Ik bouw een digitale PM-AI-agent, of ik het nu leuk vind of niet

Galen Low: Damned if you do and damned if you don't. That's been the general sentiment from project managers I've been talking to who are adopting AI into their work. But while it's easy to think of it as building our replacements and sending ourselves into involuntary early retirement, there could be a massive upside to it all.

In fact, it might give us a seat at the table as the walls between disciplines come crashing down and cross-functional collaboration gets a reset. To unpack that, I've brought in a technical program manager in the SaaS space who has started building AI agents mostly out of necessity, as she single-handedly leads dozens of programs across a team of 130 programmers, data analysts, and other technical specialists, while also servicing over 1,000 internal stakeholders.

She's going to be sharing her approach to building with AI, walking us through the tech stack she used to build an intake and triage agent that is already saving 15 hours a month, and giving her take on how pushing through our latest existential crisis might lead us to better collaboration, camaraderie, and community within our cross-functional teams and beyond.

Hope you enjoy the episode.

Welcome to the Digital Project Manager Podcast—the show that helps delivery leaders work smarter, deliver smoother, and lead their teams with confidence in the age of AI. I'm Galen, and every week we dive into real-world strategies, emerging trends, proven frameworks, and the occasional war story from the project front lines. Whether you're steering massive transformation projects, wrangling AI workflows, or just trying to keep the chaos under control, you're in the right place. Let's get into it.

Okay today, we're talking about this conundrum of building AI agents to do the project management work we used to do. We'll be digging into the process of building AI-powered project management tools and agents. We'll be discussing the implications that these agents have on human-led technical project teams, and we'll be tackling a few somewhat existential questions around the project management career path and how we can keep ourselves sane.

With me today is Alla Tarasenko, Principal Technical Program Manager at Gusto. Alla is a human-centric tech nerd with experience working laterally across domains like data and analytics, infrastructure, machine learning, security, and compliance. Notably, she's also run major programs and has overseen delivery operations for over thirteen years, working in the SaaS space with companies like Smartsheet, NerdWallet, and Gusto to roll out brand-new products, orchestrate company-wide platform migrations, build teams and workflows from the ground up, and now innovate around how technical program managers can use AI to increase their capacity without losing the parts of the job that they love.

She's a champion of helping cross-functional tech teams work better together, and as the title of this episode suggests, she's beginning to build AI agents and embedding them in her operation, whether she likes it or not.

Alla, thanks so much for being here with me today.

Alla Tarasenko: Thank you, Galen. Very grateful to be invited, very excited.

Galen Low: It's so cool to have you on the podcast. You and I, we have been crossing paths on LinkedIn for the better part of a year, and I'm glad we finally managed to connect. I was like, "Oh, my gosh, some of what you're doing is exactly what some of the folks who listen to this podcast need to hear." Namely, it's just the nitty-gritty realities of trying to increase project management team capacity by leaning on AI, building agents, and all of the sort of philosophical questions that come along with that.

I'm really excited to dive in, and honestly, I know that we can probably find our way down a whole bunch of interesting rabbit holes as we go. But just in case, here's the roadmap that I've sketched out for us today. So to start us off, I just wanted to set the stage by hitting you with a big, hairy question that my listeners want your take on.

But then I'd like to zoom out from that and maybe talk about three things. Firstly, I wanted to talk about the challenges you were facing that led you to dive headlong into AI and why you've approached the solution the way you have. Then I'd like to just lift the lid on your approach to building an AI project coordinator, not just the stack, but also the process, the setbacks, and maybe the resistance from peers or even your own inner monologue.

And lastly, I'd just like to get your take on how you envision your teams working with AI project coordinators in the near future, what it means for TPMs like yourself, and what needs to happen to get there. Sounds ambitious, but how does that sound to you?

Alla Tarasenko: Yeah, that's a lot. Yes, let's do it.

Galen Low: All right.

Awesome. Let's dive in. I thought I'd just start off with one big, hairy question. You've been diligently building a set of AI agents and tools aimed at helping engineering leads run their own projects while also keeping you organized. But from what I know of you, you're someone who is very passionate about technical program management, and you're someone who's very human-centric in your approach when you practice your craft.

So my big question is this: By building ourselves a team of agentic project managers, aren't we working ourselves out of a job?

Alla Tarasenko: Yeah, I love that question, and it's... Like you said, it's such a big, hairy one, and we can go in so many different directions with it. And before I answer, I just wanted to acknowledge, I know we chat about this a little bit in preparation for this, but whenever you say building agents, I just have this "Ah, what is he saying?

Is it about me?" Be- there's a giant imposter syndrome happening because I think a lot of us who are non-engineers and getting into this area within the last month or even a year, are feeling very much like students and learners and people who are in, in a space that's fairly new for them and where every day there is a new discovery.

So it's hard to see yourself in that light yet. And I wanted to bring that up and acknowledge that because I think with all the posturing online, a lot more people feel like me than those who say, "I'm, like, a king of all AI," and you're not alone in that. I think for working us out of our jobs, who knows what's gonna happen, right?

There's... Just wanna acknowledge there is a lot of predictions, there is a lot of if you do this, you're gonna be okay, if you do this, you're gonna be left in the dust of the beautiful future. We can make guesses. I can't predict it, but I do believe that at this conjecture, like you said, whether she likes it or not, that is a tool that we'll work with, and I do not believe that my job exists outside of AI anymore, not in the space that I'm at.

That said, I think my particular role and where I work and the type of companies I work with are very good examples of where I could actually use all the help I can get. So I'll explain what I mean. For the last about 10 years, I've mainly been working middle-sized companies, so let's say 500 to 3,000 people or anything in between, you know, growing from one to the other.

And the trend this whole time that I observed anyway in my career has been to reduce the number of operational and TPM-type people within those companies and have them only cover kind of high-level operational flows, process architecture, running large programs. And in my current job, I'm a single TPM in a team of 130 people Another job I was at, the ratio was, like, more like one to a thousand.

So this kind of existential, are we not needed anymore? What are we even doing? I've been going through this alongside a lot of other TPMs, I suspect- ... for way longer than we've been talking about AI. And when you are a TPM supporting a team of 130 people, there is so much that you wanna be able to do, and there are so many flows and people that you wanna be able to help.

You wanna coach, you wanna support different projects, especially when you see newcomer leads running those projects, sometimes for the first time in their lives. You want to improve processes. You wanna get better at how we measure our operations. There is so much you wanna do, and you have to be very picky with what you focus on.

And being able to at least take the tasks that are routine and repetitive and tend to occupy a big part of our day and outsource it to AI, that genuinely, I know that that's often the ad for AI "Oh, if you do AI, you're gonna do meaningful stuff instead." For me, that actually is true because of kind of the wild ratio of support that I and many other TPMs in middle-sized companies operate in.

So yes, that's very much my motivation, not just staying current within this tech space that we're in right now, but also to be able to focus on the stuff I love, which is process architecture and coaching the teams and doing discoveries and not having to do something like physically move tickets from one Jira board to another every week so that we can triage them, which I used to do just a month ago.

Galen Low: I like the framing of the fact that, okay, yes, we're gonna build with AI to extend our capacity, and it's gonna change our jobs, and, you know, who knows what's gonna happen with our sort of sustained relevance as this role? But certainly not doing that seems like a bigger risk in terms of staying relevant, in terms of job security, in terms of the future of the role.

The answer's probably not to just avoid AI. And don't get me wrong, I think folks listening, and generally I think people understand that, but I think you're right about the posturing. I think you're right about the sort of pressure to do these things whether we like it or not. We feel like we're being forced to.

It does seem you know, "Oh, if I am not there to move the Jira tickets then I'm out of a job. What am I gonna do?" But I think you just gave a whole bunch of examples right there of what we do instead, right? What will you do instead of copying meeting notes from one thing to another? And what I really love about A...

I mean, my jaw dropped when you said the ratios because I'm an agency guy. Maybe we're running four to 10 projects varying in complexity, but there's 12 of us in an organization of 100 or more. You know? It was like most projects had a project manager. But when you look at it at the scale of a team of 1,000 running multiple projects and you're the only one or a team of 130 running multiple projects and you're the only one, your problem set becomes different.

You're thinking in terms of operational efficiency and scale with a human lens. But, you know, the answer to the question, how are we gonna be able to do more, is not necessarily, okay, hire more people so that every 10 projects has a project manager or a program manager but to say, "Okay, well, how can we build processes?

How can we design them? How can we empower people who will be leading projects who might not be program managers by training? And how can we Set ourselves up to work together, you actually become more of a mentor and operator. You need that mindset rather than a, this collection of, you know, a dozen projects are my baby, and that's what I do.

It's quite a bit different. I know some folks listening might be like, "Duh, Galen, that's obvious." But coming from my background, where it's oh, you know, every complex project we're delivering for a client should have, you know, a very senior project manager who can interface with a client and keep everything on the rails and you know, manage the project's own process.

It's very different than the operational mindset. I think that's really interesting.

Alla Tarasenko: Yeah. And I think the key thing is that the work still needs to happen. That's something that I just always come back to when I hear people say, "Oh, we don't need project management because that," or the connection, the glue, the whatever you include into actually getting the work done across a variety of skill sets, teams, and points of view, it still needs to happen.

The question was, like, who does it, right? And what combination of people and tools end up doing that? And if I can provide some support in the sense of this is how you can see where the projects are at, and this is how you kick off a project. Here, just a simple wizard for you, which spits out, you know, whatever artifacts we need to be able to keep tracking it.

And this is some best practices that are presented as an actual flow. This will give people some kind of support and confidence. And to me, the hope is that, you know, that I'm sure I'm not the only one. Like, when you look at your day and you think, "Oh, I'm gonna do this big cool thing," build something or meet with this person and build this strategy.

And instead, you're just buried in a gazillion little things that if somebody asked you what you were doing all day, you wouldn't even be able to describe even less so in the sense of any kind of impact or outcome. And reducing that is what I think I'm trying to wrap my mind around, and that's how I'm trying to pick the use cases to focus on.

Galen Low: I like that. You know, even to zoom out a little bit, you are leading a technical program management, I guess, department of one at Gusto. You guys make HR software, but as I understand, you guys are very AI forward. You're very data-focused. It is a very technical team that's building out these products.

It's reflected in your title, right? Principal Technical Program Manager. I'm getting your title wrong. I'm so sorry. But I'm curious with the sort of ratio, and maybe, maybe the answer's obvious. But at what point was it clear to you that you needed to build something with AI, like an agentic project coordinator, in order for operations to continue the way you wanted them to?

Alla Tarasenko: I'll be honest with you, and I think that kind of derives into maybe a different topic a little bit, but I think we, we chatted a bit before about, like, how does your mindset... If you're somebody who Goes in your problem solving towards bringing people together and clear communication and making sure you have the right structure and framework and all of that good stuff.

Shifting your mind to, "Oh, I can automate this," or, "Is this something I can create?" Is actually not a straightforward thing. And again, being vulnerable saying it here because I probably should not admit it. I think it wasn't like this big, large, yeah, I really need help, so maybe the help should come from AI.

It was really more around specific pain points. For me, the big specific pain point was around the intake and triage. It was very... When I joined the team, we did not have a structured intake and triage. Was a bit ad hoc. Each team handled it differently. So put together more centralized flow through the channel, through Jira, having twice a week triages.

Quite manual, you know, pulling the tickets, discussing them, updating them in Jira. Took ridiculous amount of time. But it was an important process to switch to so that we have that centralized thinking. But it was so painful that, you know, as soon as those tool set become available, it was like, darn, would I like to not have to do this?

And then it was an interesting path of discovery of the tooling and even as I was learning about it, how the tooling changed and matured during a short period of time, like of, two, three months, that actually made what I envisioned possible. So this process that was very clunky on the input, we just redirected people to Jira and took all this time on the back end, basically is now largely run by an agent, Gumloop agent that connects to Slack and Jira.

And basically when I just started, it sounds pretty straightforward, right? But there is a, a lot that goes into it because it needs to understand the context of, like, where should the tickets go to. We have a number of different crafts within data, like data scientists and data platform engineer do very different things, and it needs to understand the sense of urgency and create high quality tickets.

We don't want just junk tickets that we then have to go back to people and ask them more questions. And it needs to be able to create those within Jira in the right space and all those things. So it builds up, and I was very specific about what I wanted. I tried other tooling. I'm not gonna say what because I don't wanna throw anyone under the bus, and it's great tooling too just for other things, but it just didn't have the features I wanted.

Then I discovered Gumloop. I did the cohort. I Tried it out. They were very workflow focused at the time, and I started building their workflow and it was insane. Just complexity of the workflow to do what I was trying to do was unjustifiable. And as I was grappling with it and trying the different things, I did another cohort with them, and that was the point when they switched to agents.

And like they say, everything changed. So this was the point where I was like, okay, now it can do what I want it to. And now I have an agent embedded within a workflow. And even then I needed to wait for a couple more new features or updates that would make what I wanted possible. For example, something as simple as not for people to be able to pose that question in Slack without a mention of the tool, 'cause you can't train 90 plus people, there are more than 1,000 people in that channel, to have to mention it every time.

So even something like that. And that was I saw a lot of people who had that shift of, oh, you know, I had these ideas, but now I can actually do them And I think that there really, really was this realization moment of AI is not gonna be in a year what it is now, just like it wasn't the same when I started working on this tool and when I finished it.

And I forgot what question I was answering, but...

Galen Low: No, that's really good because I feel like there's three really important things you said there. One, you were just waiting for the day that you could automate this triage process. And if I'm understanding you correctly, there's 1,000 stakeholders and your team in this channel dumping in requests or observations or bugs or, you know, all this information that you then need to triage and enter into Jira and then have a plan to resolve or deliver, and that obviously took a lot of time.

It's that kind of thing that sort of makes your day disappear, and it's not very gratifying. You were just waiting for the technology to be there. Second thing I like is that the technology arrived, but the first thing you tried wasn't the thing that worked. And you, you had to go through that process of running into obstacles and being like, "You know what?

Maybe this isn't the right tool." The technology does a lot of things, does a lot of things well, but just like any product, there's gonna be one that fits, and there might be a whole bunch that don't. And being able to call it and be like, "Okay, I'm not gonna mash my head against the wall anymore.

Let's see if there's something else." And then the third thing I love is the technology is changing, and it's moving so fast that in a way it's reacting to our collective frustration, our collective wish list of "Oh, if only it could do this," and we all have these sort of, you know, pain points.

And I see a lot of people, they've dismissed it. They've been like, "Oh, no, sorry ChatGPT just isn't good at this, period. So we're gonna use this other thing." I'm like, "Guess what? Wait a week." Right? It's changing so fast that you have to keep a bead on it. But I think what you said is really important, is that whatever it is today, in a year it's gonna look so different.

And part of the thing is just, okay, well, let's see what we can solve using the tech we've got today, and then let's almost paint a path into the this is not gonna be steady state for the next decade. This is for now, for the next 10 months, we can probably do this and, you know, even sooner than that, there will probably be a different way.

Alla Tarasenko: And I know people are talking about this, like, how do we isolate our contacts from our tooling, and how do we protect ourselves? Because I think about this, I know it's a cliché, but that something about building a plane while you're falling or trying to fly. I think about it every day because that's very similar to what we're doing.

It's like you're trying to create something, and all the change management around that, right? Like we all know I'm not just playing with a little tool. This just rolled out to 1,000-plus people, so people need to get used to it, they-- it needs to work well enough. You know, there's all these things. And at the same time, the underlying technology's gonna change, your ability to learn and absorb new information is gonna change.

You might have to switch to something else and, yeah, it's just an ongoing thing that you have to stay on top of. And I think we all, at some point, a whole other conversation we need to figure out ways to keep our heads together in this cycle.

Galen Low: Well, I think that is the mindset shift and the skill.

You know, building a plane while you're flying it, I think is the skill. And not to use too many metaphors in one sentence, but, you know, for project and program managers, you know, we often see our role as keeping the wheels on the bus. Now the role is building a bus. You mentioned the mindset shift. Was it a big leap for you to go from, you know, running things and making sure things are running smoothly operationally to building things with the rollout and the change management, or did they go hand in hand?

Did it feel natural and organic?

Alla Tarasenko: I'll be honest with you, and again, being vulnerable here for the sake of anybody listening who might feel the same and wants to have the it's-not-just-me moment. It was hard. It was hard even just logistically because I had this expectation that, oh, to sit down and create this workflow in Gumloop or to sit down and connect everything to my code code or whatever it is, I need uninterrupted time, and I simply struggled for weeks to allocate the time I felt I needed to.

And in retrospect, I realized a lot of that was less about time and more about just, you know, procrastination due to anxiety- Yes. Yeah ... or what- whatever you would call it. But yes, it's just even walking into that space and be like, "Okay, that's what I do now, and I might suck at it, and I don't even know what to expect, and I don't know what connects to what and what is the environment I should be in.

I don't know what I'm measuring myself against. I don't know," to your earlier point, "if I'm just working myself out of my job right now." There's all these questions, and I think it all weighs. It can weigh on you, and it, and it can make those first steps harder. I think once you make them, once you deliver something, you can gain a bit of a momentum and becomes easier, but even then, I think that shift to, like we mentioned b- before, like just shift to, "I'm not solving this with people or flow or communication, but I'm solving with this with building something," is for some, especially if you haven't coded in a long time, it can be a newer thing.

I know there are some GPMs who have been pretty heavy in automation already, and for them that'll probably be an easier shift. But even a little thing like the other day I asked support from the team in doing spot checks during UAT of a, a migration of a knowledge base. I'm thinking about this, like I need SMEs to look at the material they're familiar with so that they can see if it feels right.

And the first question the engineering manager said, "Well, why, why are we not running a code against it?" I'm like,

Galen Low: that other task that they were waiting to be able to automate can be automated now. Yeah. "Hello, we don't wanna do your UAT. Let's just..." But I think it's an interesting mindset shift, and I think it's one of the benefits I find of, you know, working in an AI-forward organization. Where your team of programmers, engineers, data scientists, you know, they also are being supported, it sounds and encouraged to use AI, that they are still enjoying the opportunity to add value to the point where they are actually suggesting it.

They're like, "Well, let's do this first, and, you know, let's use it in the process." And to be surrounded by that, I imagine, at the very least, creates a sense of belonging, even if we're, you know, all in the same boat going towards the waterfall. At least we're all here together. We're thinking about it the same way.

I know a lot of folks listening and a lot of people I talk to, you know, don't enjoy that. They're working in an organization where, you know, there are camps, and it's discordant to have a conversation about AI because some people are gonna be very resistant. Some people are gonna be maybe zealous about it, right?

They're like, "Yeah, let's never work again," you know, "Mai Tai's on the beach," and everything in between. And I think culturally that kinda creates this discord, whereas I love that conversation of "Hey, let's solve this this way." Do you feel like you building, again, I guess, like with AI, do you feel it's bringing you closer to the teams that you work with who are programmers, who are data scientists?

Alla Tarasenko: Oh, I love that question. I haven't thought about this, but I do think so because it's some- it's like a shared goal. And I also think that Gusto is very AI forward. There is a lot of encouragement and a lot of resources to be using AI. You have a lot of freedom with the tooling you use and a lot of access to training and support.

And for a good reason, because, you know, if we want AI, all that power to be available to our small business customers who actually are the case of businesses that could benefit from any help they can because they're trying to do so much with so little, we need to live that, too, right? You-- we can't do that without doing that internally.

And what I love seeing, to your point about bringing it cl-closer to the team is, if it's the team members who are also non-engineers, then you're going through a lot of the same kind of doubts and pains and learnings and excitement. And I've had numerous meetings and calls with folks, like sharing the tool that I've created and, you know, looking at what other people have created.

We've done demos, and it's all been very fun, and you get to know people beyond just whatever you're talking about, so which I always enjoy. But there is also the thought of the cross-pollination of what we are doing, right? For example let's say the data science team is working on tooling around automating some of the question answers around data, and can we bring that at some point to how we handle the intake and have some questions be answered there?

Just as a the first example that comes to mind, but I'm sure there's a lot more. Or again, taking the kind of intake triage as an example, can we work with engineers who are using AI tools to automate some of the acceptance criteria and prepare their coding work and make that part of the whole ticket creation flow?

I like that we can work together on different things in a way that probably wasn't quite possible before

Galen Low: I like that. It's more collaborative. It's not like I'm helping you solve your problem or, you know, you're helping solve someone else's problem. It's like this is our problem. We can work together on this and we have a sort of shared sensibility, I guess.

Not even process, not even just mindset, but okay, let's build a little machine. And that's normal for everyone to hear and understand now and be like, okay, and nod their head and go, "Let's go," and be excited about it 'cause it actually gives us an opportunity to work together differently. I think that's really interesting.

Alla Tarasenko: I wanna create something together about what is the specific angle, perspective, set of skills that each of us brings that will make it better. And that's, I think, what I think about quite a bit, like when we think about all those existential questions is what is it? With all the years that we've spent in the profession, like what is it that we're bringing to those conversations?

What is it that I'm bringing to this tool that an engineer can't or a project manager can't? What is that vision or what are the gaps that I will identify that they might not, you know?

Galen Low: I love it a lot because I see it posted a lot and I, I know what people mean so I don't, like I, I'm not, you know, trying to rock any boats, but I see those posts on LinkedIn that are like, "As a program manager, you know, I'm never the expert."

And I know what they mean. They mean I might not be a subject matter expert at coding or design, but I like the way you're framing it which is like I'm bringing something to the table too. And what I always say is actually we're experts at delivery, right? We're experts at orchestrating teamwork, at driving results, you know?

And I think there's something to that as well. And even to your point on this, you know, the triage workflow is something you know inside and out from everyone's perspective because you've had to and you know what's involved and you know the criteria and you know what information we'd need in the ticket before it gets over to somebody so we're not wasting anyone's time, and I just think that's, you know, it is what we're bringing into the conversation.

I think it's really cool that we get to highlight that. Hopefully people do see it that way as well.

Alla Tarasenko: Yeah. Yeah, and it's definitely something you end up... You have to revisit your kind of history and what you'd done and what you brought to the table in a different way than before. You know, like we've all been in interviews where we talk about, oh, I've delivered that project and I've done that and that felt enough.

Like I've delivered this big thing, it was successful, it was on time. I've done what a program manager should do. And I think right now this definition of what do we deliver, what is enough, how is our contribution measured, it's shifting actively. And I like to be kind of part of that conversation at the very least.

Galen Low: Right. But I think that is the thing is to, you know, almost stay relevant and valuable by staying in the conversation

Alla Tarasenko: Yeah. 'Cause at the end, we're all just learning together and figuring stuff out together. And trying to avoid any big statements or pretending like we know what's gonna happen, and that is all very valid.

But I think in the very beginning, if you're a project manager just starting or other non-engineer really just starting to figure it out, I think I would really suggest to be, like, minimalist and focused. I meet people who are like, "I really wanna try to do this, but I still haven't started, and I need to take these 10 courses.

I've heard about these 10 tools," and they just overwhelm themselves because they haven't do anything. There is nothing that will beat simply sitting down with whatever tool you have access to, and just talking to it about what you wanna do.

Galen Low: Fair, yes.

Alla Tarasenko: So just start with whatever you have. Whatever you learn will at least partially be applicable in other places.

And from the perspective of, okay, I'm now at work and I'm trying to build something, be give yourself some grace and be generous with time. Don't be assuming that you'll build something in a day, because the pieces that you think might take a long time might be, like, ridiculously quick. Actual, the actual process of creating the agent itself, and from having a lot of context, which took months to build, of course, was, like, minutes.

But then making, dealing with all the issues and fixing them, and especially the infrastructure piece oh, we can't create tickets in Jira without having a service account to now have governance and security dealing with whether or not we can do that, and so on and so forth, that is the thing that took weeks.

And especially if you do something new, obviously, you know, it will hopefully get easier with repeat. So that's another piece. Like Infrastructure, understanding the environment will probably take longer than the actual back and forth and the building with the AI. Everybody talks about how AI is non-deterministic.

We know that. I think that shows up in some frustrating ways that you sometimes don't expect. Because, you know, okay, let's say it needs to respond in certain ways, and I've tested it, and it works every time. There is a variability to how it responds, but it works. It gets what it needs to do.

And then you go in a week later, and it does something like completely different. And whoa, where does this come from? So there's definitely a certain zen that I want to develop with you. Okay, now this is happening, and like how do we figure that out? And, ah, what else? Yeah, just let yourself learn and don't get overwhelmed.

You will always have a FOMO. You will always feel like everybody around you has already created armies of agents and knows everything. Just try to solve for whatever you have, the use cases that you have at your plate that will be motivating, and you will very quickly see if it works or not, and you will learn from that.

And try to be... I think that's, in general, is a good advice for anybody working on any product, be strict with your scope. And just don't try to do too much right away, especially if it's a new tool that you're working in.

Galen Low: It's funny because I bet a whole bunch of developers listening are like, "See?

The things that you don't think will take long, take long, and the things that we th- You know, that's why our estimate is off. Now you guys get it." And I'm like, "Yeah, actually," you know, and it sounds right, that whole notion. It's interesting because that whole getting started, like it's almost designed in a way that it's like gratifying right off the bat.

Oh, I wanna be able to build a thing, chat, chat, chat. I have a thing. I have a functioning website. I have a document. I have this. And you're like, "Wow, cool." But then, you know, you touched on it earlier, which is that once you get into infrastructure, governance, and like compliance, and you built a, you know, a very narrow scope in some ways on Triage, and you also- Had that roll out to 1,000 people.

So, at this point, you're building software, right? With users, a user base. With, you know, bugs to solve for, with, you know, connectivity to maintain. And no, it wasn't just prompt, prompt, prompt, done. You have created something organic that is probabilistic, that does need maintenance, that will grow over time.

You're not done ever. This is like The Parenting Trap, right? Where you're like, "Cool, done. Oh, wait. No, I'm not done." That was just the toddler years, you know, and there's still more to go.

Alla Tarasenko: That's a good comparison. Yes, exactly. It's a bit of a roller coaster.

Galen Low: I wonder if we can dive into your stack a bit.

Specifically because in my community of digital project managers who are used to working with technology and technical teams, their big thing is "I'm done with, you know, AI note takers." Fine, whatever. How do I plug these things together? And what struck me was what you said earlier, is that it's not like you have just some appliance you're plugging in.

It's not-- It doesn't all live in one place. You've connected a number of different platforms, including Gumloop, including Slack, including Jira, and probably a whole bunch in between. For whatever you're willing to share, could you tell us about the tech stack and the flow and what's plugging into what in terms of making this intake and validation flow work?

Alla Tarasenko: Yeah. It is a live thing at my work, so I'm not gonna go too deep in it. But I think it's pretty consistent with probably what a lot of companies are doing, which is that we have access to a variety of enterprise tools like Claude or OpenAI or more specialized niche tools like Gumloop, and we use MCP with RunLayer, and we use that to be able to connect to other tools within our space, such as Google Drive or Jira or Slack, and so on and so forth.

So pretty straightforward in that way. We had a bit more complexity with how things were set up before, had to go through Bedrock, AWS, et cetera. But it's been made more straightforward recently, and I think at least all non-engineers are grateful for that. But yeah, for something like Gumloop, it's almost ridiculously straightforward the way things are set up on the background, which that you just add the Jira and the Slack to your spaces, and it operates within that.

And I use MCPs for the most part within my code code, which is what I mainly use besides Gumloop. And for things that need to run on schedule, I would use things like webhooks.

Galen Low: So can you just take me through the workflow? Something's listening in Slack to be like, "This is a request."

Alla Tarasenko: Yeah.

Galen Low: And if it's not clear enough to be able to have it get triaged into the queue, what happens and what handles what?

Alla Tarasenko: Oh, okay. So if we take the intake bot as an example, if you work with Gumloop, there are basically two ways to get it to fire when there is a question being asked in Slack. One is by embedding it and mentioning it, which is quicker. Another less quick but more natural in behavior is through a trigger that's created on the side of the agent.

So you basically link to your channel, and you say when something is asked there, and you can make some limitations like don't react to bots, et cetera, you start your engagement. So the person asks the question, the bot responds. In our case, it's designed to have a back and forth with the person because we want to extract some information first or sometimes redirect the person if they think the question is for us, but it's really meant to be for IT or somebody else.

And if they're not redirected, then it will ask additional questions. Basically, every time the person responds, there is a trigger in the thread going back to the agent. The thread gets recorded so that the agent understands that it's still within that one thread, and that goes through the workflow. And eventually, it asks for confirmation.

It tells which team it's gonna reroute to, and it goes off after the confirmation and creates a ticket in Jira using Jira MCP and a service account. So it's always one generic account that creates them, but it's still able to use the tools within Jira to substitute the name of the person instead of the service account, which makes it good for security and other reasons.

But yeah, it's actually both ridiculously straightforward from that sense and amazing at what it does because even though It will do certain things in a very... Sometimes it'll have frustrating problems. Like for example, if a person asks two questions one after another, it can sometimes respond twice.

Galen Low: Right.

Alla Tarasenko: And we've tried 10 different ways to solve this- ... with workflows, and it's still happening. Working on this actively right now. But sometimes, you know, you will have multiple people jump into a thread and have a conversation, and it just handled it with such grace. I'm like, "Oh, I'm so proud."

Galen Low: Proud parent. You're like, "Ah, look at you handling all this complex language and, you know, work politics." But it's all happening in this Slack thread, which I love because, you know, that is where the work is happening, where the question is being asked. After that, when's the next time it sees a human?

In other words, it's created the Jira ticket. It's got the service account. It's got the sort of submitter's name on it. It already has the context. It's remembered the conversation, and it's been able to gather more information. Does it do the triage? Does it decide what the urgency is, or does that kind of go to a human, like to you to be like, "Hey, is this a tier one or two, three?

I think it's a tier three. What do you think?" Or is it just go?

Alla Tarasenko: It's a great question. It suggests urgency on the basis of the conversation. That's one of the things we try to extract, like what the business circumstances are. But this version was specifically designed not to do individual level triage assignment, only team level.

Because we have a number of different crafts within data. And so the decision here is which craft should it be sent to, and then within that craft, the team has their backlog grooming or similar process where they assign and move this forward. The thing that I'm looking at at this point is does this happen effectively, and how can I help the teams with that next step, which is do we handle those tickets timely?

Do we have anything hanging? And for that right now, I've launched a script through Claude Code that once a week sends some summary around which tickets have upcoming deadlines and, you know, how many were filed, and how many have been there for longer than a week. And now I'm gonna playing with how do I turn this into something a little more directive.

So that instead of just having a list of statistics, basically metrics around the tickets, how can it suggest, like maybe send direct communication to the leads of those particular teams and say, "How about do this with that and do that with that?"

Galen Low: Yeah. Here's what I generally suggest you do today. Nudge.

Alla Tarasenko: Yeah. But there is a lot that can be done with this tool. I have a roadmap for the future versions and what can be done with it for it to be more helpful, to maybe handle more questions directly without the need for ticketing at all, about how to make the routing more precise, like at not just team, but sub-team, and then sub-sub-team level, and so on and so forth.

There's a lot to play with. But again, back to be, be strict with your scope Right. Yeah ... so that you can get a benefit before you get more benefit.

Galen Low: Well, talk to me about the benefits. I don't know if this is the right measure, but how much time do you think you're saving with this? Or a- actually, how do you measure the benefit right now?

And then also, how deep does the rabbit hole go in terms of productivity, efficiency, people doing things that they love to do? Yeah. Talk to me about the benefits.

Alla Tarasenko: Oh, God, that's such a great question. And honestly, it's something that w- I think a lot of us could get so much better at, like, how do we measure this?

And with this particular one, I did roughly measure how much time we spend in triage meetings and in doing the manual work. It came up to something like 15, 17 hours a month. The goal is to bring it down to close to zero, which is not quite there because I still oversee the flow and fix bugs and things like that, but it's definitely much lighter.

The other way I'm measuring the success is I've been running relatively regular, like couple of times a year, sentiment surveys around, like, how does engaging with Dataworks for you, for our common stakeholders, the ones who use our system quite a bit, as well as internally retro type surveying for the triage group.

Galen Low: It's funny because it's like we were talking about narrow scope, and I was like, "Yeah, you've picked a very narrow scope," and yet it strikes me that you're actually like a little product business, right? You're sending out NPS surveys to your customers. You have a roadmap. And even though this was just like a slice of, you know, what you and your team do, it's, A, a very deep slice, B, the amount of time being saved that's not insignificant.

And, you know, whether it... and I'm assuming... Well, is it 15 to 17 hours per person involved throughout a month?

Alla Tarasenko: No, it's overall.

Galen Low: Oh, okay. Overall. I was like, "That's wild." It's still a massive improvement, though.

Alla Tarasenko: And it's also like the, the sentiment matters, too, because I think there is a difference between engaging with a helpful bot that walks you through the flow and just being linked to a form.

And then people would have certain issues like, "Oh, we, we never know where that form went or what happened," whereas here it'd be like, "Here is your ticket. Let's follow up on it," you know? So yeah, the kind of the delight- Yeah. ... of the user matters, and that brings us to how do you measure that in general.

And I really like what you said about even if you created a small agent that just cuts 15 hours of work, it's still a little product. And so how do you handle the rollout and the comms around it and the change management around it and what happens to it next? And is it maintenance? Is it, you know, are you gonna contribute to it more?

Who's gonna own it? And probably more importantly, like, how do you wear more of a product hat with it? And this is something I've always felt strongly about, that if you're in program management and operations, which honestly, again, in the kind of companies that I worked with was often interchangeable or something that you did both.

When you create processes for people, you almost want to approach it as a product manager in the sense of you want to understand where the gaps are and how you solve for them, rather than just create solutions from what your idea of something is. And I think with shifting more and more to AI-driven solution, that becomes much more obvious.

Even if you were not thinking that way before, you will now because right now I've started building certain pieces for the program management support that we talked about, like the kickoff, the artifacts the tools for sending reports, you know, and for gathering information from the work management tool and, and stuff like that.

And midway of poking through it, I realized I wanna go now to the team and the people, especially people who are new at leading a project, and see Is this helpful? Or more importantly, what are your biggest problems? You've just been asked for the first time ever to be a DRI or the lead on a program on your own, so you're both an engineer and a product manager and a project manager in one for this initiative.

What is killing you? What's keeping you up at night and, and is that something I can help with? Is it something the tooling can help with? So starting this kind of learning tour next week, and it will inform a lot of what happens with the rest of those, let's call them agents.

Galen Low: Well, I mean, you know, in fairness, from my definition of, you know, agentic AI compared with what I see people get away with calling an agent, what you're doing to me is agentic proactive AI, right?

It's going in there, it's making decisions. It's not waiting for a prompt, you know. And I think for that reason, I think that's what makes it, A, very effective, B, an experience, a product, right? That is not just a conversation, but there's, you know, mechanics and criteria and your own sort of algorithms baked into it, which I think is really interesting.

And even coming back to what you were saying originally, which is what do I bring to the table, you know, as a program manager, as an ops person, or even as a product person, and some of it is that. It's like the sensibility to be like, "Hey, let's do some user research. Let's understand the requirements.

Let's prioritize those requirements. Let's do a roadmap," you know. "Let's get some sentiment surveys out every now and again." And that understanding, I think of rolling something out or change management, I think is something that we bring to the table that I think really deepens the success of the things that we build.

'Cause don't get me wrong, I, you know, I've been talking to people who are like, "I woke up one day and I realized I had 700 cloud skills," and I was like, "I don't even know what most of these do," right? It's build a thing, get to step two. Okay, I think we're done. Let's move on to another thing.

Versus, you know, having the depth and rolling something out, getting feedback on it, and improving it continuously. I think that's a really important sensibility when it comes to building with AI.

Alla Tarasenko: I love this, that, that you mentioned the change management because I've been thinking more and more that this is becoming probably one of the most important skill sets and roles for us.

Because with AI, there's both the expectation of moving fast and experimenting and just rolling things out as they are, and it's already becoming very overwhelming, I think, for folks. And I think, again, this non-deterministic nature of what is being delivered creates another layer of chaos.

So what they're dealing with, like the... what you're delivering comes with a bunch of benefits, but also with a bunch of chaos and confusion. And you still have to apply the change management in the sense of realizing its importance. But the actual things you do might need to change. Yes. And, and I almost think like they need to change towards not even like more communication, but like more empathetic communication, for lack of a better word.

I think the more there is the fast moving, the information, the bots, the chats, et cetera, the more we starve for a bit of human in the middle of all that. Yes. So somebody like, "You know, I know this is a little crazy." It does all these good things. This is something we'll need to cope with and work through for a bit, and I'm here to support you in these ways, and tell me what else you need.

Yeah, it's gonna be very important.

Galen Low: It hasn't escaped me that a lot of the you know, VPs of AI or chiefs of AI that I know have a marketing background. So not just selling something through but understanding needs and then, yeah, delivering and packaging something in an empathetic way that focuses on the benefits for people, and I think you're right.

And at speed, you know, that's what's going to be valuable. A, we're able to create so much stuff, but it's not necessarily the best ideas that are gonna get through. It's the ones that people actually use. And then that change management has to be empathetic and human, but also fast, right? 'Cause otherwise, you know, it's build a thing, right?

It took a week, and then it took us 17 weeks for someone to adopt it. It's like probably not what most organizations are hoping for. Or most builders, so.

Alla Tarasenko: I see this post, I'm, I'm sure you did, too, about, like, how, oh, in, in a few months, AI will be capable of doing what all the humans are doing, and everybody no longer, you know, needed to do their jobs.

And even aside from that statement, whether it's correct or not, I think the big piece that's missing in the conversation there is, what about adoption? And adoption is never as fast as the tech is, and that's true at the macro level, but also within the company, whether it's your little tool or whatever else is there.

So, like, how do you make it easy to adapt? How do you make it in a way that maybe change management can be minimized? Because if people have a lot of tools coming at them, and then also a ton of communication coming at them, that almost becomes counterintuitive versus helpful. So yeah, I think just something to figure out for all of us, but a fascinating thing to think about.

Galen Low: Easy solution, get rid of the humans. No, just kidding. Yeah. Maybe that's actually a good place to round this out with. I mean, A, thank you for going so deep into your triage workflow. B, I know you're exploring that idea of how you can support non-project managers within your team, right? The DRIs who you know, get deputized into a role, and what can you do to help them.

I'm just wondering what the vision for the future of your team, what does it look like? How will the humans on your team be working with AI and, you know, arguably AI teammates? And also, like, where will their AI fluency become make or break, if it hasn't already?

Alla Tarasenko: Oh, that's a great question, and I think about this a lot.

I think we all do. AI fluency is a huge part of our roles already, and it's you know, when we just started talking, you talked about choices being made. I wrote that in on LinkedIn the other day to me, the choice isn't do I do this work with AI or do I do it without AI, it's do I do the work or do I go live in a bus in a forest?

Because my work, at least within the kind of space where I work, the, the type of companies that I tend to support, it does not exist without AI anymore.

And even if I somehow found a company where AI is not used, I'd be pretty non-marketable after they lay me off.

Galen Low: Right. Right.

Alla Tarasenko: So this is not to say, "Well, we're forcing the AI, so we are," but that is just the truth that, yes, everybody is using AI to maybe a different degree, but it's pretty much universal, and it will continue to do so.

And I think where I choke a bit on the thought of what it's gonna be like in the future, like you ask, is things are changing so fast, and literally we are not talking years, we are talking months and weeks of what wasn't possible, what's possible now, that I realize I'm really struggling with being even able to imagine or envision what things are gonna look like, let's say, in a year.

You know, assuming we're all together and working together, like I mentioned, I think there is gonna be a lot more cross-pollination. I hope the positive view that I'm trying to imagine is that we will have a bunch of actually genuinely useful tools that will take care of a lot of annoying work, and we will actually be able to sit back and do more meaningful strategizing and conversations and aligning on cross-team, cross-skill set, cross-domain support and see where that takes us.

I think it's always interesting when people with different viewpoints, whether that's because of the jobs they do or whatever it is, come together, and it allows to connect dots that normally were not connected. And I don't think we often, as people working together, we don't often have just simply space to do that and creative in that way. And if we can create that space for that creativity and create kind of the habit of discussing things together as solutions, regardless of our roles, that would be pretty awesome. But we'll see I like to be a kind of a cautious optimist, but at this point, really just trying to take it day by day and learn and connect with folks around.

Galen Low: Honestly, it's a very yes, refreshingly optimistic and logical view. Like when you think about it, we specialize out of necessity because you can't know everything. But then I think it created a lot of silos in our work, where you step out of that silo and, you know, you're like, "Oh, that's not my problem," right?

That's Al's problem to, you know, deliver a thing, or, you know, that's someone else's problem to do testing on the data. And now it's like we have an opportunity to, you know, as you say, de-silo, collaborate more, have conversations together, be working together instead of working apart. You know, just throwing tickets from one place to another and asking if it's done.

Ideally, that's what goes away, and we start interfacing more. We have a shared vocabulary. We have a shared sensibility. It is almost not possible to know everything, but still it is possible to learn more and have, you know, the helping hand of AI to get you a little bit further. But I like that idea, that it brings us closer together, that we talk more.

And yeah, I know that we see that word be more strategic like a lot, but I think that is the right thing, right? Which is like we can plan and make decisions together, knowing more about what each other do so that we can run with the best ideas. And yeah, like AI is gonna be there, and it can help us get better at working together.

I don't know if that's where I wanna leave that.

Alla Tarasenko: I don't, I don't know. You know? Like we thought the social media will help us come together too. We know where that ended up. But-

Galen Low: Just wait for humans to ruin something good.

Alla Tarasenko: I like having kind of a cautious, optimistic vision. Obviously, I have many other visions in my head as well, like I think a lot of us do, and that come up in a lot of the conversations with the, the industry and, and with colleagues, and people have a lot of concerns and fears.

But I- We'll just chug along and, like I said, learn, try to support each other, usually in the time of uncertainty and change, which I've lived in before. I was I was kind of a, a kid when the Soviet Union broke up. That was an interesting kind of lesson and things falling apart under your feet.

Not quite the same thing, but I do think about it weirdly often these days. And I think the coping mechanism for us always was community and just getting closer with other people who are in the same boat as you are. And I think I'm leaning towards that now, too.

Galen Low: I like that. And, you know, I know you resisted drawing the parallels directly, but I mean, I do think when you look back at it, humans in general, not everybody to the same degree, but, you know, in general have demonstrated resilience through things that crumble beneath us, and I love that idea.

That it's it's community that sort of keeps us together and keeps us from all falling down. Amazing. Alla, thank you so much for this. This has been so much fun. You've been doing a couple sort of stints of posting every day on LinkedIn, which I applaud and also am in awe of. But if people wanna learn more about you, where can they go?

Alla Tarasenko: They should go to LinkedIn. That's really the place where I'm at. I got rid of all the other social media in an attempt to maintain my mental health, but I'm still on LinkedIn, which is counterintuitive to that. But yes, my name on LinkedIn is Alla Tara, and then Senko in parentheses, and I'm doing a May post every day challenge, which is a little challenging because I actually forgot to post yesterday. So we don't have to talk about that.

Galen Low: Well, now that we know you're one TPM among a team of 130 with thousands of stakeholders in Slack. All's forgiven.

Alla Tarasenko: But yeah, I love connecting with fellow ops and program managers, and I love actually meeting new people and having coffee meets and just seeing where everybody's at in these crazy times.

Galen Low: Awesome. I love that. I will include a link to your profile in the show notes as well. And thanks again. Thanks for sharing what you know, for being so humble, for being human. I learned a lot, honestly, so I hope my listeners will as well.

Alla Tarasenko: Thanks so much, Galen. Thanks for inviting me, and this has been a really, really fun experience.

Galen Low: All right, folks. That's it for today's episode of the Digital Project Manager Podcast. If you enjoyed this conversation, make sure to subscribe wherever you're listening. And if you want even more tactical insights, case studies, and playbooks, create a free account with us at thedigitalprojectmanager.com.

Until next time, thanks for listening.