AI can make teams faster. But it can also quietly make them worse. In this episode, Brian Milner and Hunter Hillegas dig into the risks no one wants to talk about—from eroding developer judgment to weakening team communication—and what healthy teams should do about it.

Overview

AI tools are powerful. They can generate code, draft tests, and accelerate delivery in ways that felt impossible just a few years ago. But speed is not the same as effectiveness.

In this episode, Brian sits down with Mountain Goat Software CTO Hunter Hillegas to explore where AI may actually be hurting Agile teams. They discuss the risk of losing junior developer growth paths, the illusion of productivity through inflated metrics, the danger of outsourcing judgment, and how AI can quietly create communication silos inside Scrum teams.

This is not an anti-AI conversation. It is a practical one. You will hear what guardrails healthy teams should consider, why accountability still belongs to humans, and how to use AI as a tool without letting it reshape your culture in ways you did not intend.

If your team is leaning into AI, this episode will help you do it with your eyes open.

References and resources mentioned in the show:

Hunter Hillegas
Blog: AI Doesn’t Eliminate Agile Teams — It Increases the Need for Great Ones by Mike Cohn
#169: Building Practical AI for Agile Teams with Hunter Hillegas
#82: The Intersection of AI and Agile with Emilia Breton
#151: What AI Is Really Delivering (and What It’s Not) with Evan Leybourn & Christopher Morales
Mountain Goat Software
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.

Hunter Hillegas is the Chief Technology Officer at Mountain Goat Software. With over 20 years of experience in software development, product ownership, and team leadership, he leads the creation of tools like the AI Toolkit and Team Home to support effective, engaging learning experiences. Hunter lives in Santa Barbara, California, with his wife and their dog Enzo.

Auto-generated Transcript:

Brian Milner (00:00)
Welcome back, everyone. We're here for another episode of the Agile Mentors Podcast. I'm here, as always, Brian Milner. And today I have back with us Mr. Hunter Hillegas with us. Welcome back, Hunter. Doing great. Good to see you. I'm glad that you're back here with us. ⁓ Hunter is the CTO at Mountain Goat Software. And Hunter has done a lot of programming and work with AI and some of the tools and technologies that Mountain Goat has used most recently.

Hunter (00:09)
Hey Brian how are you?

Thank you.

Brian Milner (00:28)
So we like to have him on to have discussions around AI and how that's changing and affecting the way we do things here in an agile Scrum world. And today what we thought we'd talk about is maybe a topic you wouldn't expect us to cover, but really where AI is potentially actively making teams worse. There's a lot of good things that it can do, but there's dangers, there's downsides.

to this as well. And we want to make sure we cover that and talk a little bit about that. So I know, Hunter, we've talked a lot about how AI helps teams move faster. But we want to talk a little bit about what makes them, how it's potentially making them worse. What comes to your mind when you think about that initial kind of framing of that? Are there things that initially pop in and you think, yeah, definitely here, that's a

a way that this could actually make the team worse.

Hunter (01:18)
can think of a few things and some of them are maybe theoretical, though feel very human nature-y, so I don't think they're too far afield. And I think with a lot of this stuff, we're gonna kinda have to see how it shakes out. But absolutely, I can think of a few things. I'll rattle off a few and we can kinda see what we wanna dig into. mean, the number one thing, or one of the first things I think of is...

this idea of, especially if you're a software developer, if you kind of think of that as exercising that software development muscle, if you give a lot of that over to an AI coding agent and you're not really writing the code anymore, how does that impact your ability to judge whether the stuff that it's doing is what it needs to be doing, right? I I think that's one of the real...

advantages of an experienced developer using these tools is they can tell if it's doing a good job. They can be like, the way that you implemented this is how I would do it, or this won't work because X, Y, and Z. If you are not in the weeds as much or maybe at all or over an extended period of time, does that limit your ability to be able to be effective as it's sort of like overseer?

I think that's an interesting question, right? And you can see this in other industries too, right? I think we were sort of joking before we started, know, if every associate lawyer in the firm is an LLM, how did they ever graduate to be the senior partner, right? I mean, that same idea applies across all industries, I would think.

Brian Milner (02:51)
Yeah.

Yeah, now junior developer and software, right? I think that there, you we, I've talked about this a little bit. I think there there's a potentially, maybe not today, but at least down the line, there's, there's a potential to set yourself up for real heartburn somewhere down the road. Because if, if you, and by the way, I think this is a mistake, right? I think that if, if, if a software company says,

AI can help me replace my junior developers. I don't need junior developers anymore. Then I think what happens then is, all right, we keep our senior developers, but our senior developers eventually are going away, right? Even if they stay with us the entire career, they're going to retire at some point. But what's more likely is they move on. They move on to another company. And when that happens, I have no backfill. I have no one who's coming up, who kind of understands it to a smaller degree.

that we could then elevate and groom into being more successful and more established. So I think there's a potential gap there that's going to happen for certain software companies where they've gotten rid of those junior people. And once that senior level kind of moves on, what happens? Where do you have the backfill of the institutional knowledge for the code base?

Hunter (04:04)
Yeah,

I think it's a critical thing that a lot of companies are going to have to reckon with, you know, the balance between benefit of the tools, but also making sure that you're still training up the people so that, you know, that junior developer becomes a, you know, senior developer one day and all the all of the good that comes with it. The other thing I wonder about is kind of related is the sort of potential accountability gap. So like

If I am like, who wrote this code, right? It's, that thing over there did it. I didn't do it. That's not my problem. So if something doesn't work, not just because you want to yell at somebody, but if you're in service of your users and trying to make something great, if you're constantly being able to say, well, you know, I didn't do it. It's this thing over there. I'm not sure how helpful that is, right? I mean, the problems, the underlying problems are all still there. But if you don't have that sort of same, if you have this sort of cultural escape hatch for like, the model did it.

Brian Milner (04:32)
Ha

Hunter (04:56)
That doesn't seem like a super positive thing either, which makes me wonder about being able to combat that in a way that you still have the sort of natural team dynamics with when you've got these teammates, which are not human.

Brian Milner (04:56)
Right.

Yeah, I mean, I think that has to be... I think that's a misunderstanding of AI in general, but I think a lot of people don't really see honestly, because they wouldn't do that in any other area, right? I wouldn't I use this prioritization technique to prioritize my backlog.

So I have to prioritize it the way it came out of this tool. It's the tool's fault for prioritizing it this way, not mine as the product owner. That would be silly, right? mean, nobody would think that, but I think that you're right, that there is sort of an implication there sometimes where people think, well, I didn't come up with a code. The LLM came up with a code on this. that's who's responsible. No, you're still responsible.

Use the case of.

Hunter (05:54)
Yeah, if your balance sheet's wrong, can't blame Excel, right? It's like, that's not how that works.

Brian Milner (05:57)
Right,

right, you use the case of lawyers and I think that's a good example because we see these cases where lawyers have submitted briefs that have made up things that LLMs have given them, made up case files and the judge doesn't say, well, I'm gonna hold the LLM in contempt, right? The judge says, it's the lawyer who submitted this to me who's responsible and I think that same thing applies to software.

Hunter (06:15)
Right? Exactly. Right.

Yeah. I also wonder about just the team dynamics in general, right? I mean, the communication part is a big part of all of the stuff that we teach. If you're spending more and more of your time talking to your friendly chat bot instead of talking to your team members, what is that? What effects will that have over time? Right? Will that cause knock-on effects that are hard to reason about now, but could actually be quite detrimental over a long period of time?

Brian Milner (06:49)
Yeah, I think another kind of area here is sort of the illusion of productivity ⁓ that is sort of a fool's goal a little bit in this area. I think a lot of people see AI as this silver bullet to solving the productivity problem when it's not, and it never is the key. It's never, never has been about the amount of code that can be output, right? It's the effectiveness of it.

Hunter (06:54)
Mm-hmm. Yeah.

Right, definitely not.

Brian Milner (07:16)
and the value that comes from it. And AI isn't necessarily going to change that. If you amplify, if we're already producing stuff nobody's using, then if AI increases our productivity, well, congratulations, now you're producing twice as much of stuff that nobody's using, right? ⁓ So it's more of a, you probably will end up getting better results by just tightening up your product side of the house.

Hunter (07:31)
Right. Yeah.

Brian Milner (07:42)
to make sure that what you're working on is really valuable.

Hunter (07:46)
Yeah, this sort of potential disconnect between some of the metrics that a lot of Scrum and Agile teams track, like velocity and points and whatnot. You could see those sort of being inflated by some of these tools, even if the customer outcomes don't really improve with the same. So I could see some of these things being used in ways that sort of distort some of those metrics in ways that are not healthy long-term, especially, which is something to definitely be aware of.

There's also the sort of these, so for those who haven't heard me on the show before, I don't want to make it sound like I'm an AI doomer, anti-AI person, because I'm really not. I'm a big, proponent of these tools. of course, they're like anything else, they're tools, and there's good stuff and bad stuff about them. And so I also wonder, even as good as their tools get, as they are these days for writing software, there can be a...

there's sort of this, looks right, because a lot of it sometimes might not be totally wrong. It might be real close, but it's not quite right. And so if you don't have a really great verification system, whether that's really good automated testing, human review, whatever the acceptance criteria are, that stuff's really important. And you can, of course, if you've got strong practices there, can help combat this. But for teams that maybe don't or are not as sophisticated in some of those areas, it can be

sort of even more dangerous because it can be close but not right. And so you can be like, oh, that seems fine. And it turns out not fine. And if that's not getting caught by the testing that you are hopefully doing, then you can get into some trouble.

Brian Milner (09:19)
Yeah, you know, I think there's an element of this I've tried to describe and I don't know if I'm describing this accurately, but I'll kind of frame it this way. You know, I know a lot of people humanize their, just for example, their car, right? They give their car a name and say, that's old, whatever, you know, and kind of call it a name or something.

But no one thinks, you know, the car has driven me to work the past 30 days successfully. So I trust it that it's definitely going to be able to deliver me to my work today successfully. But that same thing is not true in AI. I think because the fact that AI does respond to us in a human-like way, that we have a tendency to over inflate, I think, the

Hunter (09:46)
you ⁓

Yeah.

Brian Milner (10:06)
trust that we give it. You know, if I have given it a prompt and it's given me good answers the past 50 times I've done it, then I tend to, as a human, kind of give it a sense of trust to say, maybe I don't need to review it as closely next time because it's given it to me 50 times correctly. But the reality is you need to check every time. ⁓ Right.

Hunter (10:19)
Yeah, trust it. Yeah. Right. Right.

Yeah, because that's not how the system works at all. Even though it feels

like a person or, know, degrees of success there. You know, it's a non-deterministic math problem, basically. So.

Brian Milner (10:37)
Right, exactly.

And the percentage is the same, that it could be a problem every time. It's like flipping a coin, right? ⁓ If you flip a coin 50 times and it comes up heads, what's the odds it's gonna come up heads the 50 first time? Well, it's still 50-50, right? mean, right, exactly. And that's kind of the same thing with AI, I think. ⁓ The other thing I worry about a lot is sort of the premature outsourcing of judgment. ⁓

Hunter (10:44)
Yeah.

Right, there's no memory there.

Yeah.

Hmm.

Brian Milner (11:06)
that we kind of rely on it in an over aggressively way. we sort of aligned to what you were talking about at the beginning of the podcast, meaning a gap between our knowledge of senior and junior people. But I think that that can even infect a little bit senior people in that if we start to release and turn over judgment, then we kind of dull our ability to make those kind of judgment calls.

Hunter (11:30)
Yeah.

Brian Milner (11:30)
And

that's the thing that think AI is really kind of lacking at is its ability to judge the differences between right and wrong and good and bad.

Hunter (11:39)
Yeah, they definitely don't have that ability, least in no sort of not in anywhere close to the same way that a human would. I don't even mean to the same level, but even in its way, you know, it's completely approximated, right? A lot of stuff seems like it can do these things because habits been trained and the reinforcement learning that goes into it. But it has no innate ability to do any of that thing, right? It has it's not a it's not a person. ⁓ Yeah, I think, you know, I again, I'm

Brian Milner (12:01)
Yeah.

Hunter (12:04)
bullish on a lot of these things, but I think being very eyes wide open about the potential downsides is critically important, right? And it only will become more so as the tools get better and we see them integrated into more parts of our teams and frankly the whole economy.

Brian Milner (12:21)
Yeah, I think the Vibe coding thing is really kind of an inflection point in this area because if you don't know anything at all about coding and you go in and use Vibe coding, well, if there's a problem, let's say, let's use our scenario. It does great 50 times in a row. Well, let's say you're building a pretty complex web app or something and 50 times in a row, it's made great changes, it's worked.

beautifully, but then you get a bug. And if the AI can't figure out how to do that, you're lost at trying to figure out where that bug is because you don't really understand the principle behind how things are coded. Yeah.

Hunter (12:58)
Right. Yeah. Right. I've

seen that a lot. I've seen that. I've read some pretty harrowing accounts of people that, and I just, you know, there's like Reddit threads and stuff on some of these topics where you feel bad for these people, right? They, they built some holes, some app, usually, uh, by coded. And then you'll see their posts saying, yeah, I just deleted all of my data. Like, I don't understand why I did that. And it's just like, I'm, you know, sorry that that can happen.

Brian Milner (13:27)
Yeah, yeah, I agree. And from an actual point of view too, I think it can sometimes magnify, enhance some of the bad practices we have because if I'm using AI to do things that I would normally have gone and talked to someone else about, well, now I have a communication gap and a shared understanding gap.

Hunter (13:36)
I'm sure.

Yeah, right.

Brian Milner (13:48)
I don't know what you think about that, but that really worries me is that lack of shared understanding that maybe falls by the wayside with AI.

Hunter (13:55)
Yeah, I think there's a potential for misalignment to become more prevalent and for it to not be discovered until much later in the process. If each person, instead of talking to their teammates, is talking to their own private, effectively, AI, they're all getting, even if the answers are kind of in the same ballpark, you're still all getting different slices of stuff, right? Whereas if you'd come together and talked about it as a team,

you would have come to a shared understanding and at least we hope everyone would have been the same page and understood what everybody was going to work on going forward. So I do think that's a problem, right? In some ways it's the ability to silo yourself more, which is kind of the antithesis of what we're trying to unlock in these teams. so, yeah, fighting against that natural tendency, especially, and I'll just speak for myself, as a...

Brian Milner (14:34)
Right.

What?

Hunter (14:46)
And I think there are a lot of other software developers like me. Like I would rather not have to talk to anybody else. mean, that's just like my personality type, right? So to be able to have a tool that lets that feeds that is dangerous in some ways.

Brian Milner (14:51)
You

Yeah, yeah, mean, shocker there, right? ⁓ But yeah, I agree. I think you just kind of set yourself up for that as being a bigger problem somewhere down the road because, well, it starts to get to really fundamental questions, right? I mean, do you believe that it's better to have multiple people know how to do something than one person know how to do something?

Hunter (15:01)
Yeah.

Brian Milner (15:20)
There are people who would say, no, I think it's better to have one specialist who knows how to do that thing. And I know that I'm not that way. I know I think it's better to have multiple people. think you're covered then if something happens to that person or if they leave the company or anything else. There's just a lot that you're setting yourself up for failure if you have these kind of silos of information and knowledge. And I think AI could feed into that a little bit. think it could perpetuate

or even encourage a little more of those silos, like you said, buckets of knowledge. And I think teams really have to fight against it. I think they have to be deliberate about communicating more, having more deliberate times to talk over what's gonna happen and how it's gonna happen and the strategy and the architecture, all those things have to be kind of, you have to put extra effort into having more deliberate discussions around those.

Hunter (16:14)
I think you're right. think obviously the sort of rule book here is still being written as this stuff becomes more more prevalent. But if I was going to predict today, I would predict that the more successful teams are going to be the ones that figure that out and understand that it's not a replacement for communicating with their teammates. they need to find new ways to do it so that they could continue to get the benefits of.

these tools the added velocity but without some of these potential downsides and so there may be some some new stuff that evolves out of that which is exciting

Brian Milner (16:44)
Yeah. Well, I don't want to be all doom and gloom, like you said. I want to make sure that we don't leave anyone the impression that we're saying, you know, stay away from AI like the plague or anything. Yeah. No, I think that there's a lot that it can do to be helpful. So maybe that's a good area for us to kind of shift gears a little bit and kind of think about, you know, one of the things we do in software development is we try to establish guardrails to help us prevent

Hunter (16:54)
Definitely not my vibe, no way.

Brian Milner (17:11)
disastrous things from occurring. And if we have those guardrails in place, then they can kind of guide us and help us to go off the deep end. So let me start there with you and say, what kind of boundaries do you think healthy teams put around AI use?

Hunter (17:26)
I TBD to some degree, I mean, think some of this stuff is our best practices that have already existed, right? We, in many different contexts, we talk about automated testing, whether it's like continuous integration, other kinds of checklists, stuff that is repeatable, where we can say with some certainty, of course, you know, it depends on the quality of your test suite, but that you can say,

Brian Milner (17:28)
Yeah.

Hunter (17:53)
with some amount of conviction like, this does work. This is working. I didn't just break everything with the change that I just made. that I would bet is only going to become more more important as this continues, right? And the teams that haven't maybe been as diligent about automated testing for whatever reason will probably find more more value in going that route as much as they can, right? I predict that now that's only going to become more important.

Brian Milner (18:18)
Are there things that you think that the team should try to avoid AI doing? Are there tasks or focuses, areas that you think should always stay kind of more human led?

Hunter (18:33)
That's a good question and maybe one that I would give a different answer to a year from now than I would today as we see the sun shake out a little bit more. think your example before with the, and it kind of ties into what I was just saying, but your example previously about the lawyers who are sending out legal briefs with made up, hallucinated cases. mean, if you are presenting

Brian Milner (18:37)
You

Hunter (18:53)
LLM output as a work product with zero verification? Don't do that, please. Don't do that. That's gonna get you into trouble at some point. I mean, if you're just vibe coding something for yourself on the weekend and there's no stakes, then I guess it doesn't matter. But if this is something that you're doing in your career, please don't do that. Eventually you will get burned, even if you're lucky 49, the other 49 times. So I do think that's important. And also kind of touching on what we were talking about earlier.

Brian Milner (18:59)
Yeah.

Hunter (19:20)
Don't use these tools in a way that they try to replace your communication with your teammates, the product owner on your team, the various people that you're working with to get this stuff done. That may seem like a convenience. It may seem like a productivity win. I didn't have to go to this meeting. I I know we all love meetings and there are probably some that I wish I could get out of.

There's a lot of value there. And if the meaning isn't valuable, that's a whole separate issue, right? I don't think that's something else you could talk about in a lot of other contexts. maybe this means that there's more time in some of those get togethers to talk about stuff that you wouldn't have the time to get to, right? I would rather see that than, I just don't have to communicate with anybody anymore because I've got my magical box over here that does my work for me.

Brian Milner (20:02)
Right,

right. I know that one of the things that AI is helping us do a lot is to automate a lot of the things that previously we would take hours to do. And I know you've done that with a lot of stuff that you've done inside of Mountain Goat. I'm wondering, have you run up against any use cases of things that you've deliberately decided not to automate? Because it would be a better...

It's better to have you in the loop more closely involved or something of that nature.

Hunter (20:30)
There are definitely some things that I would, I still consider high risk or too high risk for me to feel, for trust the model. And I am probably more trusting of it than some of my peers. Like I know people that are significantly less trusting. And so there's, there's a spectrum there, but even in my position where I feel like I've got a pretty good handle on what it's good at and what it's not, there are some things that I still wouldn't like. I'm not letting it.

run wild and completely manage our online store inventory, for instance. Not that we couldn't figure out a bunch of guardrails to protect against that, but there are certain things that are high risk and they're not so automatable in terms of there's not an obvious automation win where it's taking so much time, where it's like,

I definitely need to find a way to automate this. It's just like, I think I'll just keep this as a human task for now. And who knows, my opinion may change as some of these tools get better. But I do a risk assessment with a lot of that stuff to determine, if this goes horribly wrong, what's going to happen? And I think that's a healthy way of looking at it.

Brian Milner (21:33)
Yeah. Well, I want to be respectful of time, people here listening to the show and everything. And so I want to of start to drive us into the finish line here about this. But we're here talking about kind of sort of more of the problems with it and how AI might actually be contributing to the problems a little bit more.

people are listening to this Hunter and hearing us talk about that and hearing us talk about guardrails and everything else. If you could give them one piece of advice that you've learned and kind of how you have implemented AI, what's something that's kind of hard fought wisdom that you feel like you've learned in the stuff you've done with AI so far?

Hunter (22:10)
Well, I do find it to be a very helpful sort of coding assistant slash junior team member. I mean, I still consider it to be in that role. I think for me, it's sort of healthy to know what you don't know and to be aware of the places where, excuse me, where it may not be a top performer. I mean, I remember the first time somebody wrote a newspaper article about me.

and I read it and there was a ton of stuff that was wrong, right? And it wasn't maliciously wrong. It was just like, they make mistakes. And then I realized, oh, every other article I read is, that's subject to that article is probably like, that's not what happened. And so it really kind of flipped something in my mind where it's like, aha, it is, know, to be aware of what you don't know is so incredibly important. And so I do go up and my usage of these tools kind of goes into that as well. If it's trying to do something that I don't understand.

Brian Milner (22:50)
Right.

Hunter (23:06)
then there's a much higher chance that things could go wrong in ways that I can't anticipate, right? And so I have the benefit of, you know, almost 30 years now, yikes, of programming experience where I, you know, I kind of know how it should be put together and I can say, that looks right, or that does not look right, or I can tell that I can see why it did that, but it doesn't know about this other thing that I know about. And so I know this isn't going to work. And if there's areas where I don't have enough of that knowledge, you know,

It might work out okay, but the risk level is going to be a lot higher. So I do think that that's useful and important. then I also think this stuff is really fun. So that's another, that's the sort of other aspect of this. find it fascinating to see what these things can do, especially how quickly they're changing and moving. So if that's the kind of thing that, you know, you, the listener find fun as well, then if this is not an area you've jumped into, I think,

you'll really appreciate it. And I guess the last thing is just, especially with software, the sort of, because these tools can take on a lot of this work themselves to varying degrees, but they can, the bar for like what level of software we can create gets lower, right? It's not as expensive to make software anymore. And so there's stuff that I never would have built because it just would have taken me too long, even fun projects at home or whatever, something I never would have done. And so I'm like, I don't want to work for two weeks on this, but I can do it now in an afternoon.

And so it's like, that means that I'm doing stuff that I never would have done otherwise, which can also be a lot of fun. So I do think there's a lot of, if you'd like to experiment and tinker with these things, there's a lot of opportunity there too.

Brian Milner (24:41)
Yeah, I mean, I remember, you know, I've had conversations with several people that are just friends that aren't in technology in any way, shape or form that, hey, I made this website for my wife, or I made this for, you know, this little side project. And they're able to do that now using AI in a way that they could have never done previously. ⁓ So I think it is an amazing tool and it does open a lot of doors in that way. you know,

Hunter (24:58)
Mm-hmm. Right.

Brian Milner (25:06)
I do think we're still at the buyer beware kind of juncture of this and it's not perfect. And I think as long as you realistically understand the limitations, then you can set up the right guard rails for whatever your circumstance.

Hunter (25:11)
Ha, no.

Yeah, I think that's, totally agree. You know, the hype cycle is strong, but as long as you can kind of see past that and then I think there's a lot of useful stuff it can do today. There's going to be a few more things it can do tomorrow and a bunch more things it can do the day after that. So I, I, I'm pretty excited.

Brian Milner (25:33)
That's awesome. Well, Hunter, I really appreciate you coming on again and sharing your time with us. This has been fascinating and hopefully we can have you back soon.

Hunter (25:43)
Thanks, Brian. Really appreciate it.