Shawn Hakl: [00:00] If you think about our role in this, we have more endpoints than anyone else. We’re the origination point of the transaction. So, to the extent that I make the journey for the customer to get from the edge of the network into AI as simple, secure, and reliable as possible, the better that overall end-to-end experience is. So, hyperscalers in our case are a natural partner to get that done.
Danielle Rios: [00:19] Got it.
Announcer: [00:25] This is “Telco in 20,” a podcast that helps telco execs achieve a competitive advantage with AI and the public cloud. It is hosted by Danielle Rios, also known as DR. Today we’re talking to Shawn Hakl, Senior Vice President of Product at AT&T Business.
Danielle Rios: [00:45] Hi, guys, I’m DR. The AI race is on, and telcos want in. So far, the big ideas are sovereign cloud, AI factories, and GPU-as-a-service. Telefónica is leading EURO-3C. Telcos like Orange, Deutsche Telekom, and SoftBank are building AI factories, and Telus has already sold out its initial GPU capacity in Canada. Then there’s AT&T. In March, the operator announced something called AWS Interconnect – last mile, which is AT&T’s fiber and fixed wireless, plugged directly into AWS and engineered for AI workloads. AT&T’s point of view is that enterprises need more than just compute, they need the flattest, fastest path to AI.
[01:28] So, it’s doubling down on fiber and last-mile connectivity and embedding the AWS workflow as the AI on-ramp. Today, I’m joined by Shawn Hakl, Senior Vice President of Product at AT&T Business. We dig into why AT&T is partnering with hyperscalers instead of competing with them, how it cut latency from 110 milliseconds down to 40, and what an agent consumable network actually looks like. So, let’s take 20. Shawn Hakl is Senior Vice President of Product at AT&T Business. Hi, Shawn. Welcome back to “Telco in 20.”
Shawn Hakl: [02:04] Hey, Danielle. Thanks for having me. I’m super excited to be here.
Danielle Rios: [02:07] Well, I’m super excited to talk to you. You’ve had a job change. Congratulations. You’ve moved from Verizon to Microsoft and now you’re the head of product for AT&T Business. And so tell me a little bit about your new role and the organization that you’re leading.
Shawn Hakl: [02:21] Yeah. So, I own the product organization within AT&T Business, so that essentially covers everything that we offer to customers all the way from the small business into large enterprise, government, international customers, as well as our IoT business, which is obviously in a super exciting area. As well as the newest addition to the fold is all the work we do with the hyperscalers and the CPaaS players in terms of A2P and Network APIs, so another area that I think is a huge promise in the industry, and an opportunity for us to participate in monetization that hasn’t traditionally been available to the telcos.
Danielle Rios: [02:54] Awesome. Now, AI has come bursting onto the scene. Every time I log on to X, something new has been dropped, a new model, a new product. I can’t imagine what it’s like to lead an enterprise and now having to switch to support AI. And so I would imagine now having conversations with enterprise customers has totally shifted because of AI. And so what are enterprise CIOs and CTOs asking of AT&T that they just weren’t a couple of years ago?
Shawn Hakl: [03:23] Yeah, the conversations are definitely different. As you know, enterprise networking is always sort of tracked to the application ecosystem. As application architectures have changed, the communications and security needs change with it. So, mostly the questions we’re having with customers right now is, “Okay, in the age of AI, what do I have to do? What do I change?” So, really, if you think about that, in the age of AI, you need to have the flattest, fastest path to the AI engine. In essence, if you think about an agentic AI answer or using AI in a use case, in the real world, it used to be most of the cloud applications were developed to be 25 to 30 milliseconds away from the end user. Now, if you’ve got multiple agent interactions, that can get down to about five to seven.
Danielle Rios: [04:01] Yeah.
Shawn Hakl: [04:01] And then, of course, you can never have a discussion with AI without having a discussion about security and privacy. You have a lot of high value tokens, which means important information going up to the model and important information coming back, maybe even more so. And so being able to track, manage, and monitor where your information is going, who it’s being exposed to, how it’s being used, super super important element. And then the last thing that I think that’s starting to dawn on people is, even though all of these agents may look and act and talk like people on occasion, if you are provisioning and managing them and setting them up in the same models you use for employees, it’s slow. And so we’ve built a scale business around our IoT practice, and I think a lot of those practices are relevant to how we automate for agents.
[04:43] And so a lot of that discussion is: how do you expose programmatic interfaces or agent consumable interfaces so people can set up and secure and manage those resources? And people may say, “Okay, well, why are you talking to AT&T about that?” Well, if you think about the actors involved, you’ve got the end users, you’ve got the stuff in the cloud, you’ve got your data sitting somewhere and often your SLMs sitting in a protected space, and then you have to access these models, which people are typically accessing a number of them. The glue that puts that together, the person that has the visibility to help you manage that is us. So, there’s an expectation there from our customers that we’re going to help them get that done.
Danielle Rios: [05:18] So, are you guys advocating moving AI workloads from, let’s say, your previous employer or hyperscaler and into and owned by essentially an AT&T data center, an AT&T cloud, for lack of a better description?
Shawn Hakl: [05:32] No, we’re not going to be in the GPU-as-a-Service business at scale. I don’t have aspirations to take on the cloud players. There’s a lot of capital investment involved with that that doesn’t involve putting fiber and 5G in the ground, which is where we spend our money. However, they are natural partners. If you think about our role in this, we have more endpoints than anyone else. We’re the origination point of the transaction. They are the destination.
[05:53] So, to the extent that I make the journey for the customer to get from the edge of the network into AI as simple, secure, and reliable as possible, the better that overall end to end experience is. So, hyperscalers in our case are a natural partner to get that done. And you’ll see some of the announcements we’ve made. We met with the last-mile cloud connectivity partnership with AWS, which means essentially what you can do is you can manage the network as a resource in your Amazon instance, which means you’re setting up harmonized policy from the edge of the network into your cloud.
Danielle Rios: [06:23] Got it.
Shawn Hakl: [06:23] Why are we doing that? Because it’s the right answer for the customer. You can achieve scale. You can manage agents. Without that, trying to glue all those individual pieces together becomes too clumsy, and it’s not consumable. In essence, it becomes difficult to set that up. And so a lot of emphasis on that. And then obviously we do see the opportunity for hybrid models to evolve. In other words, people will call LLMs for some degree of the transaction, but on its flip side is you start to use AI for more critical business processes. It’s not just back office, but as you’re exposing AI for your most sensitive data, we do see a tendency for people to come back, and try to put that in SLMs with say control in a hybrid cloud model. So, it’s not an either/or, it’s a both.
Danielle Rios: [07:04] Got it.
Shawn Hakl: [07:05] And to the extent that we can facilitate that, we just feel like the adoption for AI will be easier, safer, and more secure, and that’s our end goal.
Danielle Rios: [07:12] Awesome. How are your teams changing the way they work day-to-day because of AI to serve these customers?
Shawn Hakl: [07:19] This is fun. As a product person, it’s paradise because instead of telling something about how great it’s going to be, we can sit and use… Whether it’s Lovable or Claude Code or whatever tool gets invented next week, I can show and not tell. So, in other words, we can literally go from concept to functioning execution fairly quickly. So, that’s very exciting. You can, for initial testing, build digital twins of your customers and start testing your initial concepts and simulating the experience for either the seller, the customer or the ops team within a simulated discussion of, “This is how the product’s going to function, this is how it’s going to be,” so you can optimize those experiences, starting with the customer experience, but also the seller experience and the operations experience. So, those tools give us a great opportunity. And AT&T is a huge consumer. We have more than 100,000 users that have been onboarded into our Ask AT&T, which is essentially a harness that sits in front of the LLMs. We consume roughly 27 billion tokens a day.
Danielle Rios: [08:15] As an organization, just internally?
Shawn Hakl: [08:17] As an organization, yeah. I haven’t started going into the product breakouts quite yet, but we’ll get there as well. And in terms of how often we’re using it, almost 1.7, 1.8 billion of those transactions on a daily basis are made in production environment, so this is using AI in the real world to have impact. We’re not just theoreticianing it out to our enterprise customers. This is on the basis of our own experience, as well as best practices with our customers in front of people. It’s been super exciting.
Danielle Rios: [08:44] Yeah.
Shawn Hakl: [08:44] Another example, you did ask me about edge AI, so we did recently do the release where we connected the edge of the RAN, the radio access network, for IoT transactions directly via specialized packet core into GPU-as-a-Service at the edge with our friends at NVIDIA. The important thing there is we are partnering not just with the end customer, but with the development community. So, this is one where we can sit with the development community, exposed through, say, NVIDIA’s developer ecosystem, which they’re great at, and build stuff real time where we can find out some real lessons. Like, we did find that AI at the edge does have a benefit. Those distributed models do have a benefit for processing video technology or audio.
Danielle Rios: [09:23] Super low latency required.
Shawn Hakl: [09:24] Yeah. We took it from 110 millisecond response down to a 40 millisecond response for a live customer for TanMar operating across about 100 cameras in five states, and that had value. Not everything. Probably not going to build PowerPoint at the edge with Copilot, but you’re doing license plate recognition or facial recognition where you’ve got to move quick and make decisions, there’s some value there. So, not suggesting the whole industry’s going to move that way, but there’s a set of use cases where this will apply.
Danielle Rios: [09:50] Some. Yeah. You sort of see these announcements in Europe, especially the sovereign cloud and the GPU-as-a-Service, and it feels like there’s a big question there of really you’re going to start to take away from the hyperscalers. There’s this whole software stack that you need, but I do agree there’s probably a smallish percentage that latency does matter and you need small language models or super low latency. Video’s a great example. So, yeah, that’s cool.
Shawn Hakl: [10:16] Hey, listen, our friends at NVIDIA are creating a super open dynamic ecosystem that’ll give a lot of opportunity to participate in the growth. We want to be part of that and part of enabling that. On the flip side, we’re probably not going to go try and compete head on with hyperscaler, especially in the US market where we just don’t have that same dynamic. It’s just not as relevant. They’re partners to us, and they’re partners driven by the fact that that’s where my customer wants to go.
Danielle Rios: [10:38] Yeah. There’s a lot of gravity because they’ve already moved workloads to the cloud, so it’s going to be almost impossible to pull them out of that.
Shawn Hakl: [10:44] Plus, if you’re customer-centric, you partner up with people to make sure that it’s the flattest, fastest, easiest path for my customer to get there. I don’t create value by being in the way.
Danielle Rios: [10:53] Yep, absolutely. Now, you’ve used the phrase agent-consumable to describe where network interfaces are headed. I think a lot of the interfaces that we’ve built to date in software assume there’s a human operator, there’s a human on the other side. Now agents need to be able to call any piece of software made by any vendor and take action. How are you guys making your network and your products agent-consumable?
Shawn Hakl: [11:17] Yeah. So, again, a couple things. You have to start viewing your network as a layered entity. In other words, you’ve got a set of physical resources that you build upon. Inventory has to be discoverable, reservable, et cetera. So, first of all, you build those interfaces that let you capture that.
Danielle Rios: [11:32] Yeah.
Shawn Hakl: [11:32] Then you’re going to build higher order services for intelligence where you’re exposing policy in the network. And when you do that, to the extent that you can harmonize those implementations and make them as industry standard as possible, it lets people build those chains across the different entities, so that you can get from the edge into the cloud. So, there is building the baseline consumable resources, letting people do basic capabilities. Then there’s building the intelligent network layer on top of that, which lets people set up the way that they want to manage that pipe, hopefully in harmony with the entities they’re trying to get through. And then you’ll expose higher order services into the development community, so pushing that up through established developer ecosystems.
[12:09] People will not consume those services that have to relearn the whole solution. A lot of them are very comfortable building through GitHub in the Azure or the Google or the Amazon developer ecosystems. NVIDIA has done a great job that way. So, it’s making sure that you’ve got a firm view of what your baseline building blocks are, and then making people able to consume as they go up the stack, and then making sure you don’t build backwards dependencies. In other words, the ability for you to get to a physical asset you need to access can’t be dependent on you consuming a higher order service for me because I’m basically trying to force you into my ecosystem. That doesn’t work. So, it’s doing that. And then exposing all of that so agents can discover them, use them, and then you’ve got true zero-trust security, so you can successfully expose those elements without breaking anything or exposing customer data.
[12:52] So, when I talk about that, it’s making sure you have the right APIs in place, which are agent discoverable and consumable. It’s about making sure you’ve got zero-trust security in place to make sure that by exposing them, you don’t create unintended side effects because a lot of permutations and combinations. And then exposing them through the right developer ecosystem and the right tools so that they can be found, discovered and used.
Danielle Rios: [13:12] Well, I imagine in thinking of that end developer customer, you would have to really partner with your internal people at AT&T and kind of take an approach that AWS did when they first started building out APIs for every product, and making sure you guys are eating your own dog food, making your own internal teams communicate through these agent-consumable interfaces or APIs just to dumb it down. And so are you guys starting to do that, where even internally you’re consuming services from other components of the network so that you’re hardening it, making it accessible, figuring out what APIs you really do need to make this all work together, so that your developers have that really great experience?
Shawn Hakl: [13:53] Yeah, that’s 100%. So, obviously I feel like I have some advantage from my background and my time at Microsoft-
Danielle Rios: [13:58] For sure, yeah.
Shawn Hakl: [13:59] … seeing how they develop and the development practice they enforce. We’ve brought folks into the business that have a strong history with software defined networking, so the first people that did that along with a combination of folks both internally, as well as externally to understand the principles of decoupled design and layer design. And so yeah, AT&T was an early leader in SDN and NFV, so using those lessons learned and exposing that up was a good basis to start with, coupled with some folks that we brought in from the outside that have a history of delivery in this space, coupled with the appropriate partnering with our hyperscaler friends. Rather than fighting them, you learn. They want to consume things a certain way-
Danielle Rios: [14:34] For sure.
Shawn Hakl: [14:34] … so they’ll tell you what they want. So, listening is a great opportunity as well to learn a few things. So, between those combinations of that background of the history and those spaces, bringing in some extra talent and then listening a little bit more carefully instead of trying to outsmart your partner is a good model.
Danielle Rios: [14:51] Yeah, it takes real discipline to do that because internally it’s so easy to look over into the next office and say, “Hey, buddy, give me a back door. This is a pain in my butt,” versus really building that interface right. You’re using it just like your customers will use it, and that’s how you make sure it’s a really great product and has all the interfaces that you need and exposing all the right things.
Shawn Hakl: [15:11] Yeah.
Danielle Rios: [15:11] So, super cool. That sounds awesome. Well, telcos aren’t known as risk-takers, but you’re a little bit of a risk-taker. I heard you were on the Canadian National Luge team. I’ve watched Luge recently at the Winter Olympics, and it looks terrifying. This is the open sled, feet first, 90 miles an hour down an ice track. And so what did Luge teach you about taking risk?
Shawn Hakl: [15:36] One, it was a lot of fun, and two key things I’ll take away from that. What people don’t realize is though you were down on the track very fast, chances are you’ve walked it about 100 times for every time you go down that track. So just knowing what’s coming, practice, mental discipline around having a plan and being able to execute on it. That’s the discipline rigid part. The other part that I learned from that is there’s an expression that they use in the community called, “Drive till you die,” which means as you’re shooting down the track, if you start steering like you’re going to crash, you’re going to crash. If you are flying down the track, and you’ve got the right plan, and you start to go off a little bit, just keep steering and acting like you’re going to make it, and nine times out of 10 you will. So, if you’re going to take risks, figure out your plan, stick to it and then hold onto it as you shoot down the track. There’s going to be wobbles, stick to the plan and the race, win.
Danielle Rios: [16:23] Keep going. Yeah, unbelievable. I wish telco execs, in general, would listen to this advice because it’s so risk averse. And there’s been so many times where people are ready to make a little bit of a risky decision and they default to the safe bet, and then they wonder why things never change in telco. And I’m like, “Because you guys never take risks.” But Shawn, this was a really great conversation learning what you guys are doing with your network, with AI, kind of how you’re thinking about it, and so thank you so much for coming onto the podcast.
Shawn Hakl: [16:51] Hey, I appreciate time. I love talking to you. So, great opportunity. Have an awesome day.
Danielle Rios: [16:56] Yay. Stick around, we end each podcast with a “Telco in 20” takeaway. I’ve got two minutes to tell you something you need to know. In 2002, Jeff Bezos sent a memo to every team at Amazon. It mandated that every product exposed their data and functionality through service interfaces. No shared databases, no back doors, no exceptions. And according to legend, anyone who didn’t do this would be fired. Fast forward to today and think about what Shawn just told us: network interfaces will be agent consumable. In order to do that, networks need to be discoverable, so agents can find what they need without being told. They need to be predictable, so agents get the same answer every time they ask. And they need to be programmatic, so agents never have to leave their own world.
[17:52] So, take a page out of the Bezos book, build a network with those three properties, and force your teams to consume it the same way agents will. No secret processes, no back doors, no exceptions. Are you ready to build an agent-consumable network? DM me on LinkedIn or X @TelcoDR, and let’s talk. In the meantime, tune into more “Telco in 20” episodes, like and follow, and leave us a five-star review. Don’t forget to sign up for my killer email newsletter on TelcoDR.com, and check out our awesome YouTube channel, and hit that subscribe button. Later, nerds.
