The quarterly roadmaps, tiered approvals, and monolithic systems that are the backbone of traditional AI strategies in retail are exactly what holds many retailers back today.
These weren’t bad decisions. These were a rational response to a world where construction and transportation technologies were slow, expensive, and difficult to reverse. That world is gone. According to a report from the Brookings Institution, enterprise AI adoption has jumped from 55% to 78% in one year, but the majority of organizations are still deploying AI as a single, monolithic function rather than a flexible, collaborative system.
According to research from MIT Sloan, AI tools are already increasing developer productivity by up to 39% and shortening delivery cycles that once took quarters to weeks. It is clear that technology is not a challenge. Stanford University’s Digital Economy Institute found that 77% of the most difficult challenges in enterprise AI adoption are organizational rather than technical.
The operating model built for scarcity is now a major constraint in an environment where scarcity no longer exists.
David Glick, SVP of Enterprise Business Services at Walmart, joined Emerj’s Matthew DeMello to explore why enterprise AI stalls when the operating model doesn’t change, and how moving from quarterly planning to real-time iteration, federated agent architecture, and automated governance can measurably improve speed and reliability.
This article explores three core insights for retail technology leaders as they move from traditional AI deployments to real-time federated execution.
- Replace quarterly planning cycles with the introduction of a stopwatch. Compressing iterations from quarters to hours reduces rework, keeps governance aligned with the actual code, and eliminates the planning overhead that delays most enterprise AI efforts before they scale.
- Moving from monolithic AI systems to federated nano agents: Deploying a network of small, task-specific agents coordinated by an intelligent routing layer provides faster execution, clearer ownership, and more manageable complexity than a single, centralized deployment.
- Building the infrastructure to build the agent: Investing in a repeatable agent development platform rather than a one-time deployment is what separates organizations that scale AI from those that remain in pilot mode.
Listen to the full episode below.
Episode: How Walmart is redesigning AI delivery speed – A conversation with Walmart’s David Glick
guest: david glickSenior Vice President of Enterprise Business Services at Walmart
Expertise: GenAI Transformation, Enterprise Technology Operations, E-commerce and Retail Infrastructure, Operational Excellence and Scalable Systems
Easy recognition: Dave Glick led large-scale technology and operations initiatives at Walmart and previously served as CTO of Flex. Prior to that, he spent 20 years at Amazon building foundational retail and operational technologies, including Amazon’s proprietary automated pricing systems, warehouse management systems, and transportation platforms such as Amazon Flex. David holds a Ph.D. in physics from the University of North Carolina at Chapel Hill.
Replacing the quarterly planning cycle with the introduction of a stopwatch
David Glick draws a direct line between the pace of planning and AI failure. When iteration speeds are reduced from quarters to hours, the cost of getting priorities wrong dramatically decreases, and with it the overall logic of a strong planning cycle.
The practical implication for retail technology leaders is that the faster your team can iterate, the less valuable your planning overhead will be. Instead of spending three months on discovery and three months on UX to get one thing right, teams that can prototype in real time with end users can pivot in minutes. SVP explains:
“When you can only do one thing a year, it’s better to do the right thing right. Spend a lot of time on overall prioritization: three months on discovery, three months on UX. If you can do 50 things a year, it might not matter which one you do first.”
– David Glick, Senior Vice President, Walmart Enterprise Business Services
It’s a reimagining of how retail leaders should think about planning overhead. The goal is not to plan better. It’s about iterating faster and making planning less important. Three operational shifts support this:
- Prototype in real time with end users Rather than gathering requirements in advance. Glick’s team compressed work that once took months back and forth into a single afternoon session.
- Replace sequential approvals with parallel governance. Security and compliance processes should run alongside development, not after it.
- Measure in hours and days, not quarters. Planning units must match delivery units.
Transitioning from monolithic AI systems to federated nano agents
The dominant mental model of enterprise AI—one big, centralized system that handles everything—is not just inefficient; This is architecturally out of sync with how you actually create value at scale.
Glick draws a direct line from the failures of monolithic software to the limitations of monolithic AI. When hundreds of engineers checked in a single codebase, they encountered the same problems: too much coordination overhead, too slow speed, and too many dependencies. Moving to microservices solved the software problem. Nanoagents solve it for AI.
The distinction is actually important.
- nano agent is a small, single-purpose AI tool built and owned by a domain-specific team. Each solves one problem well rather than many problems well.
- super agent Acts as an intelligent dispatcher. This is a routing layer that directs requests to the appropriate nano agent without the user knowing where to look or what to request.
- thick agent It sits in the middle and handles domain-level routing before handing off work to more specific agents below.
The senior vice president explains the principle with a simple analogy. “The Swiss Army knife does many things, but not enough. A fork, a spoon, and a knife – those three things will usually make your dinner better than a pocket knife.”
For retail technology leaders, the operational implications are clear. A single monolithic AI deployment introduces a single point of failure and a single point of slowdown. Federated networks of task-specific agents create resilience, speed, clear ownership, and scale in ways that centralized systems cannot.
Building the infrastructure to build the agent, not just the agent itself
David’s most sophisticated insight in this conversation is not about any particular agent. This is about what happens when organizations stop treating AI implementation as a series of projects and start treating it as a manufacturing capability.
Glick calls this “the machine that builds machines.” The idea is simple. Rather than investing engineering effort to build one agent at a time, invest in platforms, processes, and standards that make agent development reproducible, fast, and scalable across all domains in your organization.
The practical differences between these two approaches are:
- project thinking produces individual agents that behave well on their own, but require the same effort to build each time.
- platform thinking Generate an agent factory. This is an infrastructure that domain teams can use to launch, test, and deploy agents from scratch.
Glick is candid about the current situation in most organizations, including his own. His team created three separate agent-building platforms across various functions such as finance, human resources, and operations, each optimized for its own domain. Rather than treating this as a problem to be eliminated, he sees it as an acceptable stage of maturity.
“We’d rather have two things that do something than do nothing. The people in my finance technology team built what works best for them. The people in my HR technology team built something a little different, but it works well for them and gives them access to the data they need.”
- David Glick, Senior Vice President, Walmart Enterprise Business Services
Rule of thumb for retail industry leaders: Early duplication of agent infrastructure is acceptable. There are no absences. The priority is to let the domain team build, iterate, and integrate what works. Organizations that wait for the perfect integration platform before implementing will find themselves years behind those that build sloppily and learn quickly.
