understanding its user.
- Retailers and providers need to build the bridges for personal shopping agents to effectively communicate shopper needs to internal agents. This will allow us to validate whether we can meet the needs of the customer’s personal agent and help facilitate the purchase.
We are seeing early success in item research, as customers are already using GenAI extensively for this. Agents can perform that same research, and, assuming the shopper has trained it well enough, make decisions and take actions based on that research.
We also see agents being particularly useful in shopping for essentials, given the regular cadence of purchasing everyday items. As the technology matures and users better understand how to maximize its potential, agents will be capable of handling more complex tasks.
Marketing and agent discovery
The way an agent shops is fundamentally different from the way a human shops. For example, agents may be less likely to be attracted to images or visuals designed to elicit an emotional response. So, while Walmart’s core value proposition – Every Day Low Prices – and the spirit in which we market to customers won’t change, we need to develop new pathways for agent discovery.
These new pathways include things like advertising strategies tailored for agents and the development of agent-specific SEO. This won't replace existing advertising methods but rather complement them, reflecting the evolving ways our customers choose to shop – much like the introduction of social media.
Navigating the complex challenges of agents
Developing robust and reliable agents presents unique challenges. Above all, accuracy is paramount, and we’re exploring ways to leverage agents for critical functions like governance and checks-and-balances. Often, a co-pilot model, with humans and AI working as a team, is the most effective approach.
Another key consideration is the human threshold for automation. We're carefully evaluating which actions are best suited for autonomous agent execution and where human oversight and approval remain essential. This highly intentional approach ensures we're building systems that are not only operationally efficient and trustworthy but keep human experience at the center.
What will agentic AI look like in 3-5 years?
For retailers, the potential for agentic AI is immense. Imagine complex personal shoppers that understand nuanced preferences, dynamic store environments that adapt in real-time based on customer needs, and self-optimizing logistics networks that ensure products are always in the right place at the right time. This is why we're so invested in building the right foundation today, to not only meet but anticipate the needs of tomorrow.
By focusing on strategic implementation, addressing challenges with precision and intention, and fostering a collaborative ecosystem, Walmart is spearheading the next major evolution of retail.
