The Meridian Point Podcast
Fireside Chat with Glenn Marshall: AI Coaching and the Future of Work
KUMAR: Hey, everyone. Kumar here with Glenn. I think this is the second one of these we've done, although we've done a lot of videos together, right, Glenn? We did a whole series on permaculture. We've done a whole bunch of stuff on XSCALE. And today we have a few topics. We don't know how many of them we're going to get to, but I can think of at least three. Some observations from me, and some observations from you, Glenn. One of them is this idea that humans are somehow more open to an AI coach than to a real human. We'll talk about that. Then you mentioned something about ecosystems, or the interaction between ecosystems as it relates to an organization. And something else. I don't remember what the third thing was that you mentioned.
GLENN: Hormuz. But that might be getting a bit too...
KUMAR: Hormuz, yeah. The Strait of Hormuz.
GLENN: Yeah. And a new world order, I guess. We shall see.
KUMAR: Yeah. We're in the midst of whatever is happening.
GLENN: Yeah, exactly.
KUMAR: So where do you want to start?
GLENN: Let's talk about AI.
KUMAR: Sure, let me give a little background. About six months ago, we won some work to help a company and their product managers and product owners with a class. Training, basically. And also some leaders in that company, again with a class, more training. We were able to convince them that a six to nine or ten week cohort is a better investment of their time and money than a two-day certification class. They still did the two-day certification class with some other certifications, but they said, okay, let's see how that works. So we trained a cohort of product managers, and we're going to be starting with a group of leaders.
KUMAR: Let me go back. For the product managers, they're not all product managers. There are some business analysts, some product owners, some product-adjacent roles. Let's just call it that. We already had a product offering and we enhanced it. We created a bunch of agents designed specifically to aid them through what it takes to build products: from ideation, to understanding the why for the product, to building empathy with the customers, to building a defensible journey map for how these customers would interact with the products and the pain points they may have with existing products and how they might be improved. Then we went into creating outcomes and OKRs, and what we call COBI: Customer Outcomes, Business Impact. It's our version of OKRs. All the way to actually writing epics and stories. So from the big stuff, the discovery work that needs to happen, to the implementation work that needs to happen, this course covers it all. And all the while it's aided by AI agents.
KUMAR: The way we constructed it, the asynchronous portions were devoted to more of the lecture segments, aided by AI. They did exercises. They wrote a product brief, and they created empathy maps and journey maps, all with the aid of these AI agents. The AI didn't actually do the work. It was more interrogative, is that the right way to say it? It asked a bunch of questions, more like a Socratic method, a coaching method, to understand what these people were trying to build, and then help them build it. And it was a success, a huge success.
KUMAR: Within the first three weeks, we found that mindsets for these people had shifted dramatically, especially for the group that was using these AI agents, more so than the people who weren't using them. And I started to wonder about this. Are people just more inclined to engage and be honest with an AI bot that's asking them coaching-type questions than they would with a human? I've worked in the product space for a long time, worked with a lot of product people for a long time, and I find it hard to get them to respond to the exact same questions the AI bot was asking them, just because of time, or, oh, I know this stuff, Kumar, I don't need your help. It's got me wondering. Is a bot better for these types of things, and a human better to sort of tie them together? Because that's ultimately what happened in this class. The bot was better at getting these people to engage with the basics, the fundamentals that maybe they were embarrassed to admit they didn't know, or didn't know well enough. And in our live sessions, they were much more engaged in the content and the learning. So the live sessions went so much better. That's an observation. I'm wondering what you think.
GLENN: There are a couple of things there. So you're saying that even for the people who didn't use the AI agents, they still had a notable mindset shift?
KUMAR: I'd say it was a spectrum. For the people who didn't use them, or didn't use them as much as the others, there was a notable difference in the quality of their work. Their work wasn't complete. They didn't engage with the same level of enthusiasm, because the others had already engaged with the content and had created these artifacts that in many cases they'd never created before. A lot of these people, even though they were in a product organization, were more project-oriented in their thinking. I'd say by the end of the class, all of them had shifted to some degree toward a product-oriented mindset: discovery, iteration, experimentation, hypothesis testing, all the things you'd think are table stakes when it comes to product delivery and product management. But some had shifted more than others. And the ones who shifted the most were the ones who had really engaged with the AI agents. In my experience. It's also not an opinion, it's defensible.
GLENN: Yeah, and there's a feedback loop. Just the fact that some engaged more, they in turn create a feedback loop that engages the people who didn't, and it goes back and forth. But the first question I wanted to ask was this: just getting a mindset shift at all is really hard. So the fact that you did that is notable. Trying to put the AI aside, maybe there's something in the way you packaged it up that somehow opened the floodgates, even without the AI.
KUMAR: Same exact course, just the addition of AI. Same exact course I ran several times last year, 2025, with different cohorts, and the level of engagement was not nearly the same. The story maps were rudimentary at best. The journey maps were rudimentary. All the stuff they were able to get deeper engagement and deeper understanding of was rudimentary at best in the non-AI version of the same content.
GLENN: And you figure the audience you gave it to was representative, and the only variable that changed was AI. That's interesting. Is it possible that somehow people felt more psychologically safe with the AI?
KUMAR: That's my working theory. You're working on your own, and you've got this homework to do. We've all had experience by now with these online learning things, self-paced learning. For me, they're tedious as heck. I can't focus enough to do them. But I have a sense that if I were to take this course myself, I would be more engaged, because I'm working not just with the content and whatever the assignment is. I'm working with this thing that is asking me questions, and I'm providing answers, and it's giving me feedback. It's a conversation I'm having with this bot to help me with my assignment, which is, hey, you've got to think about how to create a journey map for this product you're going to build, or have already built.
GLENN: So you brought up an important point. It is additive over what you were giving before. It's not that interacting with AI is better than interacting with a human. It's that they had an additional thing, the bot, that they didn't have before.
KUMAR: Yeah. The difference is there are asynchronous portions of this class. The way it's designed, we meet every week for a couple of hours, three hours, whatever. In the middle, they're consuming content that normally would be covered in a two, three, or four day lecture. As part of that content delivery, there are exercises that even in the old non-AI course they have to do. They watch something and they do something to exercise what they just heard, to implement what they just heard. But in this case, they're not just doing it on their own. They have a bot to do it with. They input and have a conversation with the bot. The bot is engaging them, because it's been trained to know: this is what they watched, this is the outcome we're looking for, help them get there. It's got that little bit of knowledge it needs to keep the conversation focused around the topic of that lesson.
GLENN: So this kind of feels like an advanced gamification exercise. The AI bot is basically gamifying the lecture, or the exercise rather, and making it more engaging, so you're getting better engagement.
KUMAR: Yes. But it goes beyond games, right?
GLENN: Okay, gamification.
KUMAR: It goes beyond that, in my opinion, because I've used that before too. Short little quizzes in the non-AI version that gamify things and test their knowledge, embedded in the course material. They take the quiz, they get a badge if they get a hundred percent, whatever. This is not that. They don't get anything except the satisfaction of seeing some artifact appear in front of them that's been created through this collaboration with this agent. I'd submit it's gamification on steroids.
GLENN: Probably. I can buy that. But I want to come back to the premise. Are people just more apt to converse with a machine, because the machine is unbiased? It's got no emotions, no feelings, it doesn't make any judgments about your knowledge or lack of knowledge. It's just a machine. Are people more inclined to engage with a machine to get learning than they would with a human? I'm going down a different path than I did when you first mentioned this. I was actually a bit troubled. The fact that people take guidance from AI more readily than they take guidance from a human, that troubled me. But now that we drill into it a little bit, it sounds like it's more gamification on steroids, and it's just a really engaging exercise.
KUMAR: I would submit that if you were able to construct an A/B test, where A was AI and B was a human, I think the human would still result in more engagement. But that's impossible. How can you have that kind of exciting one-on-one teaching and coaching experience in a class of twenty people? You can't. If you could, I submit the human would be better. It's like a classroom. The ideal classroom is a dedicated, engaged, highly skilled teacher, one-on-one, completely attuned and focused on the student. But we can't afford that. So we have ten to one, or twenty to one.
KUMAR: I think it's more like a peer you're talking to, someone you can feel safe with, because they're not judging you. They may know a little bit more than you about that particular topic, but they have a very narrow base of knowledge. You're just asking questions of this peer to help you increase the part of the knowledge you don't have, or don't have enough experience in. It's almost like a peer-to-peer conversation with an AI bot, as opposed to a positional imbalance, where you've got this teacher who's an expert in everything to do with product or whatever. If I were to clone myself and make twenty versions of me to come up and give feedback to these people as they go through their lessons, I just don't think I would get that kind of engagement.
GLENN: I think that's a fair comment. And it feels like you've uncovered a nugget. AI has the ability to personalize co-work, where you have a trusted colleague going through the course with you. You feel psychologically safe, you have a deep connection with them, they're empathic, they're kind, they're gentle, they're not obnoxious or blameful or impatient. And AI embodies those behaviors. So when you guide it with the material around the exercise, you have the ideal learning partner.
KUMAR: Yeah. It's always perfect every time, because it's customized to you and it adapts to you based on your style.
GLENN: So I think this is a significant advance, and it's not as scary as I thought it was.
KUMAR: I'll give you another nugget. Part of the course also includes, when they put in their assignments, a listener bot that looks to see what the assignment is, for what module and what lesson. Then it automatically responds with feedback. This is great, you could have done this. Or, what you did was wonderful. Or, hey, this was pretty close, but have you thought about this, and this, and this? So they're getting feedback on what they submit from a bot. Some of it was from me or one of my colleagues, but most of the time it's from an AI agent. I don't think any of them picked up on whether it was from a bot or from me or from a colleague. They assumed it was always a bot, but they were very receptive of that feedback. They incorporated it into their learning. And that was the kind of feedback that, if you were able to clone yourself, you would have been able to give.
GLENN: Yeah. So again, economy of scale, except on the assessment, because AI basically works for free, a flat rate every month. You can make it intensely interactive, give many more exercises, and have them graded and given feedback. So that's another powerful angle.
KUMAR: Yeah. And you mentioned before we got on the call that you were terrified of this. Tell me more. I know you said your mind has maybe changed a little, but I want to hear the terror.
GLENN: I think I misunderstood, but I'll tell you what I thought. It seemed like you were presenting material as a human, and you were also presenting material through your AI agent, and the material was just embraced when it came from the AI agent versus from the human. People were always a bit skeptical. But the agent, being fluent and perceived as private, people tended to open up more. It was just a better information delivery vector that had a higher retention rate. And my mind went to, oh my god, in the wrong hands this could be very troubling.
KUMAR: Yeah. That's not it at all, but I see where you're coming from. I know there's some research out there that the coaching industry has done, where they have avatars, faces, talking heads if you will, but they're really AI. That's the first level of coaching that someone goes to at one of these big companies that provide coaching services. They talk to a bot. They were doing research on this. I don't know all the research, I can't quote it here, but what I remember from reading the article was that people were very open to that. Only when it got to a certain place would the bot say, let me refer you to a human, or, would you wish to continue talking with me? In those cases, some would take the human, and some would continue to talk with the bot until it just wasn't useful for them, and then they would request a human. My point is that it's becoming more and more accepted. Even things like coaching, at least the first line of coaching, is going to an agent, a bot. And are people more receptive to that than to a human? That's what this article implied. I don't have it in front of me, but I'm going to do some deeper research after this episode airs, and there'll probably be a blog post about this with the research that either supports or disproves it. My hypothesis is that people generally are going to be more open to a bot than a human for simple things, basic things that they're too embarrassed to admit they don't know.
KUMAR: I'll tell you one thing. I do quite a bit of coaching, and it takes many sessions before we get to a level of comfort where the coachee trusts me enough to tell me things that I think they might tell a bot in visit one. Which is so ridiculous when you think about it, because the bot is tracking, recording conversations, doing transcripts, all this sort of stuff. Obviously people emotionally feel more comfortable. Logically it's ridiculous, because there are massive security and confidentiality problems with AI. But it's an emotional thing. Emotions, by definition, don't make sense.
GLENN: You've brought up an interesting point. You and I are both reasonably into AI, and we see a lot of advantages for us in our work today. We can be more productive, more efficient, faster, better, cheaper, all that good stuff. And over the next couple of years, we can see AI getting better and more helpful. I see that clearly myself. If you're doing something today, you could do it better with AI, in general. There are some caveats, of course. The whole agreeableness over accuracy problem is huge, but I'll leave that aside for now. So it's kind of good, albeit very disruptive, over the next couple of years. But in the next five years, I find it terrifying. It's amazing and terrifying. In five years, if we're not managing this properly, and so far we're not, this could very easily go to a very, very bad place.
KUMAR: Yeah. I don't know where it's going to go, but there are a lot of bad places it could go.
GLENN: Yeah.
KUMAR: I'm with you. In the product development space, AI has really shrunk or eliminated, in many cases, the iron triangle of cost, scope, and budget, because anybody can go out there and vibe-code an application in an afternoon, with very little budget. Can that application scale? Is it enterprise-grade? No. But they can still do it. So what does that mean for the workforce? Is the future product development team a human augmented by six agents, or two humans augmented by four agents each? I heard this term: instead of a two-pizza team, it's going to be a two-slice team. One human with several agents helping them do their work. The notion is the human is still going to be the master brain, aided and augmented by agents around them, in whatever they do. So if I'm a lawyer, I might have some agents helping me prepare briefs and drafts and filing stuff. If I'm a doctor, I might have some physical agents with sensors, robots that help me do my job. So I think it's going to fundamentally change how people think about work. And every revolution like this creates new jobs. I don't know what those new jobs are yet. That's the thing, I'm not sure.
GLENN: Well, our colleague Peter Merel has said AI is going to do all jobs, all work, except for the stuff that we want to do ourselves.
KUMAR: Yeah, I find a flaw in that logic. But let's back up for a second. I think you said something profound: the iron triangle in terms of cost. You package up the work, it will do it, so cost goes to zero. Or much less.
GLENN: In terms of cost, you just throw enough horsepower at it. Again, unlimited, multiple unlimited accounts. It can basically do largely infinite amounts of work. Let me reword that. Well, there's still the token cost.
KUMAR: Right now people are experimenting with this stuff, putting a lot of money into it and laying off people. I don't know if it's a direct consequence or not, but people are getting laid off. I read a statistic that just in the last six months or so, a hundred and twenty thousand people have been laid off. But that's beside the point. The point is, is the ROI there? If you move to a model where you have a couple of humans as master brains, with agents augmenting what they do, is the ROI there? You still have to pay for those agents, the tokens and whatever. And is the quality of the products they produce desirable? Is it feasible? Is it usable? All those things. We don't know yet.
GLENN: I want to back up for a sec. Let's pretend you had a large software project done by humans, and the same project being done in parallel, for experimental purposes, by AI, and you outfitted them both appropriately. I think the cost would go down. AI basically works per hour for free. Instead of paying by the hour, you pay by the month, and it's a very modest amount.
KUMAR: I think you pay by the number of tokens these things consume, and it can get expensive.
GLENN: Okay. Would it be an order of magnitude less costly?
KUMAR: That's the thing. With what I know right now, I know Microsoft has limited the use of, especially, Claude Code, specifically because of the token costs. The company I support is very cautiously moving toward an AI-enabled future, because they're afraid that once they turn it on, they're going to get millions of dollars in cost to pay for the tokens that developers and other people use. So we just don't know enough. What is the right mix, the right balance? It would be great to have Peter here, because he's probably done more research on it, or maybe even done some work with it, where you augment a team of humans with some bots.
GLENN: I was just thinking at a very coarse level. Paying ten developers versus paying the equivalent in AI. Surely that would be an order of magnitude less costly.
KUMAR: I would think so. I think you still need to have some humans in there, at least to me. That's why I say the two-slice team: a person augmented by five or six bots. I'm just talking about the augment-by-five-or-six-bots part. You still have a human in both scenarios.
GLENN: So that allows you to massively increase the scope. And you're saving money on the budget. The quality is the other thing I'd like to talk about, that presumably is high as well. So basically all three sides of the iron triangle just disappeared, and it all comes down to... oh, we lost... can you hear me?
KUMAR: Yeah, you're back. I just paused the recording for a second. But you're back. I think we lost audio.
GLENN: I got a phone call. It disrupted things.
KUMAR: Oh, I see. Am I back?
GLENN: You're back. Can you hear me? I'm not hearing you.
KUMAR: Must be on your end, or your Mac. I see your lips move. Your Mac obsession is costing you. This is all going to be on the video, by the way.
GLENN: I have no audio, Kumar.
KUMAR: Okay, so he has no audio recording. All right, well, we lost audio, some sort of technical snafu. So I'm going to end it here. And I'd like to end with a question, really, for all of you. If you could comment: what's your experience been with AI? Do you tend to trust it more than you would a human, or share information with an AI more than you would with a human? Or am I off base? Please leave that in the comments, and we'll see you next time. Thank you.
