Intelligent Operations at Enterprise Scale - with Dave Glick of Walmart
The AI in Business Podcast · 24 Aug 2026
Source: https://podcast.emerj.com/intelligent-operations-at-enterprise-scale-with-dave-glick-of-walmart
Transcript: https://podcasts.opennash.com/shows/ai-in-business/intelligent-operations-at-enterprise-scale-with-dave-glick-of-walmart-d24d70/ (via OpenNash Podcasts, machine-generated)

Welcome everyone to the Emerj AI in Business podcast. Today's guest is Dave Glick, senior vice president of growth technology at Walmart. Dave discusses how Walmart's engineering organization is shifting from reactively triaging support tickets to proactively identifying and fixing small, recurring issues before they escalate. He shares how AI coding agents, an internal agent registry, and a company-wide launchpad are helping engineers move faster while keeping governance intact. Today's episode is sponsored by BigPanda. A quick note for our audience that the views and opinions expressed by Dave on today's program are his own and do not reflect those of Walmart or its leadership.

If you offer AI products or services into the enterprise, you need to find enterprise leaders with relevance. That means the right title and the right type of organization, and of course, readiness. Emerj attracts VP and higher ranking enterprise audiences who are already convinced that they need to move beyond traditional IT. To learn the exact strategies we use to help leading AI brands and startups connect with their ideal enterprise AI buyers, visit Emerj dot com slash AD1.

That's Emerj dot com slash AD1. Now the conversation with Dave Glick. David, welcome to the show.

Thanks for having me.

So we've had you on before, but I'd like to understand what you're busy with at the moment at Walmart. Tell us more about what you're working on.

Yeah, a few months ago, I took a new role in a new global team responsible for our growth businesses. And so that's advertising, uh, data ventures, Walmart Plus and marketplace. And these are all the things that are driving high margin growth at Walmart.

Well, you seem to be busy. So our last episode was extremely interesting. Not so long ago, you talked about the factory of nano agents, where it's not about one monolithic agent that runs it all. It's about building very focused agents for specific tasks and then ensuring that they can deliver on those tasks. So we're sort of going to zoom in a bit closer on that specific idea. And let's look at engineering, where we are solving incidents, servicing the entire company when it comes to the data and services.

And what that looks like when all of that overload comes to up until now humans, it's the triage, it's figuring out what to focus on first, who do you need to fall into a room to solve a problem? And now we're handing that off onto agents. And this ties into what you said last time. We're using the agents, but within the engineering business unit, what does that look like now? And what are the biggest pressures that you still see, even when we've got the agents employed and what we need to think about?

Yeah, I mean, since since I've joined engineering, which was about 25 years ago, there's always been the pressure on the business and the product team. Want new features, we want to move big rocks, and that's, you know, adding new features to our seller center, globalizing the platform or launching new categories. We have to balance that with, like, making sure what we have now runs. And you know, you've heard people talk about technical debt. I don't really like that term, but like, I I think of it more of as operational excellence.

We want to increase production capacity. We want to spend our time producing more features that reduce costs and drive revenue for the business. But we need to keep keep the shop running, and so that's always been sort of the tension. And so one of the things we've tried to do is as I joined, balance this. You know, how many tickets do we have? What are things that are little? We call paper cuts that hurt our customers.

You know, one of the paper cuts we looked at the other day is, oh, we can only upload X megabyte images. Where did we come up with X? Why isn't it Y? And so we look at those magic numbers and say, can we make our lives better for sellers? And so, you know, going from X to Y megabytes means that we have less errors and the sellers are less frustrated. That seems like a good thing to do, and it was like a one line code change.

And so balancing all the big rocks we need to move with what can we do simply and easily to to make our sellers lives better, to reduce overhead. You know, we get a bunch of tickets come in in each week. And what sometimes you do is, you, oh, let's throw more L1 support at it, let's throw more L2 support at it. It's like, why don't we go and fix the problem in the first place? And you know, my mantra is eliminate, automate, optimize, right?

The best ticket is no ticket. If we get get out of a whole class of tickets, we can spend our time building features that affect customers and sellers. And, you know, one of the great things about AI, we've been doing triaging tickets for decades, but using AI to do that. When I started, we had a whole team whose job it is, we called it ticket diving, right? And go in and look at what the tickets are and write a summary report every week for that.

And now we can do that in minutes with AI, and it can find patterns. And so we're excited about how we can apply that to making our software engineers lives better. They spend less time doing repetitive tasks and more time building and moving big rocks.

How is that a complete shift within the engineering department's thinking with agents involved? Because it seems to me that there needs to be a clear shift in the way you're thinking.

Yeah, well, you know, one of the things we started last year, as we were thinking, anytime something was going not well, we'd say, how can AI help us? And one example was we were launching, uh, our careers site, and, you know, we were in the testing process. And my team was like, we have 15 bugs open, we're closing about 15 bugs a day, so we should be okay. And I went and talked to the product team and the customers, and they're like, Oh, no, there's all these edge cases we haven't thought of.

And so, like Friday night at five o'clock, I called one of the engineers and said, How can AI help us? And he said, Oh, I'll just take my coding agent and point it at the GitHub repo and tell it to write all the tests. And so I was like, Really, how long will that take you? He's like, Oh, five minutes. And, you know, it ended up taking a week to get it really dialed in. But that would have taken us months to write all those tests, and we would have never hit the date.

So a year ago we were saying, how can AI help us when something goes awry? Now, we don't have to wait for something to go awry. We're just saying, how can AI help us? And I was, I was talking a little bit about this paper cuts program that we started, which was like, oh, sellers get this error message when they try to upload something and like, let's fix that. Why do we have this five megabyte limit instead of something bigger?

All those things. And the team, the FIR, when the team started, they were like, Oh, we'll prioritize this as part of Q3. And we said, No, how about tonight? Like, let us get in the cadence of delivering things faster, because with the AI, you can deliver things faster. But but it's a a mind shift where people are thinking of, oh, we need to plan things and prioritize it and talk to the product team and talk to the business team.

It's like, no, we've got a list of a hundred things that should be relatively minor fixes. We should go knock those out a few of them every week. And if you do that over the period of, say, six months, like, things will get better. And in fact, I'm after this interview or after this podcast, I'm going downstairs and meeting with some of our relatively junior engineers. And we're gonna test whether I can, like, just feed the paper cuts into the coding agent and say, tell me all the ones that you can fix.

Go ahead and write a PR for us to review, and then we will. Ideally, the AI could march down all 100 of them and fix them all. Have we just code review those? I don't think we're there yet. So today we're doing the test of the 100 open items. You know, how many of them can the AI just knock out? And so those are all the things that we couldn't have even thought of 18 months ago, and now we're testing them every day.

Are you framing a picture where instead of having agents handling the very few incidents and coming to a point where you're sort of predicting what could be incidents and already solving them before they are incidents? Within that whole nano agent factory type thinking, where you're building these agents? And it seems to me that you are basically going in and looking at patterns showing up and then deciding, sending an agent to solve for that. How is this all managed?

Because I'm thinking about how difficult it is to manage people within a room. When you've got all of these nano agents, how do you manage them? How do you ensure that it's observable?

Yeah, well, it's. It's a good question, and that's something we're struggling with every day. I think a year ago we took sort of the mission, you know, Walmart's a people led technology enabled company, technology powered company. And we wanted to say, like, how do we put AI in the hands of every campus associate every day? That was the mission 18 months ago. And we thought that was like getting everybody licenses to Gemini, ChatGPT and copilot. And what it's turned out we've done so much more.

We we wrote our own coding agent, it's called Code Puppy, and we thought it was gonna be for the engineers. It was. It was a couple engineers wanted a command line interface tool, and they wrote it over the weekend. And then we started installing it on some of the operations folks boxes. And they're like, we love this. It's like a command line interface on teraterm, and the users love it because then they can solve their own problems.

We've got folks who are writing their own reports, and they said, we've been asking for the for this from engineering for three years, and I was able to write it myself in three hours. So we've got a lot of popcorn popping, right? We've. We've added a lot of energy into the system. But the focus now is to govern all these things is to say we've built a registry so that people can say, I want an agent that does this.

Let's go see if there's already one. We've got what's called a launch pad because we had a lot of people building things on their local machines. But how do we get that into production? How do they share that? So that's all things we're building. And so I, I don't say we have it solved yet, but we're on the right track. I, you know, I think one of the things there's always a challenge between how to do governance and how to move quickly.

And I think we've struck a good balance to say we already had data privacy policies, we already had data governance policies. Let's use those and then see how it goes. Yeah, and we can add as we go. And, you know, we're always watching the news. And we're, you know, we've got an infosec team, which is first class, who's watching what we're doing. And got a compliance team who is also using our coding agent, and so they know what it can do.

And so I think getting everybody involved has been a great, uh, opportunity for the compliance team and the infosec team and all these other teams who are in charge of keeping us safe to know what we're doing. And so I won't say we have it solved, but I think we're we're marching in the right direction.

So you sort of brought something up that's been in the mouths of a lot of our leaders. Is we looked at it actually as a concentric circle where you always had the engineers in the middle and all of the questions had to go to the engineers. And building in these layers where you've got another layer of responsible person. And they can answer their own questions, and only once they cannot answer their own question with the help of AI, then they go back to the engineer.

And I don't want to put it so plainly, but what are the engineers up to? If they're being relieved of all of the tickets, they're not handing them anymore, they're lowering the tickets. Not only are the agents triaging and solving the tickets, they're actually lowering the amount of tickets by predictively solving for those tickets. What are the engineers doing on a day-to-day basis? What are they keeping themselves busy with if they don't have to answer all these questions?

Well, I mean, that's interesting. I mean, ideally, they wouldn't be. This is kind of engineering nirvana, right? But if you think about the process at big companies, is every year we get together and do an annual operating plan or an operational plan, one or whatever you want to call it. And the business team submits, here's all we want you to do engineering. And then we go away and we come back and say, Oh, we can do these three things.

Everything else is below the line. And so for years, people have been living with this. And so now they're saying, Oh, this is below the line. Great, I'll do it myself. And so it's not we're taking that stuff off the engineers plate, we're taking that off the stuff that's never getting done. And so rather than saying the engineers have less to do, we're saying more, the pie is getting bigger. And so I, but we what we want the engineers to do is spend their time moving the biggest of big rocks.

So globalizing the platform, you say, Oh, we have X hundred services all need to be made global. And you say, Oh, if we could do one of those per day, it would still take me years to do. And we, and you know, those traditionally would be, you know, a week or two project. And so what if the AI can go through and help the engineer globalize each of those 400 services? Now we've got a global platform.

How do we simplify our set of APIs so that it's easier for sellers to connect to us? Those are big engineering tasks, which the engineers would rather be doing than solving tickets about this error message came out. Traditionally, engineers spend a lot of time being digital detectives, you know, they get a ticket and they try to have to figure out what happened. They try to reproduce it. And so that is toil. We want to remove the toil from our engineers, um, as well as from our operators.

And it opens them up to a lot of creativity. I'm sure a lot of ideas are even just bubbling up from their side. It might not even be on the list of company goals and company hopes and dreams. It that they're coming up with their own ideas as it's surfacing. So linking back to the fact that they don't have to deal with the incident reports anymore, or not as much as they used to. They can now focus on higher level tasks and just the iterations getting so much faster.

And you spoke about that last time, the time cycles are just speeding up completely. How important is it to have one single space, let's call it a platform, where all of these things come together? You mentioned already registry and a launch pad, because I've heard this before. Somebody within the company wrote a wonderful thing, but it's sitting on their own computer. And now when you try to roll it out, it's either localized or they that they want to hold onto their own information there.

How important is it to have a layer somewhere where all of this lives and it can be governed and the security is in place for that?

It is, and you know, governance is super important, you know, obviously, infosec is paramount. The most important thing we do is take care of our customers data and our associates data and all and our data. But but yes, we need governance. But you could have, we could have sat here for the last 18 months and said, let us think about everything that could go wrong and put guardrails in the place for those things. And we would still be where we were 18 months ago.

And I talked to my friends at other big companies, and they're still trying to get, you know, ChatGPT through their security process. And so, you know, we've basically said, like, let us start doing. And then again, we have our. We already have data governance and data privacy policies and infosec things that scan. And let's start with those. And what we found is, yes, we need a sharing mechanism. We didn't start with, oh, let us create a sharing mechanism, and this is how we're gonna spec it.

We started with, like, oh, we've got a bunch of agents and people want to share them. So let's build. And and one of the people in the community built like a sharing mechanism, and we crowdsourced that into our agent. And so it's not necessarily reactive, but it's sort of we want this thing to be viral and bottoms up. Because, you know, we can be much more nimble if we're thinking about it as we go, rather than trying to plan everything ahead of time.

And we've heard that as well is don't get stuck in planning or analysis, paralysis, where you sit and try to figure out all of the things before you start enrolling them. And from that mindset, and you've said that that on the previous episode, it's it's have small agents. At what point do you see? Say, okay, now this is big enough to scale, or this is safe enough to scale. Are there certain indicators that this is going to be safe?

We can deploy this now. How do you manage that part of the process?

The first thing, obviously, is staying safe. And again, we have guardrails in a variety of different places. But then it's like, how can we create these agents faster? And and we're all experiment, we don't know, nobody knows what the right answer is, so we learn something new every day. One of the things I've done is started creating AI office hours. And so every every Monday morning, if you've created something with AI in the last week, you can come in and demo it to me for whatever that's worth.

And if it's something that gets legs, then, you know, I can put some more engineers on it and so on. But I think that keeps me closer to the technology. Is you have lots of lots of people who are on Twitter saying, Oh, we're gonna use agents for everything, but don't really understand exactly what that means. And so I like to be close to the technology. And I learned something super meaningful over the last few weeks that, you know, as I do these AI office hours.

About three quarters of the things people present are things that are helping them reduce toil and not production features. And at first I thought, well, you know, maybe it's because they don't trust it for production, they don't trust the AI for production features. But then someone said to me, Well, maybe that's this is what they spend their time doing, right? Doing change requests and release management and security protocols. And and they would rather be building things, right?

They'd rather be m moving big rocks. And so we actually had a um a hackathon last week and the the theme was remove toil. And because, as you said, without the right governance, we're gonna have 100 release management agents and 100 testing agents. And, you know, 100 other things that live on people's boxes. And so what we did is said, like, let us look at the whole software development life cycle and say, oh, well, we do need one release manager that we bet on.

We need one test building thing, or, you know, maybe two or three, but not 100. And so we sort of federate this across the whole environment and people. So the engineers actually got to be the customers of themselves. I've been saying for years that when the customer and the engineer is close together, good things happen. And so this was a case where the engineer got to be the customer.

That's sort of a perfect way to start and build trust. Because if it's solving your problems within your own day to day processes and you speak about the accomplishments, you're sort of celebrating the wins, that already builds trust. And of course, it's completely different when you're in a regulated industry. All of this gets a lot more confined and you have to be more stricter. But in a retail industry, you sort of have a bit of room to play, as long as you keep those guardrails that have been worked on for such a long time.

I mean, they've been in place for a long time. It's not something new, so that makes sense to me. But there are some of our leaders that aren't there yet. The company hasn't transformed in the way that that Walmart has. What are those first steps? If you have to think back, you're talking about 18 months, but it's longer when all of this started. What are those first steps? They have to start thinking about. Let's start with the incidents where we started off today.

If they just want to start there, and then what the next step would be to get to this point where you're painting a future where everybody within the company is a user. And then you can also run this out to the public, to your users outside the company. But what is are those first steps they need to think about?

Um, I'll give you the first two steps. Uh, first, put AI in the hands of every user, every day, you know, whether that's one of the, uh, you know, the big frontier model LLMs with their UIs, or a specific coding agent or something you build yourself. Every user use AI every day. And the second is ask yourself how AI can help. Oftentimes, we only ask ourselves how AI can help if something's going wrong. But every morning you should look at what you're doing today, ask how AI can help you.

I'm going to close this off, but I'm going to ask you if there's anything I've missed that you'd like to leave the leaders with. But the way that I would sum this up is number one, you just said it starts somewhere and starts small. But what you're saying is sort of close the gap between the engineer and the user, because that's where you're going to find the uh, fastest iterations and the fastest successes. Because a lot gets lost in translation when there are too many people involved.

I like the idea of having the engineer next to the person that's going to be using it, because that closes that gap. There's no way that you can miss what's not working when you're sitting next to the person using it. And then, especially when it comes to incidents, et cetera, the ideal state is not managing the incidents or lowering the amount of incidents people have to deal with with AI. It's actually a step further. It's using the incidents and building a system where you're optimizing and solving problems even before they come up.

And it can be as simple as the user wants to upload 20 to 30 pictures instead instead of five. But look at those spots where it's such an easy fix for the agent and that will already lower your incidents. And then have a place where everybody can work on their own, but then register it so everybody knows what everybody else is working on. And then a launch pad, so all of this can be governed to a certain sense within the rules you already have.

Don't go rewriting the the rules, the governance, et cetera. You already have them, just enhance them where it's necessary. What did I miss and what else should our leaders be thinking about?

Wow, you did a great job of recapping that again. Dive in, try things out. I think that's number one, number zero, one and two. But then, you know, don't forget that. We want our business partners and our product partners, and we all want to be moving things forward for the customer, the customer, the seller and the associate. And so all these other things are distractions. But if you let the distractions hang around, they will get bigger and bigger and bigger.

So we need to spend time knocking out these paper cuts and these little grains of sand, which which often people never get to.

Oh, thanks so much for your time, David. I hope we'll have you back again and good luck for that meeting coming up.

Okay, thank you for having me.

Wrapping up today's episode, I think there are three key takeaways from our conversation with Dave. First, shifting from reactive incident triage to proactively fixing small, recurring issues frees engineering teams to focus on larger, higher impact work. Second, closing the gap between the people building a solution and the people using it produces faster iteration and fewer miscommunications. And finally, applying existing governance and data policies to new AI tools, rather than pausing to build new frameworks from scratch, lets organizations move quickly without sacrificing safety.

If you offer AI products or services into the enterprise, you need to find enterprise leaders with relevance and readiness. Emerj attracts VP and higher ranking enterprise audiences who are already convinced that they need to move beyond traditional IT. To learn the exact strategies we use to help leading AI brands and startups connect with their ideal enterprise AI buyers, visit Emerj dot com slash AD1. That's Emerj dot com slash AD1. On behalf of the team at Emerj, we'll see you on the next episode.