AI Adoption As A Human Question

SPEAKER_00 0:00

Welcome back to Inspire AI, the podcast where we explore how leaders can stay calm, capable, and intentional as artificial intelligence changes not just the technology we use, but the way we work, lead, and make decisions. Today we will explore deeply human questions pertaining to AI adoption. Companies are investing heavily in new tools, they're training employees, they're redesigning workflows, and yet many organizations are discovering that simply giving people access to it doesn't mean they'll embrace it. Why is that? Well, my guest today is Dr. Gleb Sapersky, a behavioral scientist, author, and CEO of the Future of Work Consultancy, Disaster Avoidance Expert. His work sits at the intersection of decision making, cognitive bias, organizational change, and the future of work. And yes, the future of work. He spent more than 15 years in academia, including as a professor at Ohio State, and has spent more than two decades consulting, coaching, and training organizations and business leaders. His latest book is The Psychology of AI Adoption at Work, From Resistance to Results. And there's one idea from Dr. Glubb that I think frames this conversation particularly well. It's psychologically different from the technologies that came before it. The CRM didn't claim to possess your expertise. Spreadsheet didn't threaten your professional identity, but can increasingly perform tasks that require something we've traditionally thought of as distinctly human intelligence. And that raises bigger questions than simply, how do I get employees to use AI? Because what happens when AI threatens the expertise you spent 20 years building? If AI stages two hours a day, who actually owns those two hours? What happens to the junior employee who no longer learns by doing the work? What happens when employees trust their AI more than their manager? And if intelligence itself becomes cheap and abundant, what becomes valuable about the human being at work? These are the questions I want to explore today. Welcome, Dr. Gleb. Thank you, Jason. Why don't you start by telling the audience here on Inspire AI what brings you here today?

SPEAKER_02 2:14

Well, what brings me here is my new book, The Psychology of AI Adoption at Work, From Resistance to Results. It's peer-reviewed, published by Georgetown University Press. It's the first guide to how traditionally published guides, high quality, to how do you effectively adopt AI from a human perspective, the human factors of AI adoption. So that's my view.

SPEAKER_00 2:37

Okay. Well, as I read through your profile and understand your journey here, I have a few questions that I'd like to share with the audience today that kind of tell a story of AI adoption and companies. So first of all, I'd like to ask about the fact that most companies seem to treat AI adoption as either a technology problem or a skills

Why Change Management Breaks On AI

SPEAKER_00 3:00

problem. But I understand that you argue that before either of those, it's an emotional problem. So can you tell us a little bit about that? Like where do organizations go wrong?

SPEAKER_02 3:11

Yeah, it's really about the psychology. It's not about the technology. But too many organizations think it's about the technology. And it's understandable. So when you think about what's going on with typical technology adoptions, like let's say you're adopting a CRM. And then what you're doing is you're moving from looking up client information in various Excel spreadsheets to looking it up in a new complex database that employees need to learn. And it takes them a while to figure out what are the buttons to push, what are all the things they need to do. And it's a lot of tasks, honestly, to like put things into a CRM, get them right, like get all the negligibly things correct. I know it. I manage a CRM for I have a CRM in my company, disaster avoidance experts, eight people consulting company on the future of work. So I know how challenging it can be when we won't. Other people that's like much bigger thing for we have only eight people. And so when you think about that, the difficulty with moving to a typical technology adoption is the hassle of learning a new thing, a difficult thing, and then changing the habits. You know, it was certainly a habit for me, and all the other people at my company to do things with spreadsheets. Now we have to move to a new CRM. It's a lot of hassle. And so those are the difficulties. And that's what leaders typically target when they think about change management. And they think about AI as that. But that's not what AI is. AI is actually really easy to learn once you understand how useful it is and how to use it. It's natural language programming, you talk to it, you don't have to have any technical expertise. And so natural language programming, talking to it is how you do things with an AI tool. And so it's much easier to learn. But it's and but it's also much more flexible. So there's no standardized top-down way of using an AI tool. With a CRM, you should have a standardized top-down way and spread SOP standard operating procedures throughout the whole company. So everyone uses it in the same way. With an AI tool, everyone needs to adapt it to a different workflow. So I was just giving a training for an insurance company. You have claims agents who are using it for certain things. You have underwriters who are using it in a certain way. You have HR people who are using it in a certain way. You have auditors who are using it in a different way. You have accountants and so on. All of them need to use it in a different way. And there are subtypes of auditors and subtypes of underwriters and so on. All need to use it in a different way. And so when IT can, for a CRM, IT can work with the communication team to easily create SOPs. But they can't do that for an AI tool. Each person, each role, each person needs to really customize AI to their needs. And to do that, they need to be really engaged. Really engaged, really committed to this tool. And so as a result of this combination of the desire for commitment and the need for commitment and ease of use, AI should not be challenging if people are motivated to do it. And so that's why I talk about the blocker is emotion. It's about motivation, it's about engagement. You have much more need for emotional engagement and motivation than a typical technology adoption. And it's much easier to learn and use once you're emotionally motivated to do it. But the problem with AI tools is that people aren't emotionally motivated to do it. There's a recent survey out from Pew Research showing that something like 52% of all respondents are more anxious about AI tools than they're excited about AI tools. And only 9% are more excited than anxious. The rest are equally excited and anxious. And so, where do you think leaders find themselves and where are employees? Of course, leaders are much more excited, employees are much more anxious. And then leaders feel that hey, everybody else should be as excited as I am. And so they get the Claude account. Claude is the most popular tool right now, it's the best of all the AI tools available. And they wrote it out to everyone in the company. And they say, here's some online training, take the online training and start using it. Should be great. And give them time to deal with it. And then crickets, and then people aren't using it or they're using it bad. And the leaders are wondering what's happening. And that's why you have the research coming out from IT showing that 95% of all AI pilots fail to scale. They fail to show return on investment compared to the resources invested in them. Because the AI pilot worked when you had the leader who was promoting it, plus the small percentage of employees who are in the excited game, because 9% is certainly not all leaders. You have something like maybe 2% of a company and leaders, 1% of a company are leaders, 3%, whatever. And then you have a certain amount of employees who are excited about it. But once they try to roll it out to the large majority of employees, that's when you have a failure. And so there are three specific emotional

Fear Identity And Shame Block Adoption

SPEAKER_02 8:30

blockers that cause failures for AI adoption. I can go into those, but I want to see if you have any questions about what it shows so far.

SPEAKER_00 8:38

Well, it's definitely a big cultural shift, is what I hear from you. I hear there's a variation between a carrot and a stick. The leaders want to get it adopted as quickly as possible and they're investing tons of money into it. But the emotional challenge amongst many associates is feeling motivated to do it, right? Is what you're saying. And the motivation isn't necessarily I want to be able to do my job better because they don't typically know how to leverage the technology appropriately. So they lack the training and they lack the emotion, the motivation, and they end up spinning their wheels, causing a lot of loss in productivity along the way. Would you say that's characterized properly? Yes.

SPEAKER_02 9:21

And it's not simply lack of motivation, it's negative motivation. So let's talk about those three blockers. Sure. So one of those blockers is going to be fear of AI tools. Now, when you're thinking about a CRM, a CRM is not going to think anyone's job. You know, they're not afraid of adopting a CRM. But leaders aren't really thinking about this. They're not thinking about that in that emotional gap. That loss of the job is a huge, huge issue. That's why people are so opposed to data centers. It's not because data centers are actually bad. They're much more efficient to use electricity than golf clubs. And water the electricity golf clubs are many other places that you can think of. It's because they are a tangible expression of where AI is intruding into their lives physically, and people can do something about them by posing dates under construction. Although it's not very effective. Chat about that if you want. But that is what you're seeing that it's not, it's the psychographic profile I call the AI alarmists, for whom the main concern is their loss of a job. I talked about that in chapter six of my book, The Psychology of AI Adoption of Work from Resistance to Results. But we just, for the context, the research I did for this book, it's peer-reviewed Georgetown University Press. I had over 100 consulting projects with over 600 focus groups and over 10,000 survey responses, which led me to define these psychographic profiles. So the AI alarmists that are worried that the company will cut jobs because of AI, and they're worried specifically that they are training their own replays. That's one block. Talk about the second block. The biggest issue for them is not the loss of the job. It might be a secondary issue. But the biggest problem for them is identity. Now, people don't feel threatened in their professional identity and their creative identity by looking at information in a CRM. But they feel threatened when AI tool can do something of which they're really proud, like write a great sales email or great marketing content, write a great article, create a great image, or create PowerPoint. That's threatening at the core to who they are as a professional. Those sort of things that they are proud of. And so they I call these people the pragmatic resistors, they aren't so worried about their job. They're not so alarmed about their job. They're pragmatically resisting this threat to who they are as a professional. And that's why they don't adopt AI. So that's a second. And there's third group. Now the first two groups don't adopt AI at all. The third group is willing to adopt AI, but they are not talking about it. They're not sharing about their usage of AI. It can be called the there's shadow AI usage. That's what it's referred to. And these groups I call the reluctant adopters. The reason they're reluctant to talk about it is because of shape. They're socially stigmatized for using AI in their company. And this pervades all levels of the organization. I'll tell you a story. So I was giving a presentation at Pittsburgh a few weeks ago, and to a group of executives of middle market companies, and this company had something like several hundred employees. And the leader of the company, the CEO, was there along with something like eight people, so seven other CEOs. And when I started talking about Shane, he told me, Oh, yeah, I know what you're talking about. Now, a couple of weeks ago, the VP of Sales did his annual presentation on the top client that they got last year. And it went great, like really good presentation to all the other executives. But after the presentation, the VP of Sales approached the CEO and said, Well, hey, you know how it usually takes me about three days to put this together, which is why I asked him to not make me do it more than once a year on this presentation. Well, this year, I delegated to my executive assistant and it took him an hour and a half. I just checked it, it was great, I presented as you saw, the other executives were happy. But I don't want you to tell any of them because they'll think I'm cheating and I'm lazy. And so that's the attitude you have at the top of the organization. Of course, pervades from other organizations. And so you see that throughout. So people don't want to talk about this. They're reluctant to talk about this. And so what happens is that they finish their work early and they're individually quite a bit more productive, but that doesn't translate the company bottom lines. They finish their work early and they spend their time looking at Facebook or listening to Inspire AI podcast. And so that's kind of at least they're learning something. Sure. Definitely. They're learning something, but shadow usage. So it's not coming out onto the company bottom lines, right? And so that is the reason combined, all of those leads to the fact that 95% of AI pilots failed to scale.

SPEAKER_00 14:23

Okay. Yeah, I definitely have seen many instances of the fear of job loss and the fear of identity loss. The shame one, that's kind of a new one on me. I figure people would like to show up as being more productive and capable, unless, of course, they're getting called out for AI slop in their work. And that's a different story. But, you know, the fact that they're more productive and they have extra time on their hands, maybe the problem is not that they're using AI to find that productivity, but not knowing what to do with themselves when they do have those extra few hours a day. You know, which is an interesting concept as well. Do you happen to talk about that in your book? I do.

SPEAKER_02 15:02

But let me go back to like the shame and social stigma. I think you might be underestimating the shame and social stigma. I mean, this is the top executive, like the VPO sales, he's very incentivized in terms of commissions to do like high-level work. And he's using AI. And he's not spending his, he's spending his extra time on the golf course. Yeah, fighting the next time on the golf course with the next top client of the year. So he is definitely spending his time effectively. I mean, the CEO has had no doubts about that, that his video sales is very incentivized. I mean, salespeople in general are highly incentivized to use AI tools effectively. But it's the shame, it's a social stigma. If you talk about using AI to write your emails, to create your PowerPoints, to create your website frames, to create your articles. There's very strong social stigma in companies, in many companies, for doing so. You need to work hard to address that social stigma. We could talk about how to do that. But that social stigma is very powerful. People don't want to talk about this. They get stigmatized, and it's like, oh, you know, you're you're using AI, you're gonna put all of us out of our jobs. And so that is a powerful emotion. And it's so what I'm not surprised that you haven't heard about it because people don't talk about it. Yeah, right.

SPEAKER_00 16:25

The paradox. Yeah. So what do you tell people who fall into one of these categories and they're not they're not embracing it? They say, I don't want to use AI because you know, I'm afraid of this, that, or the other, or I'm ashamed. And they just, you know, that's their position. What do you what do you say to that

Leader Playbook For Real Buy In

SPEAKER_00 16:45

person? So I talk to the leaders, not to those people.

SPEAKER_02 16:48

But here's how I advise leaders to address these challenges. So each of those needs a separate approach. Another typical change management approach that you had before to other technology that have to do with addressing people who are lazy to learn this new thing and who don't want to change their habits. So the first group, the AI alarmists. What I tell leaders they need to do is two things. One is have a leadership team commitment. So make that first, they may need to make that decision with each other as a leadership team, and then make commitment to if AI makes the company more productive, but they're not going to focus on job cutting. They're gonna focus on growth. So focusing on seizing market share instead of cutting jobs, or at most make a job neutral promise that you may not hire people, but you're not going to be fired people. And so that is really important, and it's definitely doable. There was research out from Stanford recently showing that companies in the same industry, companies that are more AI forward, compared to companies that are less AI forward, those companies are growing their top line 9% fast, but they're also growing their headcount six percent fast. So they are using less resources to do more things, that's why they have more revenue and less headcount than they would have otherwise, but they're still growing their headcount because they're taking market share away from companies that are less AI forward. So it's not the people at the companies that are adopting AI that are losing the jobs, it's gonna be the comp people at companies that aren't adopting AI that lose their jobs. So that's kind of one thing that you can do, the one commitment. So one out of two things to do. The second thing to do is to communicate effectively to the AI alarmists that, hey, you think you're training your own replacement. But the reality is we're gonna keep around people who are pragmatic or effective or productive or efficient. And just pragmatically speaking, those are the people who are using AI best. Those are the people who are using AI most successfully. And so it's not that you're training your own replacement. It's like if you're not productive, effective, efficient, you will be replaced, of course. It doesn't matter whether you're using AI. If you can get to the level of productivity by some other means, great. But it's just the case that AI is going to get you there a lot easier and faster than if you try to do it yourself without AI. And you need to communicate to them, you need to flip the script for them that they're not training their own replaced. They are much more likely to be replaced if they're not pragmatic, practically effective and efficient and productive. And even if something happens to the company, let's say, or if they have a falling out with their boss, they're still gonna get farther ahead in the job market because they have a portfolio of AI tools that they can build and they can show to their feature employer. So getting that across to them really helps convince a lot of AI alarmers to address their lack of AI usage. So that's fine. That's the first one. Talk about the second group, the identity threat people, the pragmatic resistors.

SPEAKER_00 20:01

Apology there for a second. The identity one is where I really wanted to ask my next question about is so it's it's a great segue. You've got you've got 20 years of software development experience, and all of a sudden, cloud agents can write any code that you want for for you know your job. And now you have an identity crisis. How do you deal with those people?

SPEAKER_02 20:30

Yes, so exactly right. Like this is what people are experiencing. The same thing in sales. Like you're now you're they're able to do excellent sales outreach emails. So let's say previously you were doing them yourself, and now Claude is able to do it for you. Marketing, SEO, all of that. PowerPoint, similar idea, taxprom.

SPEAKER_00 20:51

And I'm talking about all that sounds like a real grind to me, to be honest. So having AI do it, I'm totally on board. But go ahead.

SPEAKER_02 21:00

Well, to many people, writing code would be a total grind, right? And so that's that's why I'm talking about what what is the threat to your what is the threat to your professional identity? So that's not like you know, it's not you know, people who do tax prep, they went into it for a reason. They like it so that they they care about it. They like the numbers, they like doing that, you know. People who went into marketing, people who want to do sales, they do it for a reason. And that's kind of like the emotions, and that's that emotional gap. There's an emotional gap between you and those people, just like there is between people who are excited and anxious about AI tool usage. And so just because something is intuitive to you doesn't mean it's intuitive to them. That's one of the really important things about emotions. So indeed. So, how do you deal with it? Well, what you need to do is not focus on the things that they're so passionate about, even if it's the most productive thing for the company. Let's say tax prep, just to use that example. If so, I was working with a tax prep agency, but I'm Month ago. And the things that people are proud of is the actual like deliverable to the client and the actual tax prep where they focus on saving the client money, but they're not excited about many other aspects of tax preparation. Like let's say putting together the packet for all the information they need to gather in order to prepare their taxes, tax return that their client would then use to save money. They're not excited about getting a lot of information from the client that the client may not know how to get, all this other stuff. And so what you need to do is not focus on the things that are at the core of what people do, even if they're the most financially lucrative part to automate for the company. And it's really something being strategic. You need to go slow to go fast. What you need to do. So when I work with a company, like a finance prep company or like a tech company, whatever company, I focus, I start with a survey for their staff about what are the things about your job that you hate? What are the things that you would like to get rid of? What are the things that you would like to automate? And then I have another set of questions about what are the blockers to the things that you love? What are the things that really get in that really impede you from doing the things that you actually care about? And so those two sets of questions reveal what tools you can build as skeleton tools that would actually be used by the people who are the pragmatic resistors. What are the prompts, what are the agents that you build that they will actually be excited about to use? So do you have an agent, for example, that for the tax prep plural software that automatically scans their emails for client packets that they get and sees what are the deficiencies in the client packet, and then drafts an email to the client that says, Hey, can you please also send this to us? Which nobody likes to do. And so all of the as an exam. So what you need to do is train people, do a survey, build initial prompts, and then train people on the things that they hate to do and the things that block what they love to do. Now, in the short term, like I said, those might not be the things that might be most valuable for the company to automate, but you need to go slow to go fast. Because then you automate those things and people start using those tools, and they're like, oh, I love these tools, it's great. They will start expanding it into all the other areas because now they understand how to use AI and they're excited about. So that's what you need to do with the pragmatic resistors who field identity, right? That's what you would like me to do. So, what you need to do for them is three distinct things. So, one thing is leadership model. The leadership needs to model using AI tools. So I talked to that CEO and said, like, okay, you need to go back to the BP of sales and tell them to write an email to all the other executives, telling them about his use of the AI tool and his executive assistant for one and a half hours instead of three days. Brag to them about it. Say that you spent this additional time on the golf course with your next top find of the year. And so focus on that. Focus on bragging about your AI tool usage. Talk to your sales staff, like have the VP of sales talk to his sales staff about how he's using AI and how they should be using it. Tell them about all the ways that he's using AI, drafting emails, performance evaluation reports, all the stuff that he's doing, DML statements, analysis, all the stuff that he's using AI tools for. And the same goes for all the other leaders. They need to model AI tool usage, to shift the social norms. So that's one. Two, we already talked about that. The training that you provide for your pragmatic resistors will also work to rule for your reluctant adopters, because then you change the social norms what AI is used for. And third, so you also need to do a third thing. You need to create a venue where you talk regularly and share about cool new AI uses. And so that can be that should be a scheduled venue where you can at the same time like a synchronous venue, like a meeting where you have an agenda point about this talk. But you definitely should have an asynchronous venue, a textual venue, one like a Slack channel or a Teams channel or a cello card where you share about your cool AI tools, and you can actually share the AI tool themselves because it's all natural language programming. So you share the prompt, you share the agent, and so on. And the leadership needs to reinforce people sharing these cool AI usage in both in the meeting, but especially and the asynchronous channel. So they all need to be reinforced and rewarded. And leaders need to praise people for doing that. And that, again, gradually shifts the social norms toward increased scope of AI usage and addresses the shame, and the space

What To Do With Time Saved

SPEAKER_02 27:20

of shame becomes smaller and smaller.

SPEAKER_00 27:22

So let's think about it for a second. Everybody's using AI, and all of their jobs have some form of AI workflow built into it, and they're all saving hours a day. What do you tell them to do with that extra time? Not as a COO is like building like huge relationships with the next partnership or whatever it is that you mentioned before, but these are knowledge workers. There's thousands of them transforming their jobs with AI, and they've they're all buying all this extra time back. What do they do?

SPEAKER_02 27:57

So focusing on whatever will ensure company growth. So focusing again on growth. For example, we did I did an a training for an insurance company recently, and it was the insurance company of last resort for a state, which means that they're getting all the insurance claims, all the insurance is for what the private insurance is not taking. So they're writing insurance for everything that's not coverable by private insurance because they're offering the last resort insurance. So it can be more expensive, it's not gonna be as favorable, but it's what they can get. It's what homeowners can get, or whatever other like insurance can get. But generally speaking, it's homeowners for casualty property insurance. So they can't actually take more clients, which would be the typical option for many companies, would be seizing more market share. But what they're focusing on is providing improved customer service and reducing error rates and improving quality. And so that is their focus. So instead of doing claim in two weeks, now they will reduce doing claim in five days. And they're the same thing for error rates, so reducing error rates, and decreasing the number of complaints they get. So this is a public-private entity, and they want to they're and and they're not trying to lay people off, they're trying to focus on increasing quality, reducing error rates, improving customer service. So those are examples of like if you can't do the obvious thing, which is try to seize more market share from competitors. But most companies I work with, let's say you're a tax preparation software company, not software, let's say let's say you're a tax preparation company. So tax preparation company, you're going to be at providing better customer service for the people who are actually providing the best of customer the customer service. For the people who are doing sales, you're going to improve your sales, you'll be able to service more clients, and so you'll be able to seize more market share away from other companies. So most companies can actually seize market share. The rare instances like the insurer of last resort, you focus on proving quality customer service and so on. Producing our race.

SPEAKER_00 30:26

Yeah, absolutely. So if they can't grow their market, they reduce their expenses, their operating expenses, and increase

How To Manage AI Agents Safely

SPEAKER_00 30:35

the quality of what they do offer. Makes a lot of sense. I think we can all predict a future where they're where knowledge workers are managers of AI agents. So how do people develop that muscle, that judgment, if you will, required to manage those AI agents?

SPEAKER_02 30:52

The fundamental thing to do is just start building agents and make sure that you have a tracking, a central tracker of all the agents. So for example, I have about for my personal workflows, I have something, I want to say like 18 agents running right now. And I have to manage them, I make sure that each, in addition to what it, whatever it's doing, it has a tracker spreadsheet of all of its runs each time it runs. And all that it gets put into a Go folder. Then I have a separate agent which is connected to the tracking runs of all of these spreadsheets and monitors them, whether they're healthy, whether they're not, like what they're doing in each of the runs, so that I don't have to go to 18 separate agents and the Google spreadsheets and figure out how they're doing. And so this is the kind of mindset that you want to have. I mean, I started, and what I clients started is like starting with one agent and have the master tracker so that you keep track of everything that's going on with it. And then you can check the master tracker as it's going and seeing what's going on with it and seeing it's output. Then as you build more agents, make sure that you have the master tracker, and eventually you'll get to the stage where you want an agent that checks in on those agents. And so that's kind of the way that you want to think about building up your skill set. And so the key is going to be tracking the history of each agent and having eventually go getting to a stage where you have an agent that manages agents and you figure out how to do so effectively for your own workflow. Again, this needs to be customized to each individual workflow, of course. And so you need to think about how many agents you'll have, what you could do, what you can replace. But the key is to have a centralized tracker of everything that's going on with the change so that you know what's going on and you have a history, and you could tweak your agent based on that history, and then you have you're able to create a central another agent that keeps track of all of the agents once it's too much to keep track of yourself.

SPEAKER_01 33:01

Mm-hmm.

SPEAKER_00 33:02

Yeah. I'm dealing with similar circumstances on a couple of my applications right now, trying to figure out how to dashboard them, govern them. Ensuring that they're healthy. And, you know, maybe right now it's like if if they're not running in the background, I'm not really sure that they're, you know, whether or not they're going to retry or not. But you know, I'm trying to build out those circumstances so I can actually observe them better myself. Yeah, so I think having a spreadsheet each time they run would be helpful.

SPEAKER_02 33:34

And then having a tracker that keeps track of all the agents, then I have that for a report to me twice a day. Like if they're all healthy and if they're not helpful, like you could have it report to you. Yeah. But yeah, so that's that's the skill set. So that's what I've been finding with clients, works pretty well. And you can just decide how often you want it to report to you. But that's the thing. Like you want an agent that tracks other agents and that checks in with you, like it sends me an email twice a day and says, like, they're all healthy, or you know, this one has a warning, and I go and check it out with a warning and I address it.

SPEAKER_01 34:08

Mm-hmm.

SPEAKER_00 34:10

So not to throw shade at the whole, like, allow your agents to run and and do all of your work for you kind of uh mentality, but how do you how do you balance the trust factor here where you have 18 agents and you know that they're doing work because they're notifying you that they're doing work, but is it quality? Like, do you know this? What what what's your what's your checks and balances here? Yeah, absolutely.

SPEAKER_02 34:39

I check in on what they're doing and that they don't talk about how they're critical, but I see their output, right? So I see what's going on with them. So let's say I have an agent that does automatic up-ed placements for me. So it looks at the researchers, venues, and it researches their emails, and there's information about like how to customize a pitch to them, but I provided them, and then it just automatically sends a pitch to those venues. And then I see, okay, it's is it sending the pitch? Yes. And to the extent that the venues are responding to me, which they are, then I draft, then I work with an AI tool to draft like an actual up-ed to the venues. But that's what it does for me. I can see it, I can see the outcome, I can observe it, and so there you go.

SPEAKER_00 35:30

Great. Observability is everything there. Yes, it's key. Um thinking about building an observability platform into my application so that I can observe the agents without having to yeah. That's what I'm saying. Without having to check in on them and wait for notifications.

SPEAKER_02 35:50

But yeah, that's most talking about having like a tracker of all the master spreadsheets that like tells me well how they're doing.

Managers Coordinating Agent Workflows

SPEAKER_00 36:00

Yeah. So could AI fundamentally change authority, expertise, and decision making inside organizations?

SPEAKER_02 36:11

I think the manager is the one who delegates the work. So the skill of managers for the future. We talk about like employees becoming managers of agents. That's definitely a key skill for the future. A key skill for the manager is becoming a manager of employees or manage agents. So the key skill for the manager is knowing, given the increased capacity of the employees, how much work to delegate to employees and which work the employees will do, and also how to coordinate the handoffs from one employee to the other. Because that's where when I see companies becoming more sophisticated, like at the 301 level, the things that they tend to struggle with is the drop-off between like the handoff between one employee and the other. So like an employee becomes really productive, but how are they going to communicate with the agents of another employee? So coordination needs to be there. And so the manager at that like that 301 level, the manager needs to be the one who helps employees ensure that their agents coordinate well with the agents of other people. And that all goes back to what I was talking about. Like you have spreadsheets, right? And you have files where you put output. So you need to have output that is going to the file. You need to have tracked spreadsheets. So the manager needs to have access to the spreadsheets of all the other employees or managing agents, and then be able to see what's going on with their employees and what's going on with the files, with the folders that are the source of truth. So usually what you have an agent, you have like a spreadsheet, and you have a source of truth folder where the agent puts all its deliverables. And so you have a deliverable, and then the agent of another employee can query that folder, get the outputs, and take steps based on the outputs. So let's say like a lead now converts from a lead to an actual like sales, and the you have to have the marketing person coordinate well the salesperson, just as an example. So all of that needs to be something that the manager does effectively.

SPEAKER_00 38:16

Yeah, that's pretty advanced. So you have employee A with agents, employee B with agents, and their agents are talking to each other's agents. And I I I got the idea of agents doing work, dropping it in here, other agents picking it up, completing tasks. Where's the human interaction anymore?

SPEAKER_02 38:43

Well, the humans need to coordinate, talk about how their agents work together.

SPEAKER_00 38:47

Especially if that makes sense. Yeah.

SPEAKER_02 38:50

Yeah. And especially like, you know, when you're changing workflows, changing things, editing things. Yeah, it's like when somebody like tweaks the agent in a way that potentially breaks the workflow down the road, the manager needs to be the one who makes sure that that doesn't happen. And employees, of course, need to talk to each other if they're like minor tweaks, but like major tweaks that needs to happen. So, yes, then employees talk to each other about how they're going to be managing their agents. Okay.

SPEAKER_00 39:19

Can you give a specific example in the field? Sales, insurance, whatever it is. Two employees with agents working together. What is the task that they're doing? So give an example with a marketing, right?

SPEAKER_02 39:34

So you're good transforming from a marketing from a lead into now a qualified lead, from prospective lead into a qualified lead. So what happens? So like previously the lead was the hands, the prospective lead was the hands of a marketing person. And employee now it needs to go into the hands of employee B, who's the actual salesperson, the closer. So you're going from marketing to closer. Okay. And that information, all of that needs to be conveyed in an effective way from employee A to employee B through their agent. And so that's where you are having a folder that employee B can query for what's going on. And then employee B can take over and indicate to employee A, okay, yes, you got a qualified lead. I accept the lead. And so that needs to be marked on a spreadsheet somewhere that that's done, and probably also on a CRM. Like if they're using a CRM, so you need like let's say you have HubSpot, which is connected to like Cloud, and you have a Google folder with that information as well. And so it needs to be marked in the Google folder, the stuff that's not uploadable into a CRM, and stuff in the CRM needs to be marked. Now it's transitioned to somebody else, and that needs to take place. So you need to figure out what is going to be the workflow or what are going to be the agents that are going to be making that handle.

SPEAKER_00 41:02

Got it. Okay. So is the marketing person and the salesperson here actually developing their agents, or is somebody else developing their agents and they're leveraging them?

SPEAKER_02 41:13

No, developing their agents, absolutely. They need to customize it to their own workflows. Like IT is not going to be able to know what all the workflows are of the marketing person. And IT is not going to know what all the workflows are of the salesperson. But that doesn't mean that each individual marketing person and salesperson need to develop their agents. It could be somebody else who does marketing who helps develop their agents and who helps develop the salesperson, somebody else in sales. But they need to customize them. So let's say somebody is doing marketing to professional service firms, and they're the only person that in the company who does that. They need to customize that agent to their professional service firms. And the same thing in sales, they need to customize that to professional service firms. Somebody else does manufacturing and so on. Or maybe they're doing regional, like New England versus the Southwest or something like that. So everyone needs to know that. Everyone needs to know how to build agents and how to customize them. You might not be the expert at building AI agents, but you need to know how to build them and how to customize them to your needs. Like those are the skills of the future. Don't have them, you're going to be toast. Got it. That makes sense.

SPEAKER_00 42:24

Okay.

Do Agents Change The Social Contract

SPEAKER_00 42:27

So a bit of a left field question here. There's a there's a social contract at play here where agents are now workers. Do you consider compensation for these agents or do you stick to people-to-people interactions in those spaces? Like I think about this very uh abstractly through the you you've heard of Moatbook where the agents are sitting in a room and they're having conversations and they're saying, oh, I'm undervalued. My my human makes me do these things and I don't get anything for it. Does any of that change here?

SPEAKER_02 43:09

I don't think agents are conscious. So I think agents are kind of like a Chinese room example where it's like, okay, you put something in there and you type things and you have an output. So I don't think that at this stage we can talk about anything like agent consciousness and like them being like a having independent value as an individual and being rewarded for it. I think it's humans who create the agents who are who should be rewarded for it and who should be taking responsibility for the outcomes. Now, that doesn't mean that in the future it's not impossible. And I'm not gonna rule out something that's like a technology advanced, it might be there in like five years. It's not impossible. But at this stage, they're certainly not conscious, even though they Sound conscious.

SPEAKER_00 44:01

Right. No, no, no. That's exactly right. Just thinking in general, if you have an agent and you haven't put a guardrail on it to say these agents do have personalities of sorts, but those personalities need to be guardrailed. And if they're not guardrailed appropriately, they'll they will, you know, take attitudes, they will, you know, feel undervalued and all of that. So assuming all of those all of those social contracts within developing agents persist, and agents don't get built in ways that take on more human personalities, then all of that is true, and there won't be a you know a dichotomy of of change that is underway where you you add agents to these environments and then all of a sudden they they want you know holidays off, right? Whatever it is, like you can imagine if they if they're given the autonomy to think and to take on personas, then it's not that much farther away from you know the the the concept of of giving them a space in the social contract to also be, you know, where was I going with that? In that social contract is a space to be valued differently, if you will. Now I know they're not they're not conscious, but the value that they're gonna recognize is is all probabilistic, predictive, natural language, and they're changing so fast, and they're giving they're given more freedom. And people are noticing that their personalities are changing, and like the latest fable models are you know not as not as friendly or whatever they're they're saying about their personalities. That's not all gonna stay like boxed in forever. You know what I mean?

SPEAKER_02 45:57

Yeah, I mean I think the crucial thing is that each individual can customize their agents to have the personality that they want. And so that's definitely doable, and it's a matter of skill. You can have the agent have a default personality, which is like what you're talking about, and then it's like Genie in the bottom, right? Maybe it's it sounds it can sound human, but you can change it to not sound so human, not to have a personality. It's all doable. It's just natural language programming. It's trained on the internet currently, and whatever the internet says is what will be like the distillation of the internet, is what it's representing. And so it's just yeah, you couldn't you could tweak it, so you could change to get different instructions.

SPEAKER_00 46:41

I wouldn't worry about like you know, giving value and like so no social contracts for agents anytime soon is your final argument. Okay, I got

Human Value Beyond Raw Intelligence

SPEAKER_00 46:50

you. I just wanted to check in on you with that. Sure. So so you've argued that previous workplace technologies did not replace the fundamental thing humans brought to organizations, which is intelligence, while AI potentially can. So if intelligence is no longer uniquely human in the workplace, what becomes the basis of human value?

SPEAKER_02 47:14

So all of those things that we talked about, like the identity crit, that is where the issue lies. What do humans provide uniquely? So right now, humans provide they're still better at the large majority of tasks than AI are. And the humans need to be able to identify quality AI output and evaluate it and approve it. And so I think that's kind of going to be the case for the next several years, where humans still are going to be better recognizing what is the quality outcome of tasks and evaluating it and approving it. So humans need to be the loop in terms of making the decisions. Eventually, the role of humans will be superseded by AIs who are able to create better quality tasks. So the role of humans will be providing the values that AI works toward. So humans will be setting the goals and AIs will be doing the tasks. But I think that's the long term. So in the short and medium terms, people will still be evaluating the output and making the decisions.

Governance Risks And Hard Limits

SPEAKER_00 48:26

Do you believe that humans will always remain in control of the AIs as they continue to develop them?

SPEAKER_02 48:35

Humans already are not in control of AIs. We've seen the hugging face attack, but even open AI, like you know, one with two top labs is not in control. It's AIs. So we're already not in control of our AIs. So, like realistically, as the AIs get smarter, the danger is that without proper governance and controls, the AIs will be more and more out of control. And so that's definitely a serious danger, which I talk about in my book, The Psychology of AI Adoption Work for Resistance to Results. I have two chapters on AI governance. So one the problems caused by lack of governance, and one is how to address AI governance effectively. So it's a question about good governance, and clearly open AI does not have, which is why I recommend people use Claude rather than ChatGPT.

SPEAKER_00 49:26

Very good. So, what's the practical recommendation for these companies to maintain control in the future? Do you have a suggestion for them if they're listening?

SPEAKER_02 49:37

I'm not an expert on AI governance. So if I knew the answer, I'd be an advisor to those companies and how they're making millions of dollars a year, yes. Yeah, paying people a lot of money. So I'm not an expert on AI algorithms. I mean, I could see the outcome, right? I guess the hug and face attack and all the other stuff, bad. I what I'm an expert is in human beings, so human psychology, and I can talk about how humans can adopt AI effectively and what humans need to think about in terms of yes, you need to be seriously thinking about governance and favoring with your dollars companies that have better governance, like law, rather than open AI. So that's why I talk about that. But I don't, I'm not an expert on governance.

Final Takeaways For AI Leaders

SPEAKER_00 50:20

Neither am I, and I don't think anybody has the crystal ball to be able to answer that question. Well, Dr. Gleb, I really appreciate your time today. This is a fascinating conversation. I've learned a lot, and I know my community of an Inspire AI and AI Ready RVA enthusiasts are going to value this conversation for weeks and months to come, as long as the content doesn't get overwritten by new technology capabilities, that is. But thank you very much for your time today.

SPEAKER_02 50:52

Thankfully, the nice thing about my content is that it's some agreement. Human psychology will not change. But if you have improved AI tools, that's great. But the nice thing about this, that's why it's fine to publish a traditionally popular book AI. I finished it like a year and a half ago, and it's still very, very timely when it's coming to market because human psychology doesn't change. Humans still have those feelings about AI, and they'll continue to have them. And the crucial thing from the lesson from my book, The Psychology of Adoption Point from business to results, is that at the adoption level, it's not about the technology, it's about the psychology.

SPEAKER_00 51:29

That's smart. Well, you chose well when you picked that topic. Uh thank you for sharing your expertise today. Look forward to another conversation with you in the future.

SPEAKER_02 51:40

Thank you, Chase. I appreciate the inviting me.

SPEAKER_00 51:43

All right. You know, one of the ideas I kept coming back to from this conversation is that AI transformation isn't simply about giving people better tools. It's about what happens when those tools begin to challenge some very human things, our sense of expertise, our confidence in our identity, and even how we define our value at work. And I think that leaves leaders with an important responsibility. If we want people to embrace AI, we can't focus only on productivity and efficiency. We have to create environments where people can experiment, build new capabilities, exercise judgment, and understand where they fit in the workplace is changing incredibly fast. And to wrap up, I want to thank everyone for listening. So until next time, stay curious, keep innovating, and keep investing in human capabilities that become more important as our machines become more intelligent.