The need to get this right is at odds with the excitement to go fast. Stop Requested. This is Stop Requested. by ETA.

I’m Christian. And I’m Levi. These are real conversations with the innovators, operators, and advocates driving improvements in public transportation.

Today’s episode is a. little different. We’re kicking off a new AI roundtable series with ETA CEO John Maglio and Stephen Kuban, CEO of Kuban Transit Solutions.

Rather than focusing on one specific technology, we’re comparing notes on what we’re seeing and hearing from transit agencies right now. We’ get into practical use cases, policy, and workforce questions, the challenge of connecting agency data, and what AI could eventually mean for more proactive transit operations.

Here’s our first AI roundtable with John Maglio and Stephen Kuban. Welcome back to Stop Requested. Uh, today, we’ have a very, uh, exciting episode. This is the first AI series, uh, that we’re doing here at

Stop Requested, and, uh, the show style is gonna be a little different today. We’re gonna do it more as a roundtable, uh, and there’s gonna be four of us in this conversation. You’ll hear the different voices, and I’m one of the participants of this roundtable, Christian Londono from ETA

Transit, and joined by me, uh, are… Hi, I’m John Maglio. I’m the CEO of ETA. I’m Stephen Kuban. I’m the CEO at KTS, where we’ support transit agencies and the public sector with AI adoption and integration, workforce development, so really on the cutting edge of AI. Very excited to talk about, hear you, what you guys have been seeing. And I’m Levi McCallum, Director of Operations at ETA.

Uh, so, you know, Stephen, John, you guys are, are talking to agencies all the time, um, you know, in, in your, in your work. I know that, um, you know, agencies are increasingly interested in moving beyond, like, what is AI to how do we actually use this? What are you seeing?

What are you hearing? Um, I guess, like, I’ll go first. Uh, people wanna go beyond AI, but there’s still confusion, I think, in many cases, about, like, what AI actually is. Um, I, I do a lot of speaking at conferences and put my hand up,

“Have you heard of Claude or whatnot?” More and more hands are going up for that. Um, but as I ask, “Are you using it more than writing emails?” Still not seeing a lot of hands.

Um, there are folks who are starting to realize, “Whoa, I can do some pretty complex data, analysis of data I’ve never looked at.”

Uh, there are people also realizing, “Hey, I can use this to, like, make better strategic decisions,” but those are still the minority. Right now, it still feels, feels there’s just a lot of uncertainty about what this is. Um, that, that’s what I’ve been seeing. It is starting to change. More and more hands are going up, uh, but, uh, as someone on the frontier, I, who often thinks I’m behind, um, there’s, it’s still early days in our industry. That, that’s what

I’ve seen, but, like, John, you, you sit in a totally different seat than I do. Like, what’s your perspective been? Yeah, sure. I’ve spent a lot of time talking to, uh, executives at different transit agencies, um, and it’s different at different agencies, but, but as I talk to

CEOs, uh, the message I’m hearing is that they wanna leverage AI to solve real-world business problems. Um, so there’s been a lot of discussions about unifying data. Um, of course, as I talk to leaders in the IT space, for example, I was at APTA

Tech in St. Louis, um- Mm-hmm … a-and there’s a lot of discussion about, like, what should our AI policies be, right?

Um- Mm-hmm. And as you’re seeing, I, I think it’s still in its infancy, you know, uh, from an enterprise perspective, uh, but interest and, uh, excitement to see what’s possible i-is definitely on the certain surface. I’m getting a lot of policy questions, like a lot of policy questions.

It’s very transit, isn’t it, to start with the policy? For sure. What are the main concerns that you’re hearing? And the reason I say concern or maybe a, a challenge is because u-usually that’s the, uh, that’s the first lever that you pull, right? Is like, we need to have a policy around this and make sure that there are some guardrails for use or that it’s not abused, that sort of thing. Some, some, some people are just, like, afraid of getting hacked. I think that’s, like, one of the big things. Um, I, I like to see there’s, like, the desire lens and there’s the fear lens. Um, you can build a policy under the desire lens, like, you use these tools across your work to be more effective.

But a lot of the ones that I’ve seen are more desired lens. So it’s like, “Hey, you shouldn’t put anything into here. Don’t put any identifiable information. Don’t put the names of your colleagues. Don’t put any data in here.” Before you know it, the policy essentially says use AI, except you just can’t use it for anything useful. That’s a lot of policies I’ve seen and I, and I help coach through.

Like, there, there are ways to have a very effective policy, but, um, I, I think it starts with just understanding what is possible first and foremost, and what conditions do you need to make to realize that possibility safely and effectively.

Um, and, and so I’m starting to see more of a push now towards, like, adopting a team plan for Claude or for ChatGPT or whatnot.

Um, or of course, there’s always Copilot, but let’s not get into Copilot yet. John? It’s interesting. I’m seeing both sides. So I’ve talked to a number of agencies who are limited to using only Copilot.

Presumably, they’re in a Microsoft ecosystem. But interestingly- I thought I would hear more about concerns than I am. So I think we’re in this place of excitement. You know, we, we need to answer those questions, and those questions need to be asked. I’m not hearing them as often as I thought I would. Um, I think- That’s a good point … I think a lot of the people I’m talking to are so jazzed up about what could be possible. Uh, yeah, listen, there’s a real risk, right? That we adopt technology too quickly. We don’t do it safely.

Uh, we spend money on AI that doesn’t really solve real-world problems. Yeah, interestingly, I’m, I’m not getting as much, you know, pushback on, on what could go wrong as, as I expected to, and ultimately we need to. We have to answer those questions.

Hmm. Yeah. I’m… I, I see it’s like two, two sides right now. The, the leaders, like, a lot of the more innovative and, like, thoughtful leaders are really embracing these things, right? They’re like: “Whoa, look at what I can do when I’m using this. I can spin up this dashboard. I can, like, get my procurement out faster.” And they’re going, right, with or without a policy. Some of them are like: “Uh, I don’t have a policy, but, like, right now I’m still good to be doing this.” They’re so motivated. But then there’s the flip side of like: “Okay, now I want my organization to be doing this in the same way I’m doing it, but, like, how do I bring this out to my, to my staff, my day-to-day staff of schedulers, dispatchers, my bookkeepers to use this?” And that’s, like, a whole different type of organizational adoption because they’re not naturally sitting in the same chair as these leaders. And so, like, I’m seeing that, that tension between I know what

I’m capable of doing, but how do I safely bring the rest of my organization on this ride? What, what do you think, Christian? So, you know, having conversations with different folks, uh, also, um, you know, transportation planners, folks in different levels of, of the, uh, transit industry at different organizations, they’re using AI. And yes, some places have policies, and the policies, uh, that came out, or at least some that, that I know of, they came out, um, you know, a couple years ago when this started building pro- uh, popularity, where about don’t use it. Don’t have AI tell you how to do your job. Don’t ask them, like: “Hey, how should I, you know, make this decision?” And then follow what AI is telling you. Uh, the same rate, don’t be up- up- uploading information. But at the end of the day, people in some organizations are using AI without asking permission, or without even following- Mm-hmm … the policy. They’re just going online, finding their LM of their choice, and, and they’re getting work done. Some are finding a lot of efficiencies and, and they’re getting some of the analysis done through the AI and, and some of these leaders in organizations, they don’t even know. Like, it’s, it’s not organized, right? Like, think about your workforce. There are- Yeah … different people at different skill levels on how to use the technology. They’ve been teaching themselves by, you know, trial and error. There’s no, like, formal training where agencies are saying like: “Hey, let’s formalize our understanding of AI. Let me level everybody up, and let’s start using, you know, AI or some of these tools for A, B, C,

D. And this is the allowable, you know, um, things that you could do with a AI and, you know, and getting everybody level up, right? Like, embracing the technology. I see a lot of people just randomly using it, and they’re using it whichever way they think is gonna help their jobs and, and, and nothing formalized. Mm-hmm.

Well, like, we’re starting to do what we’ve been doing trainings to, to help with that, and, like, to formalize the training and the staff. It’s still early days. Like, we’ve just been starting that.

But, um, but that’s definitely the big thing with AI now is that it is a huge force accelerator for the individually motivated and, like, the ambitious to be using it. And you do start getting, like, divergence and, like, drift, and I’m gonna use ChatGPT and I’m gonna use Claude, I’m gonna use Copilot, and, like, no one’s really using it the same way. And, and that can be, that can be challenging for an organization, which I understand also leads to the reaction of we just have to put a policy in place. But hmm.

You know, I, I, I was thinking about this AI transformation in the public transit industry. A- a- and this is my question. I, I wanna hear your guys’s, uh, perspective, but, um, y- y- you know, in my opinion, it’s the public sector that is leading the transformation, right? We see different, uh, companies that are implementing AI and kinda like, uh, presenting versions or implementations of AI to the public transit industry.

Uh, do you think– Do you feel the same way that mostly it’s on the private sector, uh, not as much in the public sector? And do you have some maybe examples that any of you could share, you know, about how this is being used today? So you mean it’s more on the pub- the private side who’s driving it than the public, right?

This week. I mean, definitely. Def– I think there are just, like, different sets of incentives and motivations. Like, in the private side, you’re doing everything you can to find an edge, right? Like, can I get that proposal out faster?

Can I, uh… That’s, that, that’s natural. Um, that’s why… But, like, there’s so much productivity, and not just productivity, but outcomes, meaningful outcomes you can drive more effectively than ever. That’s where the public side is starting to recognize that, and when they’re mission-oriented, like really mission-oriented, then I see that change. Like: “Hey, I want to move more people more effectively.

Therefore, I should be using these tools ’cause I can do that.” That’s where I see the motivation come from, coming from, but it’s, like, an underlying mission as opposed to necessarily economic competitive productivity, and they operate on different timelines and forces. That’s my perspective, at least. So Steven, what do you think is working today, for agencies? You, you mentioned procurement on the, the private side. Are, are there any other areas where y- there’s a more concerted focus, uh, to have that application of AI? Oh, yeah.

Like anything operational, anything back-of-house operational, you can make so effective. One of my favorite stories is, um, NTD.

So you, you guys, I’m sure your ETA data often is used in NTD reporting, right? So, like, I worked with somebody who managed to build a NTD report for his agency in, like, a weekend, and, so ne-next year he’ll file it in twenty-five minutes instead of previously it would take three or four weeks pulling all these different data sets together, ’cause it’s a lot more than CAD/AVL, right? It’s tons of data sets. Um, another one was so cool, a small agency that struggled with no-shows on their paratransit, and every no-show is just a missed seat that it’s so expensive and they’re like, “We can’t afford to pay for IVR in our software.” And so they built, like, an Excel-based tool where you’ download the trip requests, drop into

Excel, and it, tells you, “Hey, if you call these six people, they’re the ones who are. most likely to no-show tomorrow.” So now their night dispatcher, they just call these six people. A bunch of them will cancel their trips, or they’ll say, “Oh, I, forgot. Thank you for reminding me. I’m gonna show up.” And so they’re seeing meaningful results in, like, minimizing their no-shows because they built a tool in-house instead of relying on something else from another vendor. And so, like, these ideas, they’re really limitless, these ideas. Like, I’ just like to think about it. What are you struggling with?

Where are you being, like, blocked? You can build something and solve that, and you can do that, like, in an. afternoon once you know how to use these tools. So it’s working across the board and from what I’m seeing.

Yeah, I’m seeing the same thing, guys. Um, we’re seeing early adopters taking initiative, um, and using these tools to improve, you know, uh, uh, efficiencies.

What we’re not seeing yet, a-and I’ don’t think this is unique to transit, is sort of the enterprise version of this, right? Mm-hmm. So instead of Bob wrote this tool over the weekend to create NTD reports, which is amazing, right? There’s a real product market fit there. This is a real-world problem we need to solve. It takes too long to do these things.

Operationalizing that, I, I think is, you know, what’s coming next. Um, a-and just to go back to your earlier question, what are we seeing in private industry? Same thing, right? Doing this as a, at an enterprise level i-is kind of a, a different animal than individual contributors using these tools, but I, I do think that’s where you start.

This episode is brought to you by ETA. For decades, transit agencies have been locked into legacy CAD/AVL systems built for another era. Expensive upgrades, rigid architectures, and software that lives in server rooms instead of the browser. ETA’s the modern replacement. Our web-based CAD/AVL delivers industrial-grade reliability with cloud-native speed and flexibility.

It integrates with what you already have, replaces closed ecosystems, and gives your team a single real-time operational view without starting over. Replace legacy.

Move forward. The future of CAD/AVL is open. See it at etatransit.com. Mm-hmm.

I, um, like, I think about, this as the Industrial Revolution. It’s a massive, massive change in how work gets done and how work is valued.

And I think about this as like a multi, multi-year change where it starts with, like, awareness of what’s possible, and then, like, the se– next step is transitioning. So, transition from the way you used to do it to a slightly faster way of how you do it. And then it’s transformation, which is why are, you even doing it in that, process in the first. place? You can rethink everything.

And transformation, I, think, John, is. what you’re getting at, that. that’s the enterprise organization. You need all of these different data sets connected. You have to know what your internal working process is, how to pull these data sets and do all of those things, but we’re still not there. Like, we’re at the very, very, very start of potential transformation. That’s like what we’re doing at KTS is figuring out how do we structure our data and build automations and workflows to use this. But we’re still just, like, transitioning our workflows, like have a conversation, write a proposal. That’s what these stories are. They’re all individual case studies of transitioning work that’s been painful. The next step then will be all enterprise transformation. But that’s a big question ’cause you gotta, you gotta figure out how to so– cut through these data silos and where that lives. And there’s a whole open question still being solved about data, where it lives and how it’s operationalized.

Working at a transit agency, I, I know how disparate each individual group can be. They have their own fiefdom. They have their own system. We have our way of doing things right, and, uh, oftentimes that is associated with the software that the department is using, whether it’s finance or it’s maintenance. You know, everyone has their own little specialty. John, how do you break down those silos and start thinking about the organization’s data rather than a department’s data?

Yeah, it’s a great question. Uh, and, and I, I, I do– W-with any technology system, it all starts with the data, right? Um, you know, it’s like building a house. You need a solid foundation.

A-and I do think there will be an inclination for each group to race to buy the product that will, you know, help make their lives easier, uh, which will only exasperate the problem that we’ve got all of these disconnected systems because for AI to really be put to work, it needs to look at a picture of the entire organization.

What do I mean by that? Hmm. So as a transit agency, I run, I don’t know, maybe a dozen different systems, all provided by different vendors, uh, that were never intended to work together, right? So if I have questions, uh, that, you know, I’m just looking for truth, you know, the way it works today is I’m gonna go to each individual system and ask, the question within the scope that that system understands.

Um, you know, AI on top of that system doesn’t solve that problem, right? So from my perspective, and, and I’ve been hearing a lot about this as well, right, in, in the AI subcommittee meeting, uh, at, at, at Tech and just, you know, uh, kind of talking to people. There is, there is, uh, I,

I, I think, uh, y- the, the observation that we, we’ve gotta get the data part right first. And, and to answer your question, how do we do that?

Um, you know, it’s not a trivial task, right? So… And A- and AI doesn’t make this magic, right? So, so the first thing I think about is, you know, w- what we shouldn’t do, right? We shouldn’t upload this data to ChatGPT, right? Particularly not HR and finance and things like that.

Now it’s out in the public domain, right? So agencies are gonna need some type of data store that, that pulls these data sets together. Um, a- and then the challenging work, you know, consuming it is fairly straightforward.

It is normalizing it in such a way that an AI engine tuned for that data can understand it. Um, and once you’ve got that figured out, well, what happens when the interfaces change? Vendor A just changed the name of a column, right?

Um, a- and, you know, that breaks. So, um, creating tools to maintain the system, monitor what’s happening, um, a- and in the best case scenario, we’re, we’re, we’re toying around with this now, is the AI agent finds the problem, creates a ticket to fix the problem, fixes the problem, and then adds it to the bill, right?

Um, because this does create a, a, a pretty hefty level of complexity, and I, I think the need to get this right is at odds with the excitement to go fast. What do you think is the, the primary reason that agencies shouldn’t be just using ChatGPT,

Gemini, Claude? Um, you know, they have it at their disposal. Some of those even have free accounts. Why not just upload it there? What, what’s the issue? Yeah, I’ll give you three answers. Uh, one is, if it’s not breaking policy today, I’m relatively certain there will be a policy in the future, uh, that, uh, that indicates that that’s a no-no.

Uh, number two, um, you know, security. Anything you put out there is in the public domain. Number three, you think about how these commercially available LLMs are built. They don’t understand your transit data. Let me give you an example. If I uploaded a bunch of data to

C- ChatGPT, it won’t know that we were on a detour for three hours last week on this route, right? So there’s context. And, you know, th- these different AI systems and machine learning systems require a group of data scientists trained in that domain to make sure these systems don’t hallucinate.

So let me put that on a bumper sticker. You know, if we, if we upload this stuff to the internet and ask questions, oftentimes we’re gonna get incorrect answers. A, a lot of that is changing. Like, a, a lot of that is changing pretty quickly. So, um, things in the public domain, like if you’re on enterprise plans for your, you know, enterprise whatnot, that’s not being trained on. That’s, like, yours and locked in. You’re, you’re right that there’s, like, a challenge in interpreting and studying the data, and the context is king, right? I liked your example of you can look at this data, but you don’t know that you were on a detour last week, right?

Um, and I think that’s, like, going to be solved by, by how we start connecting all these disparate systems and putting the human in the loop on these things. So if you were to use, say, an off-the-shelf team or enterprise version of Claude or ChatGPT, you could run, like, “Hey, here’s the data structure of my CAD/ABL data.”

It will flag, “You were really late on Route 7 last week, and you were missing your time stops.” And, you know, you chat with it and say, “Well, we were on detours.” And so it could, like, preserve and save that. But the question is, where are you storing that inherent tacit knowledge? Where are you gonna be storing that there were detours here or that there were other contextual things like so and so calling in sick?

Question is, where does the data all live to provide the context, and then so we can provide enough guidance to be interpreting and analyzing these datas? I, I see, like, the future of this as, like, AI almost as a copilot, which of course Microsoft has co-opted into their

Copilot, where, like, a human is driving, like, the decisions and questions we need to ask. And as long as you know where the different data lives, you can pull, pull, pull it all together, analyze it, and make decisions off of it.

But it’s not this, like, silver bullet. AI is not the silver bullet that’s just gonna make everything go away on the reporting side. It’s something that makes it faster, but the human is, like, the driver in the end. Uh, Stephen, in, in your example, you’re describing the, the human that’s prompting, uh, you know, to be able to get the information to them. W- is there a world in which I’m no longer prompting, or prompting is maybe a sort of last step or a last resort that

I’m notified of- Yeah … the- these things that are happening in advance? Yeah, it’s agentic stuff. Like, we’ve got agentic stuff in our organization. Like, we have an agent that automatically every day looks and scours to see if there are any RFPs out there, and if it finds something with an appropriate title, it knows how to log in because it has Like our company One Password, it can log in, it can find it, it can download the bid documents, it can process the bid documents, summarize it, and then put it into our system of record that says, surfaces in an email,

“Hey, there’s an RFP. You should consider this.” So agentic things are absolutely a huge, huge thing. I see agents on two sides. A, there’s, there’s agents on the looking back, like what has happened and is not time sensitive. Those are things that I think frontier models are very capable of today. But then you have a- agents in the moment.

This is happening right this instance, and this is where operational technology lives. That frontier models I think will always struggle to be good at because you need real-time data feeds, you need the humans who know, like what it means when a driver blows past a time stop in real time and how to respond to that. I think that agentic stuff will always live with the, with the software itself for that in the moment. Yeah, I, I would agree. And, and to carry that example further, you know, um, with the right data structures, these systems should ultimately be able to make predictions, right?

Mm-hmm. So think about this. We- we’ve seen hedge funds that use, you know, uh, quant technology to figure out what a stock price is going to do tomorrow, right? If we have the data structures in place, and we have a good architecture, it’s reasonable to think that there is a future where the system can raise its hand and say, “Hey, something doesn’t look right, and this is what I’m predicting is about to happen,” to allow a human to go take action, and then, Steven, maybe in the future, you know, a human just approves the action, right? Mm-hmm.

Mm-hmm. O- of which, like a decent amount of that’s kind of already happening, um, or is possible. It’s not mature yet, but it’s still very green. I wouldn’t expect transit as an industry to be adopting that yet. I think like those types of actions are…

It’s, uh, um, y- you– That, that’ll– that will not be driven by like the IT analyst who is like an innovator using Claude. I think that’s gonna be driven by the software vendors. And those who choose to adopt AI and make it central to their operation,

I think those are the ones who are going to have like a very, very bright future. Yeah. I’m glad you bring that up, right? Because, you know, th- there’s a lot of excitement out there, and I think your point is that, um, e- even if some of these things are possible, um, we, we should, we should take our time going through this, right? One step at a time, um, and not swing for the fences because, uh, some of the tech isn’t ready. A lot of the architecture isn’t ready.

Um, and, and, and I think if we start incorporating, you know, th- these tools into our workflows a little bit at a time, we start mastering them, and then the, the final product is, is just gonna be much more stable.

Mm-hmm. Mm-hmm. I think that having like a good flexible data structure and data architecture is, that’s the key to success for like the software vendors these days for the operational side, um, for operations to thrive.

I’ve heard of different vendors who like it’s, you can’t even get a proper report out of it. Like, if you can’t get a proper report from something that happened last week, how could you make any s- smart decisions about what’s happening right now? And so ironically, you know, the whole Tides push for open data structures and data standards and whatnot, those who are adopting something like that are starting to have to define what are the core axioms of an operation. I think anyone who’s adopting that is gonna be in a good place to architecturally shift. But many, I think, just won’t. And in the static world, you know, the world up to twenty twenty, that’s fine. You’re locked in. But as more and more agencies are realizing, “Oh, I can build a custom dashboard of what happened last week, like in real time, and I can change it in five minutes?” As more and more are doing that, they’re gonna say, “Why can’t I do that right now?

Give that to my dispatcher, so you can manage and get the predictive things right now.” And there is gonna be a massive demand for technologies that can support that.

Yeah. Let me ask you this question. I’ve been thinking about this. As you know, we’re huge supporters of standards. Um, I mean, just, just take my business for example. We’re developing software at five X velocity from a year ago, right?

Hmm. Um, standards are amazing, but are they gonna be able to keep up with the pace of innovation in an AI world?

I don’t see how. No- nothing is standard anymore, John. Right? It’s a real problem to solve, and

I don’t have the answer, right? But, but yeah. I mean, you know, you’ve got committees and meetings and approvals and, um, you know, we’re ha- we’re having trouble keeping up now. I think where, where standards are safest is in defining an existing operation. So where an operation itself is well known, that’s where GTFS comes in. It’s a fixed route with time stops, right?

You can define that operation, so you can have a standard that says, “This is what we’re doing.” Um, the latest one, of course, was on-demand transit. That’s the world I came from. There was a lot of struggle in the back offices of different companies there to define what exactly is on-demand transit,

’cause everyone kind of had their own definition of it. I think that’s why GTFS Flex has struggled to fully get adopted, and what about GOFS? There are all these, all these things there. For the most part, that’s getting sorted out. Now I’m thinking like I’m not seeing as much innovation in the mode of operation per se, like fixed routes, deviated routes, on-demand transit. Those are kind of known. So I could see how a data standard could be developed to define how you operate something on the road, so that just defines what it is. But in terms of how you operate in the back office, that’s, I think, the world that we’re in now, which is gonna move a million miles an hour. So things like, um, was it TODS or TIDES? I, I don’t remember exactly. The one that is operational data standards,

I think that one’s gonna be a lot, lot harder to get any consensus around data standards because now how the back office operates, that’s where all the innovation is happening. Just think about how many data sources aren’t captured by standards at all. I don’t see standards for maintenance data, finance data, HR, workforce management. Um- Nope … it’s a lot of customer facing, and it was a great place to start.

But, um, yeah, no, I, I, I think, uh, having conversations like this with, with other industry leaders, um, you know, it, it, it’s gonna be incumbent on us to figure that stuff out because you’re right, agencies are going to expect this. A-and their constituents are going to expect the efficiencies that it’ll create. Of course. Of course.

Like, I think about it, the ridership per public dollar spent is almost, like, the best way to view it, right? And that’s not operational cost per passenger, that’s everything that goes into the back-end operation. The whole size of your overhead, your team, whatnot. Um, just with the pressures, especially in the US, around funding for public transit, you have to drive that down, and AI is gon- one of the ways to really, really make huge strides there. So the public’s gonna demand that you do that, and that demands that the data will talk to each other. That demands that your workforce is going to be effectively using it. That demands that the skills around judgment are developed enough that you can make good calls using this data. There will be demands for the operational software to be keeping up as well.

Like… Yeah. Christian, do you have a take on this? Yeah. It, it, it, you know, just, just also, you know, as it relates to the standards and, and, and, um, where, uh, AI has been, where it’s at so far and what it seems it’s heading to and how rapidly it’s, it’s changing, right? Like, when you talk about the standards and, and all this type of things, you know, the, the AI capabilities a year or two years ago, and the capabilities today have changed exponentially.

There’s… A-and, and when you’re reflecting about how you could use this technology, you know, now the opportunities are expanding. And, and I think that as it’s heading into the future, of course the, the opportunities are only gonna keep expanding, right? And, and it’s not just that it can do tasks quickly or that you can feed it some data. It’s, it’s also the intelligence that we have never had access to, right? Because certain things, i-i-in terms, in, you know, you were talking about, like, the predictability.

Systems that are telling you what’s about to happen or, you know, the example you were, uh, explaining earlier about, um, you know, the, the no-shows and, like, giving you specific people that are maybe historically trending as no-shows, and it’s like, all these people, and some of those could be maybe somebody that has Alzheimer’s or something like that, and there’s a reason why they end up being no-shows.

But it’s giving you that intelligence that otherwise we didn’t have the capacity to extract from all the data in our systems. And, and, you know, I think that that’s really cool. But like, a, you know, to your point of the standards, at the rate this is developing and changing, it’s just very hard to be able to say, “Okay, wait, wait a minute. Let’s get organized.

Let’s create the standards and, and, you know, get everything in a way that everybody can understand it, and then we move forward,” right? Like, things are moving at a rate that is not waiting for that. So, you know, what, what I wanted to ask you, um, you know, as, as professionals in this industry, in terms of the growth of capabilities and where it seems AI is heading, you know, could, could you, you know, just throw some ideas at where you think this might be going, especially in that realm of the predictability?

What, what some of these systems do for agencies, uh, as we, you know, to the future. I go back to, to predictability, right? I think this gets really interesting. And listen, it’s cool enough if I could get a single source of truth and ask my operation questions, um, that, you know, are really difficult to get to today, if not impossible. Because all that information is spread out between different systems, um, a-and oftentimes won’t agree, right? So with that holistic view, right, if I could just have a conversation with my system as if

I was, you know, uh, just another human being. Um, and not only get data, that’s what reports and dashboards do, but the insights that you’re talking about, Christian, right? This is what LLMs do well. I, I think that’s exciting enough, but where I get really jazzed up is thinking about how we could free up human resources to do more interesting things in transit agencies to improve service delivery and provide great customer service and improve safety records and things like that. It, it’s where these systems can start to be predictive, use agentic actions to take some of this, uh, more menial work, you know, off the hands. of, of our users. I mean, imagine yourself as a dispatcher with, with all the phones ringing, you know, everything going on, right? Even the best systems today will do a pretty good job of telling you what happened. But I, I think the future is that these systems will, you know, kind of monitor the situation in real time. Um, you know, raise an alert that, “Hey, we see a situation brewing,” a-and then give you actions that, that you can take a-and then eventually say, “Hey, I think I’ll take those actions for you. Would you like, me to?” That, that’s what

I get excited about. Mm-hmm. And, and I see that, like, the, again, the delineation I mentioned earlier between, like, the operational vertical and then the whole business, which has all those things. So, like, the operational vertical of I see what- You need what action you need to take because this thing’s on fire right now, and you solve your dispatchers.

Like, oh my gosh, what an amazing tool that, can be developed within an organiz- within a software to do that, whether it be CAD/AVL, on-demand transit, maintenance, um, just things that’ are happening in real time. That’s, like, so exciting.

And then, and then you have, like, the organization as a whole because in a business, let’s say in a transit agency, CAD/AVL is one thing, paratransit is another thing, finance and payroll is another thing, like, grant management is another thing. They’re all connected, but, um, no one tool can ne- can report on the business as a whole because a business is a collection of tools.

So how can any one thing know the context of everything? And that’s where I’ see this opportunity for this, like, operating system, let’s call it, in, like, a, a transit space can, can be emerging. And, like, we’re building one at KTS for our internal operations, not a transit one, right? It’s like, where– how do business decisions get made? What information do you need from each of these different operational verticals to enable that decision? And your process of making those decisions will be unique. Every agency will be unique. So I think, like, where this operating system comes into place is, like, providing that platform that holds everything together so that the organization as a whole, the, the transit agency as a whole, can make decisions and move forward. And that’s bigger than any one software. That’s, like, an entirely different thing. I think that’s em- that’s emerging. Yeah. You know, I’d probably go a, a slightly different direction and say that, um, something I’m looking forward to, I don’t know if I can necessarily predict it, but, it’s, it’s the workforce side. Um- Mm-hmm … you, know, how, how people start to, uh, start to think about AI as a, as a partner, as a, as a, you know, a thinking partner, some-something to help, you know, brainstorm a new service change or, uh, you know, perhaps it’s, um, you know, being able to remember things about the organization that may have been? from generations past. We’ve got a lot of folks who are retiring. Uh, you know, maybe there’s some attrition at the agency and, you know, you’re, you’re down a few people.

Uh, well, who, who knew that? Well, it was, you know, Greg, and Greg left. It’s like, ah, man, it would be nice to, have his brain.

Well, I mean, h-here you go. You, you have some way to be able to record what the thinking was for the agency at the time, um, a-and then kind of make sense of where you ended up, right? It’s like, uh, uh, there are so many things that, uh, when I was at V/Line, I’m like: Well, you know, why, why is that stop there? Why don’t we move that stop to here, and then we close this stop? Well, you know, there, there’s obviously a reason for it, but,

I was in the dark, and I think that, um, an LLM or AI, you know, generally as a sort of second brain can help give some light to some of those areas that, you know, previously is just like, “Oh, that’s just the way that it is,” and you shrug your shoulders. Yeah. And to your point, Levi, about workforce development, uh, we have to develop standards and training so that if these things become available, you know, our, our staff not only knows how to use them, but is comfortable using them. Yeah, a hundred percent. I mean, if you’re just, you know, throwing a, a Claude subscription at your staff and saying, “Good luck,” uh, y-you’re probably not going to get the results that you’re looking for. I mean, there, there’ has to be, um, you know, some onboarding and some, uh, some guidance, you know, maybe a bit of formal education around, like, how to prompt the system to be able to get the, the answers that you’re hoping to get, um, un-until everything is agentic, right? Yeah. I, I… You know, that’s, that’s a great point. And, and, you know, you have to train the, the workforce, right? Uh, it, it, you know, like you mentioned earlier, this is an industrial revolution right now, which is, uh, you know, heavy. Any industrial revolution is heavy in, in disruption, right? And, and it’s thinking about how organizations and leaders are ready for that disruption management. But then the, the opportunity is not just on, uh, changing the policy and allowing people to use Claude, and maybe having a policy that says like, “Oh, you know, uh, we’re gonna have, you know, Copilot, and, and this is some of the things that we’re gonna allow people to use.” I, I think that is progress, and it’s happening in some of the organizations, but, um, you know, I think it was you, Levi, who mentioned it. Employees in different areas of the organization, uh, they use that one system, and they kinda have some expertise and intelligence that is only in that silo.

So you go back to those silos, right? Like you, you know, you have an expert in the transportation planning and operations and maybe in HR and finance.

You know, HR dealing, of course, with that knowledge management issue. Historically, there’s no transition. People leave, the intelligence leaves. So it’s, it’s how can you, um, combine of that intelligence of a transit agency?

So today, everybody has that silo intelligence in the one system and the expertise of their employees, but you don’t have the intelligence that comes from combining intelligence from different professionals in different systems, right? Because it– today, agencies, are they looking at all the data. combined and see the implications, right? I– Do I look just at the maintenance data, then, you know, I know how we’re performing data-wise, and maybe our roll calls are not as high. But what’s the impact to the dollar sign? And, like, some of the other things that maybe if I’m in maintenance, I’m only seeing, you know, like, the intelligence of my maintenance system. So I think that’s what is, missing in, in being able to combine the different data sources is, uh, what I think has to be, you’ know, the the minds of these leaders and, transit, agencies is how do I combine that intelligence and, and not only allow my employees to know how to prompt AI and use Copilot, but Copilot is now gonna have the intelligence of my agency. So how do I get to that level? Oh, I-I’ve certainly thought a lot about that

’cause like I’ve been work-working with transit agencies to figure this out. And, um, the way I see it, let’s go back to what we’re talking about, transition, transformation.

Um, it, has to start with transition and individuals. So the ultimate goal, yes, is we have a shared company brain that everything runs through. Like, that’s what we’re building internally at KTS, right? It’s not easy to get there. But it starts just with can every person in your organization use these tools to do their job more effectively?

So I like to think about the individual journey and the organizational journey. The organizational journey is everything you’ just described, Christian, but that’ you– an organization cannot change until the, individuals go through their own journey, and that’ is one of fear, of discomfort, of insecurity in many cases. Like, if my value is no longer how good I write an Excel formula, what is my value? That’s an individual journey everyone goes through. So when we approach our training, we always, we, we don’t even touch the organizational stuff in our first round of certifications.

We just talk about making individuals into practitioners with AI. And so there is a light that happens where you, as your, in your own domain, your own world, you’re gonna find that pain that you had, and suddenly that pain is gonna go away. And that’s different for everybody, right? For some people, that’s filing

NTD. For some, it’s writing a good email to management. For some, it’s just how long it’ takes them to build a schedule by hand. Everybody’s pain is different. So when we approach our training, we talk about the individual journey first. Often, that even starts with stuff outside of work. I’ve seen people go, and they use Gemini to build a coloring book for their kids before a Disney trip. They find their individual value, and that moment where that something goes off in their head where they say, “Oh, this could be really, really good.” So the practitioner starts with helping p-people, people to find that, and then we always suggest, and w- you know, we always push for essentially capstone building.

So let’s, as a collective, as an organization, let’s try and do this project that has been on our backlog for so long. For some, people, that’s, build a new dashboard. For others, it’s like, the no-show tool or the NTD. It’s some project I’ve wanted to accomplish.

Now that we’ve found our individual journey, like we’ve got our own light, let’s try and solve a company problem with this. And that then starts to create the whole momentum for the organization as a whole because individuals are starting to do something really cool, and the organization is benefiting.

And this one, we, we call this the, uh, the, the launchpad, right? It’s a certification for practitioners. It takes three months, even up to four months for individuals to go through that journey, but then you’ll have the conditions in which individuals know how to accelerate their work in their own individual personal way. Well, now the next step that we have is the manager and the executive transitioning, which says, as an organization, as a manager, now how do you manage your whole team who are doing all this work in different ways? How do you start using cohesive skills and effective governance and, like, workflows? How do you start that transformation journey only once the people are, like, on their individual journey? And then you have the executive side over top of everything, which is IT policy and how you start doing future procurements and how you store data and where should you store your company brain and all of those things. But in my experience, it starts with individuals, that scheduler, that thirty-year long maintenance manager.

Let them find their win and their spark. Yeah. Listen, I, it’s funny. I tend to think at the enterprise level, but, I mean, there’ are folks out there’ that, like, you said, haven’t even used this to write an email yet, right?

Um, I-I’m curious, uh, you know, what, what’s your advice to, to managers to help get them off the fence, so to speak? Like, to get them to, you know, dip their toes in the water and try something, right? Just to get folks comfortable with this technology. Um, yeah. So, like, on– We just released a new website a couple weeks ago that, that shows this, like the three steps, right?

Is, um, you gotta see it, to believe it. You gotta see it first. You gotta know it’s possible. Then you gotta believe that’ you’re capable of doing it, and then you have to succeed in doing this thing. So I think often managers might come say, “You guys have to use AI.” But, as a, a staff person, I, don’t even know what I would use it for. I get that all the time. Like, what do I even use this for?

Like… So my suggestion, my advice, is to work with the individuals in their own workflow and identify where is your individual pain. So if you’re working with a maintenance manager, they’re like, “Ugh, I hate having to write my email at the end of the day.” Well, work with them and say, “Hey, what happens if we could do this and get it done for you in five minutes instead of thirty?” When they see that, then they believe they can do it because they can apply it. So

I say it starts with it, like show them a use case, give them the capability, and say, “Look how easy it is for you to do this use case,” and just provide that culture that allows you to experiment and to play with it. Um, then everything will flow from there. That’s where

I would start, John. I, I wanna go around the room and, and ask, i-i… what is that first step that a transit agency could take to make some, you know, meaningful progress towards, uh, you know, LLM adoption or, you know, maybe more generally, uh, with, uh, thinking about incorporating AI into the organization?

If, you know, a transit leader, a CEO, an executive director could just snap their fingers, what’s that first thing that you think they should do? You know, I think it- it’s different for different parts of the organization. I, I love what Steven just said, right?

Um, we just gotta get folks comfortable, you know, get using it, and, uh, you know, see the benefit of it.

Um, I think at an organizational level, I, I think it starts with hand wave the how for now, a-and, and start talking about the what. What, what questions do I wanna ask the system? What processes are taking me too long and would benefit from some type of automation?

Um, rather than jumping into AI for AI’s sake, right? W- you know, s- kinda start with the, the, the, the result in mind first.

Um, a-and, you know, Steven, I hear you. It, it, it’s good if we start bottoms up and get people using it. However, I, I don’t know if

I would do it serially. I, I think at different parts of the organization, there are big problems to, challenges to overcome, whether they be cybersecurity, data engineering, um, you know, access to different, you know, parts of the data a-and different tools.

Um, I don’t know. I’d probably have a, a different track. I, you know, I think getting individuals comfortable with the technology, getting them, you know, excited about what’s possible, a-and then thinking with the end in mind. You know, what do we hope to accomplish here, while starting to put together, you know, policies that help us overcome, uh, challenges that we know we’re gonna have. It’s pretty similar to kinda what I was saying. Like, it might sound like I’m saying get the individuals using AI.

No. First step, if you’re a leader, you start using AI. Yeah. You start.

Like, find how does it work for you as a leader. AI adoption is not a staff problem, it’s a leadership. It’s the biggest challenge facing leadership. As a leader, as a CEO, I don’t care the size of your organization, it is your responsibility to recognize what’s possible and figure out how your life can be changed by using this. And only once you have gone on your individual journey and you have that awareness, will the rest of the organization be able to follow.

Let me ask you this, Steven. Assuming there are gonna be some people that are slower to respond to this, or maybe even resistant, you know, what does that future look like for that type of leader? Transit’s in kind of a spiral right now in terms of we’re not getting enough money, so we can’t serve as many people, so we have to cut funds, so we’re not getting as many… We’re in a spiral. You’re seeing it across, like, coast to coast.

AI is the chance to break out of that spiral, because you can do a lot more with a lot less. Those who refuse to adopt it will continue in, in this spiral, period. And only those who adopt it are going to find that they actually already hold the capability to break out of it. A-and Christian, I just wanna make sure that, uh, make it around the, the table to, uh, you know, for your answer. If a, a leader could, you know, snap their fingers and, uh, have some step towards progress, what does that look like? What, what would they do, in your mind? Yeah, I, I think, um, education is, is first. So 100% I agree with Steven in terms of you have to start using it yourself.

Uh, I’ve spoken with, you know, CEOs of different transit agencies and, and they share how they’ve used AI themselves and how impressed they’ve been with some of the things that it’s doing, uh, for them and, and their colleagues. But I, I think the first step, uh, as a leader would be education, right? If people are afraid of what they don’t know, uh, people don’t understand where it could be harmful and where it could be beneficial.

So I would say training for the staff, especially those administrative staff that work in front of their computer, spending a lot of their hours in front of the computer, and also sharing with one another what is working. Like, where we’re finding efficiencies.

Because, you know- Yeah … I think, uh, all together as a, as a professional, I’ve learned a lot of the tools that I use and my shortcuts in Excel and this and that throughout my career from colleagues. Like, “Hey, use this command, you can do this and that,” and it’s like, “Oh, wow, thank you.” And I still use some of those today. So I would say, uh, first, you know, education. Uh, make sure that you’re leveling everybody up, uh, on the knowledge of AI.

And then second, just start by sharing how, how people are using it and how they’re finding efficiencies. I think that, that would be a good, uh, starting point. Well, this has been a wonderful conversation, uh, and thanks to our, our special guests here today.

Hope you join us for a second episode that we’re gonna have coming up on cybersecurity. It’s really been a blast. Thank you all very much, and thank you to our listeners for tuning in. We’ll be back next Monday with another episode of Stop

Requesting.