AI can help product owners move faster, but faster is not always better. In this episode, Lance Dacy and Brian Milner explore where AI genuinely improves product work and where teams still need strong judgment, clear priorities, and real customer understanding.

Overview

As development teams adopt AI tools at a rapid pace, product owners are under pressure to keep up. Brian and Lance discuss how AI is already changing backlog refinement, product discovery, stakeholder communication, and day-to-day product work. They also explore why many teams are still using AI too narrowly and missing larger opportunities to improve decision-making and collaboration.

The conversation stays grounded in practical application rather than hype. Lance shares where AI can save product owners meaningful time, where human judgment still matters most, and why teams need to be careful about treating AI-generated output as automatically correct. If your team is trying to understand how AI fits into modern product leadership, this episode offers a realistic starting point.

References and resources mentioned in the show:

Lance Dacy
#117: How AI and Automation Are Redefining Success for Developers with Lance Dacy
#164: Why Innovation Efforts Fall Flat with Tendayi Viki
AI Doesn’t Eliminate Agile Teams — It Increases the Need for Great Ones by Mike Cohn
Subscribe to the Agile Mentors Podcast

Want to get involved?

This show is designed for you, and we’d love your input.

• Enjoyed what you heard today? Please leave a rating and a review. It really helps, and we read every single one.
• Got an Agile subject you’d like us to discuss or a question that needs an answer? Share your thoughts with us at podcast@mountaingoatsoftware.com

This episode’s presenters are:

Brian Milner is a Certified Scrum Trainer®, Certified Scrum Professional®, Certified ScrumMaster®, and Certified Scrum Product Owner®, and host of the Agile Mentors Podcast training at Mountain Goat Software. He's passionate about making a difference in people's day-to-day work, influenced by his own experience of transitioning to Scrum and seeing improvements in work/life balance, honesty, respect, and the quality of work.

Lance Dacy is a Certified Scrum Trainer®, Certified Scrum Professional®, Certified ScrumMaster®, and Certified Scrum Product Owner®. Lance brings a great personality and servant's heart to his workshops. He loves seeing people walk away with tangible and practical things they can do with their teams straight away.

Auto-generated Transcript:

Brian Milner (00:02) Welcome in Agile Mentors, we're back. I'm here again, Brian Milner, and I have with us ⁓ once again, the friend of the show, Mr. Lance Dacey with us. Welcome back in, Lance.

Lance Dacy (00:12) Hey Brian, great to be back. I feel like it's been a while, so it's fun to be

Brian Milner (00:15) It does feel like

it's been a while, yeah. ⁓ Always happy to have Lance here. And ⁓ in case you haven't been a listener of the show and aren't really familiar with Lance, Lance ⁓ is a certified Scrum trainer like myself. He also has his own company called Big Agile ⁓ and teaches ⁓ the same kind of classes I teach. So we often compare notes and talk through things. Plus Lance was a mentor of mine.

And still, I would consider you a mentor of mine. ⁓

Lance Dacy (00:47) Well, likewise,

I think we all learn from each other. So that's good. But thank you. I'm flattered.

Brian Milner (00:50) Yeah,

I agree. ⁓ So what we wanted to talk about for this episode is really ⁓ kind of trying to focus in a little bit about how AI is changing things in the product area. So specifically for product owners and the work of what product owners do. ⁓ And there's a lot here. We were trying to...

narrow down our scope and what it is we might want to just focus on here, but ⁓ like a good product person, right? So, I mean, that's a good place to start is sort of like, how are they using AI and what's the good uses of AI and just the normal routine things that product owners do like backlog, ⁓ maintenance, creation of items, refinement of items. ⁓ Where is that actually useful and where is it just

Lance Dacy (01:23) Like a good product person.

Brian Milner (01:47) a fluff that's, know, like it'd be easier for me to just do this on my own.

Lance Dacy (01:53) Man, I tell ya, I was thinking the other day, ⁓ you know, we often say, even in our classes when we're trying to ⁓ get people to learn something, and we do what's called a teach back, right? And we think the best way to learn something is teaching it.

And I can't tell you how happy I've been the last, well, I'll say the last six months I've worked on a course for AI and product management, but I've delivered it, let's see, three times. And it's like the most exciting one, not that they're not all exciting in their own way, but this one's really cool because I'm learning so much. But more importantly, I'm learning how it applies because you can't.

can't interact with every product group you see out there, right? So when you get these people to your classes, you're like, hey, that does connect there, or that doesn't connect there. And I was actually surprised, Brian, how many people show up and really don't know what this thing is, which is OK. I'm not condemning them for that. But their organizations are telling them, hey, you need to start using more AI. Now, development,

When you see that thing write code for them and stuff, that's easily translatable. And yes, you can get in the debate. Is it good code or not? But

If you think, like some of these people come in thinking AI is intelligence, artificial intelligence is what we're calling it. And I'm like, if you think AI intelligence is intelligent, I guess you would say, and you just feed it a whole bunch of information, like they're just loading up an LLM or something, then I find you're going to be dangerous to yourself and to the organization. So what I kick off the class with is like, if you can learn how it actually works, which is really, ⁓

of just call it the world's best autocomplete right now so you can feed it something and it'll autocomplete but when you start realizing where it's going to do a probability analysis of what it's going to come back and tell you you get really scary i think because it's like

Brian Milner (03:35) You

Lance Dacy (03:47) It's going out to the world. Who knows where it's getting its information from and the probability of the return is just, it could still be a bad thing. So I'm like, look, if you think AI is intelligent, you're dangerous to yourself and to your organization. If you treat it and you learn how it works and you treat it and learn how to control it, you become dangerous to your competitors. And that's kind of like the tagline I've been trying to help product people know that I find there's a lot of ways we can use it in product. ⁓ that's really tangible and speeds up.

Brian Milner (03:49) Mm. Yeah.

Lance Dacy (04:17) a lot of things that we do. So, ⁓ you know, we got a lot of places we can start with that. I don't know where you want to dive in.

Brian Milner (04:22) Yeah. Well, no,

no, I think you're absolutely right that, you know, there is a word of caution kind of like, you know, that you have to put out in front of any of this kind of discussion to say, ⁓ you know, recognize the level of the technology and just know where it's at at the moment. ⁓ That it's it's it's it's good. It does a lot of things really well and it can help speed things up. ⁓ But it doesn't mean that you

you abdicate your responsibility. It's still your responsibility. It's not the AI's responsibility. You're using that as a tool. And I think sometimes people lose sight of that. They just feel like, well, I'll just go put it in the LLM and it'll pop out this list of things. I'll copy and paste it and I'm done. Well, good luck with that. You're gonna, right, right.

Lance Dacy (05:14) That's you're dangerous to yourself

and the organization because you're going to be in a legal office here pretty soon when you start feeding it stuff you're not supposed to. And it's like, if we just, if you don't know much about it, then just treat it like a teenager. Like I've got four boys of my own, just treat it like a teenager. You know, how much oversight do you give your teenager? Well, I mean, parenting, you're trying to pull back a little bit, but

When you give them a task that's really important, you're going to, you're going to, we call it human in the loop, right? You never want to take your judgment out of it because there's nothing that can replace your experiences and your opinions. But at the same time, how can we, you know, I think of it like automated testing. You remember, you may not be old enough.

Brian Milner (05:54) Yeah. yeah. No, I am. I definitely am.

Lance Dacy (05:59) So

I remember when automated testing came out, the testers were enlightened because they were now free to do the more human things. And I feel the same way with product managers, instead of spending our time doing the tactical writing and things like that, why not let other tools that can do that better? But the trick is learning how to control it. So most people treat, if you treat AI like a little, you know, smart intern, they're not smart. So just, just think, Hey, I have to give it boundaries.

And if you know that it's going out to the world to give you a response, you can then at least start saying, I need to tell you who you are. I need to tell you what our product space is. I need to tell you the output and the formats that I want. You're going to narrow that field a lot more and you're going to get a lot better results. And so if you learn how to control it, I think it can be a force multiplier.

Brian Milner (06:50) Yeah. I

also think that like you need to, like if you're working in this area, you've got to think of it differently than the general public. Cause you know, we're working in technology and a general public member is going to go and just enter something into chat GPT or something else. And it's going to spit out an answer. And that's what they think the extent of it is. But I mean, we live in a world where you can create multiple agents and you can set up workflows where

You know, one of them, well, the first one might spin out something, but then you have three or four others that then go through a process to check it and verify it and make sure that it's, it's correct. And that's what we did with anything else. We verified and made sure that it was correct before we move forward with, with code or other things. That's what developers are doing now when they use AI is they have to go through processes to check and make sure that it's verified and, correct. So why would that be any different from the product area? And I think that.

product people have to really latch onto this. This is not AI 101. The game now is a game of setting up things that can do some of the more mundane, tedious kind of work for you, but you can...

It's inexpensive enough that you can create multiple agents that aren't just all doing the same thing, but they each have a different role. They each have a different job. And that way, by the time it does get to the human in the loop, you're much less likely to have that junk. You're less likely to have things that are gonna be errors because it's run through a process of multiple checks. that... ⁓

Lance Dacy (08:26) Mm-hmm.

Brian Milner (08:40) That's kind of the world that the developers are living in. And I think that the product area has got to catch up to say, all right, we got to learn this too. We got to get on board if we're going to keep up with what the developers are doing.

Lance Dacy (08:52) You know, we and I teach in my course a prompt craft that's really the important part of, think, starting with this thing so that you can somewhat control it. But, ⁓ you know, just as you were you were talking about with all these agents going out and do that, there's this example I have called a counterfactual example where you can feed it a prompt and then nest, you know, a scratch pad or something with it to have it nest itself many times and give you different answers and where it got it so you can start

kind of evaluating that. I called it a, it was almost like a Monte Carlo simulation for product ideation. I think what this really helps with is ⁓ what I would call the blank page problem for product people. Like you're supposed to be the domain expert, right? But I'm more of a, I'm not as a creative person, I'm an executor. So AI really helps me just to bounce ideas off. I'm like, I like that. I like that. Like it's hard for me just to come up with something sometimes. So the blank page problem,

Brian Milner (09:37) Yeah.

Lance Dacy (09:52) Sometimes product people have that. Luckily you have stakeholders and user research and things like that. But I find it can give you a first draft.

Like you were just saying relatively fast. So you spend your energy critiquing rather than thrashing, you know, almost like what I was saying with the testers. It's like I can be free to be a little bit more creative and explore. I mean, isn't that the point of Agile anyways? We don't know what we're doing until we know what we're doing. So if I can get something spun up and start critiquing it, that's a much better use of my time. So, you know, we talked about just before we got on here.

⁓ What are some of the areas that we'd have to narrow down of where people are using this? And I think like the AI and backlog creation and refinement, I find we could save a huge amount of time from the blank page problem ⁓ using AI to draft.

I mean, whatever you want to call them, acceptance criteria, conditions of satisfaction and acceptance tests. That's another blank page problem because you like, I like to use the word demonstrate next to them, but it's like, Hey, for each user story, follow the invest criteria and give me eight.

acceptance criteria for each one and as soon as I see that man ⁓ it just starts ⁓ clicking but you know the danger zone's passive ⁓ acceptance where you just copy and paste it into JIRA and off you go so I think prompting is one of those product competencies that if you have no idea where to start

I think learning how to prompt can speed up AI and backlog, or I'm sorry, backlog creation and maximize your time instead of giving vague one-liners. You I find that all the time, even in our certified Scrum product owner classes, it's like you say, hey, let's pick a product and then write a vision statement for it. Well, I mean, normally you're a domain expert of the product. That's not as hard, but that's the blank page problem, right? Where you're sitting down going, I can get AI, me 10 things. And I'm like, boom, off and running very quickly.

Brian Milner (11:48) Yeah, yeah, I'm the opposite. I'm the creative side of it to me is like very easy. And that's the part that comes really simply to me. It's actually the follow through that I start to lose interest in and I start to feel like, my gosh, what am I supposed to do next? And I don't really know how to carry this next thing out. So.

Lance Dacy (12:08) And

then you got 30 people overlying on what you're going to say.

Brian Milner (12:11) Right, so

my trick that I've always learned with that is ⁓ because if I ask it to give me an idea, I'm pretty selective about those kinds of things. And mostly ideas, quote unquote, that I get from an LLM, I just kind of reject. Because they just feel like they're not very unique or they're not very creative and truly creative. And so what I found is I have to get it to do like 10.

because one of them is never gonna be enough. If I get to do 10, then maybe one or two of those would be a spark of something I might wanna build on. But usually if I just ask for one, give me an idea for whatever, that idea that comes back just never actually pans out for me. I need a selection of them to kind of work through.

Lance Dacy (13:02) But the great thing is, don't you find that that's a very cheap to do though? So give me 100. Like it takes eight seconds, whether you're doing 10 or 100. And what I was guiding this one lady who basically had just heard the name AI.

Brian Milner (13:07) Yeah.

Lance Dacy (13:15) ⁓ in one of my classes, so it's very interesting to see the different ⁓ disparities of knowledge there, but I was like, hey, you you load up whatever tool you want, chat, GPT, Claude, know, DeepSeq, whatever it is that you use. You can usually create projects.

and you can give it project instructions. That already is going to narrow down that blank page to a little bit better options. But outside of that, you load up books. Like I was experimenting the other day, I'm not a big marketing person. I own my own business, but business is me. I'm out there teaching and coaching. It's like the Scrum Master problem. We work on the business, not in it. Well, I do both and guess how well that goes. So it's like, I need some help marketing, not necessarily just high

Brian Milner (13:58) Hahaha

Lance Dacy (14:02) I'm a marketing person, why can't AI do a lot of this? So I fed it a couple of books that I found that I didn't read the whole book, but I read some chapters. I'm like, I really like that. And it aligns with what I was trying to do. So I even narrowed the prompt further by saying, look at these books. You know how fast it can read a book, right?

Brian Milner (14:19) Yeah, right.

Lance Dacy (14:20) faster than Mike Cohn. There's nobody

on earth that I see breathe faster than Mike, but AI can, COBOL can, but anyway, she was just blown away when she tried it and it put a book in there. So your execution problem is the same way. You can load up books that like what follows your philosophy and narrow it down even more. So it's just amazing how, you know, we're different, right? So I need more help on the creative side, but I'll execute a plan. That's my deal.

We can use it in either way, you know, it's amazing.

Brian Milner (14:51) Yeah.

I'll tell you another area that I feel like AI ⁓ for product specifically ⁓ can really lend a hand. And I think we're going to see a lot of advancement in this area in the next five, 10 years. What am I mean? Not 10, five years, ⁓ right? Five months. ⁓ But that's in kind of traditional product discovery and the discovery side of the product world. ⁓ Because there, I mean,

Lance Dacy (15:10) Nuts.

Brian Milner (15:22) Quite frankly, that's an area where I think it can excel because there's just so many signals that a product person has to keep track of and ⁓ research and be out there looking for. But that's what AI does well, is takes lots of data and then can distill down into here's really what you need to be aware of. I think product discovery,

gets a lot easier now in the world of AI.

Lance Dacy (15:53) Well, it's like, my gosh, where you could sit with 20 stakeholders and not dive through the data that you might otherwise have with AI. I read a statistic the other day, like 74 % of product ⁓ professionals are already using AI to analyze user research data. So if you're not doing that, you're already behind. And ⁓ not road mapping, not story writing, which I think can help there.

It's making sense of other humans, know, messy feedback that you get. So where do you even get, I mean, you sit down, you're a product person like, well, I want to see what people are thinking about our product. What am I sending them a survey? That's terrible. So what can we do? Well, you can go now load up all your data from Twitter or X, whatever, you know, social platforms. They let me read sentiment. Are you going to read all that with 20 stakeholders and go through the data and make sense of it? You know, it's, terrible. So feed that into AI service now.

I was working in our lab and I give them product analytics, product solution interviews, ethnographic research samples, put all that together and say, me five themes of what's happening for a customer. You know how quick that can happen? And so you're spot on, I think. ⁓ Product discovery and research.

I mean, we can really speed up our testing and I don't mean, you know, I don't mean feature testing. mean, what are the things, what are the business problems in the market that our product can solve? That's really what a product owner should be doing anyway. So you can run 20 customer interviews. AI can cluster those seams. They can even surface. We had it surface contradictions. So maybe what's important to see, you don't just react to one thing you see, you're like, well, there's, you know, like Kanoa.

analysis. Here's some people think this is an exciter feature and then 30 other ones think it's a reverse. Like you do that, I'm leaving your product. It's like cluster those together, surface contradictions, you know, I'd call it identifying ⁓ patterns. Used to take days. I don't care who you are. It takes days and not necessarily with more people does it speed up. I don't know, you know, it's like so and then we introduce rag, know, retrieval, augmented generation. That's the goal.

for your product because nobody else has that right. So you start narrowing your prompts, your prompt crafts down with with ⁓ rags. Every org is sitting on, ⁓ you know, years of institutional knowledge that nobody can find. I promise, you know, so now we can go find the data. You don't have to look at it or make sense of it. Feed it in. Show me a table. I was just one of the first things I was blown away with. They eyes like create me as I was like, I just threw in a mess of

Brian Milner (18:27) Yeah.

Lance Dacy (18:40) data say, I get a spreadsheet of this? But I was appalled that it was that quick. I'm like, okay, you know, to me, discovery velocity is the next competitive advantage. Let's track velocity on that. think you're absolutely right.

Brian Milner (18:42) haha

Yeah, I mean, I think that this is kind of what I was trying to refer to with the agent in the world of now AI agents is that, you you talk about things like, you know, sentiment analysis from X or Twitter or whatever, you know, like that's one channel and you've got channels like that. You've got channels like your support team. You've got channels like, you know, what kind of support calls are we getting? ⁓

What are the reviews people are leaving for this? What's, what's, right.

Lance Dacy (19:28) How long were they on a call? Right? Like

you wouldn't normally kind of even think about that, but now that is an option. That's that institutional data that you can say, how long are you on a support call for that problem? Sorry to interrupt you, but I'm like, man, that's an amazing part of it.

Brian Milner (19:40) No, no, no.

Yeah. And kind of what I'm driving at is that, you know, there's lots of these kinds of signals and there's lots of these things that ⁓ even if you are dipping your toe into this and saying, well, yeah, I can create a prompt to do that and I can send out something to go and do that's still behind because what you got to realize is that there are people out there who, know, there's a agent that does one of those things and another agent that does another

And then there's another agent that's sort of the manager of all of those things that's pulling that information and accumulating it together. And then that product person is getting an email at 8 a.m. every morning that says, here's your signals for today.

Lance Dacy (20:26) And it's doing all that while you sleep, right? that's the, and you know, the agent thing, that's scary too, because how much of it's going in the wrong direction while you're sleeping as well. Like we do a simulation in our class too, where I give them a file that's just kind of jacked just to see if they can, you know.

Brian Milner (20:28) Right.

Lance Dacy (20:44) kind of look through it and say, ⁓ this is kind of weird. I don't know. But that agent stuff, you know, I haven't done as much as I would like with that because it's still, I want to, it's almost like I tell people all the time, I do scrum training and they're like, well, is there used to be, are there tools out here that can hold our backlog, our stories and our velocity thing? And of course there is now, right? You put it all out there and you publish a graph and nobody looks at it. Right. So I like the manual. I like to know how it's doing manually, you know, so I know how to

speak to the data. And when somebody asked me why there was a dip, I had to go harvest the data. So I'm kind of that old school way first. Once I know what's going on behind the scenes, then I can automate it because now I know how to inspect it. Right. So I think a lot of people are.

The tragedy is they'll say, this will do all this automatically, but we're not doing a good job saying where did this data come from? Is it curated? Is it biased? Is it PII? it, you know, sensitive, sensitive data? Like what are we doing as part of our definition of done when using AI? What are the acceptance criteria? you know, we talk about the five Q checklist to just kind of think through as I'm using AI's are these following this checklist just like we would everything else, but.

so powerful those agents. I just haven't embraced those as much as I would like right now. I don't know if you have, you may have a lot more experience with that and can talk me off the ledge, but I'm just so old school. I want to know what it's doing first so I can trust it, you know.

Brian Milner (22:07) Yeah.

No,

I think you're right that you do have to always keep that in mind, the trust factor of it. ⁓ I mean, I think there's a couple of solutions there for that. I mean, one of which is that, I think sometimes we're limited in the way that we view those things because we think about humans. And so we think, this is how we do it in a human way, but you're unlimited as far as the agents are concerned. So each one of those signals, before it goes into the flow, I can have three.

you know, research assistant, assistant, ⁓ know, ⁓ agents that cross check the data that comes back, make sure it's real, make sure it's, you know, like all those kinds of things I can build into my flow and, you know, be, be a little more solid on, on, you know, what I'm actually acting on. ⁓ and I think that's, that's part of the key, but I think, you know, at the end of the day, we started here, you're talking about this and I think this still applies very strongly.

the human judgment has to be involved in some place. ⁓

Lance Dacy (23:16) Yeah, it's not Skynet.

You everybody's so afraid of, you know, the pop culture in Skynet. It's like, if you really knew what was going on, you're not going to be that afraid of it. You're going to be afraid that you're using it the way you are. You're going to be like, I should probably scale back a little bit. I'm not saying that won't ever happen, but you know, so many people are afraid of Skynet. It's like, well, you're never going to take the human out of it. ⁓ Now, whether we become managed by AI, that's a different question.

Brian Milner (23:28) Yeah!

Right. Yeah.

Yeah, that's, that's, ⁓ who knows? Who knows? ⁓ but, you I, I do think that that human element is, is at least for the time being that this is how I always have to press preface anything I say about AI today is right now, you know, it needs that huge human judgment. It doesn't do a very good job of making a judgment call. And we still need to be that, that factor that does that.

Lance Dacy (24:10) Well, or we bake that in, right? So the other thing is you're right. If you just, you know, I feel like there's two ⁓ flavors of AI. There's the general versus the integrated. And so a lot of people right now are just experimenting with the general AI, which is like hiring a brilliant freelancer, you know, consultant. You hired them this morning.

They know a lot about everything, but if you brief them, if you say, here's what I want, you have to start from scratch every single time. ⁓ and you don't watch what you hand over, you know, you're going to be in a serious trouble. likewise, or counter to that, you have an integrated AI specialist that's embedded in your tool chain for years and it's just learning. And so it's not just pulling from everything out in the ether. It's got a lot more.

⁓ consensus, you your backlog data, your team's data, ⁓ you know, you don't have to set the context necessarily because it's always there learning from that. So I feel like the practical risk when it comes to that is if the product owners don't know which one, you know, they're holding and how to use it, they, they, you know, paste sensitive strategy into your general LLM. And then you're in the chief legal officer's office.

you know, for three hours or it doesn't go up for years and then you're fired. You know, it's like, ⁓ you know, wrong tool, wrong job. You got to be careful of that. We know that I'm just saying with AI, a lot of people aren't really thinking about that. I feel like are the people that I come across. You know, there's two types.

Brian Milner (25:24) Yeah, yeah.

Yeah, I shared this story with you prior and I just want to share it to the audience as well, but I had someone come through class this week who ⁓ kind of asked me a question in one of our open Q &A sessions about how ⁓ he kind of told me that ⁓ in his organization, the developers had really grasped onto this AI thing and that was speeding up their productivity and they were getting more and more.

productive, they were getting faster and faster what they were doing and they kind of have started to see this imbalance develop between the developers and the product side in that the developers are going so fast that product can't keep up, that they're kind of running out of things to do and they're having to fill in with backlog, like tech debts kind of stuff and all that kind of work. And I think it's sort of ⁓ definitely a warning sign for us to say, yeah, we've got to be able to do this.

Maybe this is a good way for us to kind of start to bring this to a close, ⁓ if you're, Lance, if people are listening here that are in the product area and aren't really doing any of this stuff, where do you think there's kind of one area that they could begin to start to use this, begin to dip their toes in the water and start to make a positive impact that maybe could help them catch up with the developer side of the house?

Lance Dacy (27:00) Well, I mean, obviously I'd say take any kind of course to get the language. the, problem, think first thing is that a lot of people are kind of coming up with their own language for this. remember when I first started this, there was an actual role called a prompter.

I don't know if that's still around anymore. ⁓ Somebody who learns how to prompt, you know, I kind of call it prompt crafting, getting good, just like you'd craft a book or something. So I try to get familiar with the terms ⁓ and take a class kind of just to see that, whether it be online or in person or, one of our classes are great too, because we, I feel like we have a lot of good product knowledge with, with Scrum and Agile and how that can help you accelerate that. But that's kind of, you know, not to sell, I'm not trying to sell those things, but I'm like,

you start getting exposed to a lot of things that you didn't otherwise know. So I would start with something like that. ⁓ But I think just like when you're writing software, I feel like one of the hardest things to do is have a real problem to solve. So, you know, I used to try to learn C sharp and Java and all this stuff and they give you these methods and examples, but it's like.

That's almost like how a user story on its own doesn't have much value. It's like the value is really at the larger thing. And it's like, you got to sit down and say, what problem am I solving? So you can start connecting the process, right? And you don't have to go build a perfect product. That's a sample. You just have to learn to do the process of it. So I think the next step would be ⁓ where you would sit down and say, out of all the things that I have going on, like we do in our class, we call it an AI adoption.

Brian Milner (28:14) Yeah.

Lance Dacy (28:41) I don't know if you've ever seen one of these, I think the Scrum Alliance actually has, or is probably where I got it from, where it's like, here are some tools I want to assess. They're high impact, but I'm uncertain, and I want to learn more in research. So figure out what tools those are. On the other side, we have ADOPT. These are high impact, and we have a high degree of certainty of what we want. We want to scale and standardize those. And the other quadrant is TRIAL. We have a high

certainty, but low, immediate impact. And, you know, we want to pilot these things and then you have a hold low impact, low certainty, just put those on hold. So I feel like once you learn the language of what's going on,

you can then sit back and go, okay, what am I really certain about some things that I could start really implementing? What are some things I need to go learn more of so that you can start separating the morass of, you know, things out there that can help you tackle that? So I'd go to some education just to make sure I know what the terms are. And then I would start classifying how would I apply it to my real world? What are some ways that I really think AI could speed up? And if you don't know what that is, you might learn that from a class.

that you would take as well. Just getting involved, even a user group of some sort, right? So just start getting exposed to those things. And then based on, you know, the certainty that you have of it, start practicing with it. I can't say it any other way. Practice, is that a right term? Like starting, or maybe experiment, you know, and go pick two or three tools. Like try it in ChatGPT, try it in Clawd, try it in Grop, try it in DeepSeek if you want your stuff to go.

to China. don't know. But there's all kinds of tools out there. Some of you are going to be limited and governed to your own organization like co-pilot. That's probably a safe way to start using it, but it may not be as powerful as other things. run your ideas. It's not expensive. Run them through there and see what each one of them does. Right. And compare that to, you know, how you would do this in the real world. And then I would start learning how to prompt.

Brian Milner (30:22) Yeah.

Lance Dacy (30:48) ⁓ and narrow down those results because at the end of day it's a probability machine. All it's doing is trying to figure out what's the next word that should be here.

And what you want to do is really tighten where it goes and gets those words. And you don't really know that until you start seeing what it does and being familiar with, you know, hallucinations and what all these things are. So those are kind of the three big things that I would say ⁓ is to get familiar with the terminology, try to figure out where it would apply, what's the most risky, less risky, start experimenting and playing around with it, and then get in user groups and networking to bounce ideas off of you. None of us are smarter.

than all of us together. ⁓ I don't know how many organizations really know, you know, they have some kind of AI governance or provenance type scenario, but almost like a continuous improvement community of practice on in your organization, where could we apply these things? I'd take that to the next level too.

Brian Milner (31:44) Yeah. Yeah. And I like your, your point there about, ⁓ even just kind of like sitting down and trying to map out what's your, what do you spend, where do you spend your time? You know, as a product person, where, what kind of activities do you do on a regular basis and which one of those things really take a ton of time and which things are very simple that you can get done really quickly. Those things that take a lot of time. just analyze each one of them and say, is there any way that AI could speed this up and make this faster?

Lance Dacy (32:14) Yeah, it's like, you know, I think we teach this in our advanced class. I think you do too as well where, you know, the product owner is kind of the first job to crack under scale, right? So you start getting a lot of teams. We do the four D's. You remember that the delay, discard, delegate or do. And what you start out with is just write out some sticky notes of everything you do. And what are some things you could delay decisions, you know, to the latest responsible moment, delegate, whatever. Anything in that delegate column might be a good, right candidate to

Brian Milner (32:29) Yeah.

Lance Dacy (32:44) start

seeing where you could, you know, let AI start taking some of that over. And I would say the prompt crafting is the core product owner competency that I think I would focus on next because a vague prompt gets vague answers. You won't see the value very quickly. You'll be like, this is terrible. But if you were to do a structured prompt with clear roles, ⁓ clear constraints, ⁓ reasoning, you know, requirements and how you want it in an output.

You can have something you can put in front of a stakeholder in eight seconds that you might be blown away with. You're like, I would have never thought of that. And it's an HTML format. I could just load it up on my blog. A prompt is a brief, basically. So I would actually spend time, once I narrowed everything down, how to get better at prompting and doing prompt craft. I think that's the big next skill, in my opinion. I'm a fronter.

Brian Milner (33:37) That's awesome.

Yeah, no, no, no. think you're right. I know these episodes are things we could probably have every other month and have brand new content to talk about.

Lance Dacy (33:50) mean, we

haven't even talked about documentation hygiene. know, that's the other big one. ⁓ But, you know, closing that out, I know it's like the POs who will thrive aren't the ones who necessarily know AI the most. I think it's going to be the ones that understand it well enough to know when to trust it, you know, when to challenge it, when to put it down and make the call themselves.

Brian Milner (33:55) Yeah

Lance Dacy (34:15) To me, that's the skill, which comes with maturity as well, because some of this is new. so that's where I would start getting educated very well.

Brian Milner (34:24) I agree. Well, Lance, I can't thank you enough for coming back on the show again and working through this complex topic here with me. But thanks for sharing your research and the stuff that you've been talking to your classes about here with the rest of the listeners.

Lance Dacy (34:42) Absolutely. Thanks for having me. And I'd love to hear from the listeners and give us some topics that we can kind of narrow this down and give some focus sessions. I would love nothing more because I learned from them too. You know, we don't claim to be the experts. ⁓ We just claim to try to help experts.

Brian Milner (34:52) Yeah.

Yeah. Awesome. Well, thanks again, Lance.

Lance Dacy (35:01) Yep. Have a good one, Brian.