A Conversation about Bridging the AI Dividend Gap

[00:00:00] So, if you want to just completely waste your company's technology budget this year, I have a totally foolproof strategy for you. Okay, well I have to hear this. Right. So, just take all your brand new, incredibly expensive, generative AI licenses and hand them exclusively to your highest performing, most credentialed senior employees. Which sounds completely insane, right? I mean, it's totally counterintuitive to how any business should operate. Exactly.

[00:00:24] But the newest data, and I mean literally hot off the presses in 2026, it shows that giving generative AI to your resident experts yields basically a zero percent return on investment in a lot of core business functions. Yeah, which is wild. It is. But if you look around the corporate landscape right now, that is exactly what the vast majority of organizations are doing. They're hoarding the most powerful productivity tool of our generation for the people who need it the absolute least.

[00:00:50] It's just a massive misallocation of resources. And I mean, it completely upends the whole narrative we were sold, right? Oh, for sure. We were told it was the great equalizer. Exactly. Yeah. For the last few years, the dominant storyline was that generative AI was going to be this great democratizing force. You know, the rising tide that lifts all boats. Right. A way to take all that institutional knowledge locked inside the brains of the elite workers and just instantly distribute it to the front line.

[00:01:15] But we are finally moving past that theoretical hype cycle. We're actually getting our hands on massive longitudinal data sets about how this is playing out in reality, like on the factory floors and the call centers in the cubicles. And the picture that data paints is dramatically different from that utopian sales pitch. Oh, totally. Night and day.

[00:01:36] Which is the whole mission for our deep dive today. We are pulling apart a newly published 2026 paper by Dr. Jonathan Westover from the Human Capital Leadership Review. Yeah, it's called the AI Dividend Gap. Right. And our goal today is to figure out who is actually capturing the real value of AI in the workplace right now, like why the standard corporate rollout strategy is just fundamentally broken.

[00:01:59] And most importantly, how you, whether you're a leader making purchasing decisions or just a professional trying to navigate your own career, how you can actually capture this technological dividend instead of being crushed by it. Well, to really understand what Dr. Westover is uncovering here, we first have to ground ourselves in a bit of economic history, specifically the framework established by those economists, Dharun Ashimoglu and Simon Johnson. Oh, I love their work.

[00:02:24] Yeah. They make this point that honestly should be taped to the wall of every executive boardroom. The distribution of benefits from any new technology is never a law of nature. Right. It's not inevitable. Exactly. It's not some predetermined physics equation where you plug in a new software and prosperity just automatically flows to everyone. It's always a choice. Always. I mean, think about the steam engine, right? Or early factory automation or even the personal computer. In every single instance, the technology itself was neutral.

[00:02:54] Yeah, it's just a tool. Right. It was the society, the employers, the power dynamics of the era that dictated whether that technology created a broad middle class or whether it just enriched a tiny handful of factory owners. While making the actual work much more dangerous and miserable for the laborers. Exactly. And today with generative AI, we are in the exact same crucible. We're making choices right now about workflow design and access and training that are going to dictate the shape of the economy for the next 50 years. We aren't just evaluating software.

[00:03:24] We're evaluating human architectural choices. Yes. Perfectly said. I love that framing because it completely strips away the excuse of technological inevitability. If it's a choice, we can look at the choices being made today. And to do that, Dr. Westover's paper leans heavily on the October 2026 Gallup American Job Quality Study. And we should really pause on the scale of this for a second because it's massive. Yeah. It's a collaboration with Jobs for the Future and the Families and Workers Fund.

[00:03:54] They surveyed over 15,400 U.S. employees between late January and late March of 2026. Right. So this isn't some tiny little pulse survey you see on LinkedIn. No, it's a comprehensive x-ray of the American workforce. And the most jarring finding from that x-ray is just the sheer lack of penetration. I mean, we live in a media environment where it feels like literally every single person on Earth is using a large language model to do their job. Oh, it totally feels that way.

[00:04:20] But the Gallup data reveals that only 28% of employees actually use AI on a weekly basis. Only 28%. Yeah. And more importantly, a staggering 54% of the workforce has never used it in their current role. Never. Over half the working population is completely disconnected from the most important technological shift of our time. Okay. Let's unpack this because it gets so much more revealing when you look at who actually falls into that 28% of regular users.

[00:04:49] It is heavily, heavily skewed toward people who already hold all the advantages. Oh, absolutely. We're talking about 40% of college graduates using it regularly compared to just 17% of workers without a degree. Huge gap. Yeah. And if you look at the corporate ladder, 37% of managers are regular users versus only 25% of individual contributors. And the industry breakdown is just a total chasm.

[00:05:13] What's fascinating here is you've seen your ubiquitous adoption in these white-caller sanctuaries, like professional services, financial activities, information, media. You have massive chunks of those workforces using it daily. Right. But then you look at healthcare support or manufacturing production lines, retail sales. The data shows that between 70% and 80% of those workers have never touched it for a work task. Okay. So the visual that comes to mind for me here is a marathon. Okay. I like marathons. Right.

[00:05:42] So imagine a city marathon where the organizers have developed this revolutionary new energy-returning running shoe. It practically propels you forward. Like those super carbon-plated ones. Exactly. Yeah. But instead of handing those shoes out to the people struggling at the back of the pack, the runners who desperately need the mechanical advantage just to finish the race, the organizers are exclusively handing them out to the elite runners who are already in the lead group. Which is just wild to think about. It is.

[00:06:09] Meanwhile, the people at the back are left running barefoot on hot asphalt. We're taking the ultimate productivity accelerator and gifting it almost exclusively to the people who are already the most credentialed, the most secure, and the highest paid. That captures the dynamic perfectly. And I mean, it introduces the core concept of the paper, which is the AI dividend gap. The dividend being the time saved, the mental energy conserved, you know, the reduction in drudgery. Right. The actual tangible benefits. Exactly.

[00:06:38] And what makes the Gallup data so vital is that it doesn't just ask about software usage. It cross-references that usage with a very strict definition of job quality. How are they defining job quality? They measure things like financial well-being, workplace safety, opportunities for growth, agency and autonomy. Essentially, do you have a say in how your work is done and are you paid fairly for it? Yes.

[00:07:04] And prior waves of the study found that roughly 60% of the U.S. workforce does not have a high-quality job by those metrics. 60%. Right. Wow. Now, overlay the AI adoption data on top of that. Among the daily or weekly AI users, 52% already hold what Gallup defines as a high-quality job. Okay. But among the non-users, only 32% have a quality job. The technology is completely bypassing the roles that actually need structural improvement.

[00:07:33] It's actively compounding existing advantages rather than leveling the playing field. So if the technology is bypassing the people who need it most, we have to ask how that actually happens logistically. I think it's tempting to just picture, you know, a stingy IT department refusing to hand out passwords. Right. Just locking the doors. Yeah. But Dr. Westover argues it's much more insidious and systemic than that.

[00:07:56] He maps out this four-stage pathway where the so-called AI dividend basically leaks away before it can ever reach the worker. Yeah. He breaks it down into access, use, benefit, and reward. Let's dig into that first stage. Access. So access seems pretty straightforward, right? Mm-hmm. Does the company give you a device, a license, and the basic authorization to log in? Right. Can you physically open the program? Exactly.

[00:08:21] And the issue here starts with how AI tools are currently packaged and sold by vendors. They're predominantly designed for traditional knowledge work. Drafting emails, summarizing PDFs, writing code. Right. All desk job stuff. Yeah. So when a company buys a block of enterprise licenses, they naturally distribute them to the people whose daily tasks look exactly like the marketing materials. They hand them to the HR department, the analysts, the middle managers.

[00:08:48] And the floor supervisor at the logistics warehouse or the intake coordinator at the medical clinic, they just get skipped over. Even though they probably have incredibly repetitive administrative tasks that an AI could automate in seconds. Exactly. They don't even get to the starting line. Which is a tragic missed opportunity because there was this major 2024 study by Elendu and colleagues. Oh, the task exposure paper. Yes.

[00:09:11] It demonstrated that roughly four out of five U.S. workers have at least 10% of their daily tasks exposed to automation or augmentation by large language models. Four out of five? Right. The potential utility is literally everywhere across almost all wage tiers. But the initial access is gated by corporate hierarchy. Okay, wait. So that's the first leak in the pipe. But let's say we fix that. Let's say a company decides to be incredibly progressive and buys a license for every single employee.

[00:09:41] That brings us to stage two, which is use. Right. Actually using it. And I want to push back on the conventional wisdom here for a second because if my boss puts an incredibly powerful tool that makes my job easier directly onto my computer, the logical assumption is that I'm going to open it and use it, right? That is the assumption, yeah. So why is there a massive drop off between merely having access and actually utilizing the tool? Is it just a lack of training or is there something deeper happening here?

[00:10:07] Like if I'm a frontline data entry clerk, I might actively avoid using the bot because I'm terrified it's going to eventually realize it can do my whole job and render me obsolete. That fear of obsolescence is a huge factor. And it really bleeds into the broader issue of psychological safety. The paper brings in some fascinating data from a 2025 Boston Consulting Group survey. Okay, what did they find? So they looked at over 10,600 employees across 11 countries.

[00:10:34] And they looked specifically at situations where access was already granted. So everyone had the tool. Equal access. Right. Okay. They found that regular use among frontline white color employees completely flatlines at about 51%. Meanwhile, over 75% of leaders and managers in the exact same companies are using it several times a week. Wait, so same access, but totally different usage rates? Exactly.

[00:10:59] The BCG data revealed that this gap is heavily tied to enablement and permission. Only about a third of the frontline respondents felt they have been adequately trained. Ah, there it is. And here is the magic number. Regular, confident use sharply increased only among employees who had received at least five hours of dedicated training. Five hours? Yeah. Anything less than that. And the tool just sits there as a source of anxiety. Five hours is a significant investment.

[00:11:29] You can't just send a company-wide email with a five-minute tutorial video and expect people to change their entire workflow. You really can. And this gets to your point about fear. If you're a senior manager, your job security is generally higher, right? Right. So if you spend an hour messing around with a new chatbot, trying different prompts, and you end up generating a bunch of useless garbage, it's just considered professional development or exploring new efficiencies. Oh, totally.

[00:11:55] But if you're a frontline worker, constantly monitored for metrics, time off task is heavily penalized. Exactly. You don't have the slack in your schedule to experiment. And worse, you don't have the psychological safety to make a mistake. What if the AI hallucinates a policy and you send it to a customer? You are the one getting fired, not the software vendor. Precisely. That dynamic of risk aversion explains so much of the demographic divide in usage.

[00:12:21] And Dr. Westover highlights a specific demographic gap that is absolutely wild to me. It's not just about corporate rank. There is a massive gender divide. The 2025 Daniel study by Homelow Investor Guard is incredibly revealing on this front. Yeah. Break that down for us. So they looked at roughly 18,000 workers across 11 different occupations that are highly exposed to generative AI. And because it's Denmark, they were able to link the survey responses directly to rich administrative records. Right. They have great data over there.

[00:12:50] They do. And they found that women were 16 percentage points less likely than men to use CHAD-GPT for their work tasks. And just to be clear for everyone listening, this isn't just a quirky anomaly in one Scandinavian country, right? Not at all. The paper cites a massive global synthesis by Otis and colleagues. Oh, the 2025 one? Yes. It encompassed 18 separate studies and over 140,000 people worldwide. Wow.

[00:13:15] And that persistent gender gap in generative AI usage holds true across almost all regions, all sectors and all occupations. That is staggering. And the most critical finding from that global data is that this gap persists even when you completely equalize access. Wait, really? Yeah. Even when the man and the woman are sitting at the exact same level in the same department with the exact same software license, the usage gap remains. Okay. So how do we explain that?

[00:13:43] Because it can't just be a blanket statement that women are inherently less interested in technology. That makes literally no sense. Right. It's not that at all. Sociologists point to the exact same dynamics we just discussed regarding psychological safety and risk at work. In many corporate environments, women are historically penalized more harshly for mistakes than their male counterparts. Right. There is way less margin for error. Exactly.

[00:14:07] If the technology is known to hallucinate or produce biased outputs, the reluctance to adopt it isn't a lack of technical capability. It is a highly rational, self-protective calculation. That makes total sense. Furthermore, women disproportionately shoulder invisible labor in the workplace. You know, the mentoring, the cultural cohesion, the emotional intelligence tasks. Which are precisely the kinds of tasks that AI is currently terrible at handling. Exactly.

[00:14:34] So the tool doesn't actually fit the unmeasured demands of their role. And using it incorrectly carries a higher social and professional penalty. It makes total sense why the usage drops off. It really does. Okay. So we've navigated access and use. If an employee actually makes it through that gauntlet, so they have the license, they get the five hours of paid training, they feel safe enough to experiment, we arrive at stage three. Benefit. Benefit. Does this technology actually make the day-to-day experience of doing a job better?

[00:15:03] This is where the data finally offers some brilliant optimism. Oh, thank goodness. Right. Among the people who are frequent users, the self-reporting gains in the Gallup study are astronomical. We are looking at 63% saying AI allows them to execute their tasks faster. That's huge. It gets better. 56% report that it actually helps them find more creative solutions to problems. Wow. And almost half say it improves the overall quality of their output and allows them to focus on the most engaging, interesting parts of their job.

[00:15:33] That is the utopian promise actually materializing. Like, faster work, higher quality, less boredom. And here is the crucial nuance. While college graduates and higher earners were more likely to say, AI made them work faster, the reported gains in creativity and quality were remarkably consistent across all education levels. Wait, so the quality boost is universal? Yes. That means a worker without a college degree finds the tool just as powerful for elevating the quality of their thought process as an executive does.

[00:16:03] Provided they are actually given the access and the training to use it. The ceiling of human potential is genuinely raised. Okay, so the tool works. The benefit is completely real. Which brings us to the final stage of Dr. Westover's pipeline. And honestly, the place where this entire narrative takes a deeply cynical turn. Yeah, stage four, reward. Right.

[00:16:25] Because in a rational world, if a worker figures out how to use a new tool to become 63% faster and significantly more creative, there should be a tangible reward for that. Like a promotion, a bonus, or maybe just the ability to log off at three on a Friday. That would make sense, yeah. But Dr. Westover identifies what he calls the reward leak. The reward leak is arguably the most concerning metric in the entire study.

[00:16:48] Nearly one third of all AI users, about 31%, report that their employer's response to their increased productivity is simply to assign them more work. They are literally experiencing expanded responsibilities explicitly because they are using AI. Exactly. They are being punished for their own efficiency. And if you're part of that 63% who self-reported that AI makes you work faster, you are even more likely to get hit with this.

[00:17:14] The data shows 37% of those high efficiency users are just being loaded up with more tasks. It honestly reminds me of being a kid and having to do yard work on a Saturday. Oh, I know where this is going. Right. So your parents tell you to rake all the leaves in the backyard. Normally it takes you three hours. But one weekend, you figure out this incredibly clever system using a tarp and a different rake, and you finish the whole yard perfectly in 45 minutes. You're a little efficiency expert. Exactly.

[00:17:40] So you walk inside completely thrilled, expecting to spend the rest of the afternoon playing video games. And your parents look at you and say, wow, you have two extra hours. Go clean the garage and wash the car. Yeah. And what lesson do you learn from that? You learn to rake the leaves as slowly as humanly possible next Saturday. You learn to completely hide your efficiency. And we are seeing that exact dynamic play out across the corporate world right now. Which is terrifying for innovation. It is. The demographic breakdown of this job intensification is alarming.

[00:18:09] 38% of the youngest workers in the study, those aged 18 to 24, report that AI has expanded their responsibilities. Oh, wow. Think about the long-term impact of that. We have an entire generation of early career professionals whose very first experience with generative AI is that it's a relentless, unyielding engine for burnout. They're being trained to associate this miraculous technology purely with exhaustion.

[00:18:35] If we connect this to the bigger picture, if organizations do not fix this reward leak, they're going to face massive, quiet resistance. Workers will simply stop innovating. Let's pull back and look at the sheer irony of everything we've discussed so far. Companies are reserving AI licenses for their top performers and managers. Right. The frontline workers who could benefit most don't get it or they aren't trained on it. And the few people who do figure it out are just given more work until they burn out. It's a disaster.

[00:19:05] It really is. And this brings us to a massive, glaring question about business strategy. If we ignore the ethical implications for a minute and just look at cold, hard capitalism, is giving AI exclusively to your top performers even a good business decision? Like, does it actually generate the highest return on investment? The causal evidence strongly suggests that it is a catastrophic misallocation of capital. Giving AI to your best people actually provides the lowest possible return. Really? Yeah.

[00:19:35] Dr. Westover highlights a landmark 2025 study by Eric Brynjolfsson and his team to prove this. They studied the rollout of a generative AI conversational assistant to over 5,100 customer support agents. Okay. Over 5,000 agents. That's a good sample. Very good. Overall, the tool generated a highly respectable 14% increase in productivity across the board. Which, I mean, most executives would be absolutely thrilled with. Oh, they would be. But the aggregate number hides the real story here.

[00:20:05] When the researchers segmented the agents by skill level, the results were shocking. What did they find? The novice workers and the lower skilled agents saw a massive 34% increase in their productivity. 34%? Yeah. But the highly skilled, highly experienced top performers, the people who had been there for years, they saw almost zero measurable effect on their productivity. Wait, wait, wait. Almost zero? The top performers got no faster at all? Almost none.

[00:20:32] The entire 14% aggregate bump was driven by the massive, massive gains at the bottom of the skill distribution. Okay. But I can hear someone arguing that's just customer service, right? It's heavily scripted. It's repetitive work. Does that pattern actually hold up when we look at complex, ambiguous knowledge work? It absolutely holds up. The paper brings in another famous experiment from 2023 conducted by Delacqua and researchers. They looked at 758 consultants at the Boston Consulting Group. BCG consultants.

[00:21:01] So these are elite, highly educated knowledge workers doing complex strategy tasks. Exactly. The top of the top. They were given GPT-4 to assist with their workflows. And the below average performers improved their task performance by staggering 43%. 43% for the bottom half? Right. And the above average elite performers, they only improved by about 17%. The technology acts as a massive leveler. Okay. We need to understand the mechanics of why this happens.

[00:21:28] Like, why does a large language model turbocharge a novice but barely move the needle for an expert? It really comes down to understanding what generative AI actually excels at. It is, at its core, a delivery mechanism for tacit knowledge. Tacit knowledge. Okay, define that for us. Tacit knowledge is the unwritten, unspoken expertise that a top performer possesses. It's the instinct. Let's use a culinary analogy. I love food analogies. Go for it.

[00:21:55] Imagine a novice cook trying to make a perfect steak, and Gordon Ramsay standing right behind them. Gordon Ramsay isn't just handing them a recipe card. A recipe is explicit knowledge. Right. Anyone can read a recipe. Exactly. But Gordon is watching the smoke point of the oil. He's listening to the pitch of the sizzle. He's correcting the angle of the cook's wrist as they baste the butter. That is tacit knowledge. So how does an AI act like Gordon Ramsay in a corporate setting?

[00:22:22] Well, in the Brynjolfsson customer support study, the AI was trained on the chat logs of the absolute best, most effective agents. It analyzed their phrasing, their de-escalation techniques, their subtle ways of navigating complex technical issues. Ah, I see. So when a novice agent was struggling with a live customer, the AI would pop up on their screen and suggest a response that contained all of that embedded expertise. It was literally whispering the master's instincts into the novice's ear. Wow.

[00:22:51] So it allows a worker with two months of experience to perform like a worker with three years of experience. Exactly. But if you hand that exact same AI suggestion to the actual three-year veteran... They already know it. They're the ones who generated the data the AI learned from in the first place. Precisely. The AI is just suggesting the exact thing they were going to type anyway. Which brings us right back to the tragic irony of the Gallup data.

[00:23:13] Organizations are spending millions of dollars to buy advanced tools for the executives and senior managers, the very people who already possess the tacit knowledge and will gain the least from the software. Yeah. Meanwhile, they are actively withholding the tool from the entry-level and frontline workers who would see a 40% performance leap if they just had access to it. It is the ultimate organizational irony. Okay, I have to play devil's advocate here because here's where it gets really interesting.

[00:23:40] I can hear a chief operating officer listening to this right now and screaming, Yes, I know it boosts novices. But if I hand a deeply complex, occasionally hallucinating large language model to my least experienced workers, it's a liability nightmare. That is the fear. Yeah. Right. An expert with 10 years of experience knows when the AI is making up a fake legal precedent. A 22-year-old novice is just going to copy paste the hallucination and send it to a client. And that COO is voicing a very legitimate existential fear.

[00:24:10] This is where the concept of the jagged technological frontier comes into play, which was explored in that same BCG consultant study. The jagged frontier. What does that mean? AI is not uniformly intelligent. It is brilliantly capable of doing highly complex things like writing Python code, but it will confidently fail at basic arithmetic or logical reasoning. The border between what it can do and what it cannot do is invisible and highly jagged.

[00:24:38] Give me a concrete example of how an elite worker falls off that jagged frontier. In the BCG experiment, the researchers deliberately gave the consultants a specific business case study that fell just outside the AI's capabilities. It required subtle counterintuitive logic that the AI's training data would lead it to get completely wrong. Oh, that is a devious experiment. It was.

[00:25:01] And the AI generated a beautifully formatted, highly persuasive, brilliantly written analysis that was fundamentally incorrect. Oh no. And the consultants who used the AI for this task were 19 percentage points less likely to get the right answer than the consultants who were forced to do it manually without the AI. Because the AI output looked so polished, the experts just turned off their brains and trusted it? They outsourced their judgment. They fell asleep at the wheel because the machine was so articulate.

[00:25:31] This is the ultimate paradox. You cannot just automate the workflow and remove the humans. Because you need the human's domain knowledge to catch the machine when it crosses the jagged frontier. Right. But if you only rely on your few elite experts to monitor the AI, they become a massive bottleneck. The only sustainable path forward is broad distribution coupled with intense, continuous training on critical thinking. This feels like the perfect pivot point.

[00:25:57] We have spent the first half of this deep dive thoroughly diagnosing a very sick patient. Very sick indeed. We've seen the broken promises, the demographic disparities, the punishing job intensification, and the completely bizarre strategy of withholding the tool from the people it helps most. Yeah, it's bleak. But Dr. Westover doesn't just leave us in despair. The entire back half of his paper is a pragmatic, evidence-based playbook for leaders to actually fix this. He outlines five specific responses.

[00:26:26] Let's start with the top of the funnel, addressing the access and enablement pipeline. So the first response is structural, designed for broad access. This means completely abandoning the outdated IT mindset of running small exclusionary pilots where only a handful of chosen managers get to play with the new toy. Right. It can't just be a perk. Exactly. You have to treat AI as core infrastructure. Dr. Westover points to JPMorgan Chase as a masterclass in this approach.

[00:26:55] JPMorgan built their own internal platform, right? The LLM suite? Yes. And by mid-2025, they had deployed it to over 200,000 employees. They didn't gatekeep it. They didn't decide that only investment bankers got it while the retail branch staff didn't. They treated it like email or a word processor. A fundamental utility required to do modern work. But building for broad access also means configuring the tool for the actual realities of frontline work. What do you mean by that?

[00:27:24] Well, a chatbot interface that requires typing long, complicated prompts is useless to a maintenance technician wearing thick gloves on a factory floor. Oh, right. Of course. You have to design interfaces like voice-to-text, specialized models that seamlessly integrate into the physical environment of the worker. Which transitions directly into the second response, targeted enablement. As we discussed with the BCG data, dumping software on a worker's computer without training is totally useless.

[00:27:52] Enablement means explicitly funding the time to learn. You have to pay people to get good at this. The primary example he uses here is Walmart. Okay. What did Walmart do? In late 2025, Walmart partnered with OpenAI to create a formal certification program. But the genius of the initiative wasn't just the curriculum. It was the audience. They explicitly targeted frontline retail associates and they integrated it into their existing paid education benefits.

[00:28:18] So they were paying their hourly workers to build AI literacy on company time. Exactly. I want to use a musical analogy here. Giving out access without training is like dropping a grand piano into the center of a company break room. Okay. I love this. Sure, you've provided access to music, but the only people who are actually going to sit down and play it are the people whose parents could afford to give them piano lessons for 10 years when they were kids. Yep. The privilege is baked in. Right.

[00:28:45] So if you want the whole company to make music, you have to pay for the piano lessons during work hours. That is a brilliant way to conceptualize it. Enablement bridges the gap between theoretical availability and actual capability. Dr. Westover also highlights a fascinating partnership between Microsoft and the AFL-CIO, you know, the massive labor federation. Oh, what did they do together? They built AI curricula specifically tailored to the needs of organized labor and union leaders.

[00:29:13] But there is a crucial tactical warning here regarding how you actually offer this training. Right. The targeted outreach component. Exactly. Think back to the global data showing those persistent gender gaps or the reluctance of frontline workers to experiment. If corporate HR just sends a mass email saying optional AI workshop on Friday at 4 p.m., who actually shows up? The only people who show up are the ones who already feel secure, confident and have slack in their schedules. Exactly.

[00:29:40] An open invitation usually just reinforces existing inequalities. So how do you actually reach the under-users? You utilize peer networks. You identify respected influential workers on the frontline and train them to be peer coaches. Oh, that's smart. Yeah. A warehouse worker is infinitely more likely to adopt a new AI scheduling tool if the person teaching them is the veteran shift supervisor they've known for five years, rather than a 25-year-old IT manager beaming in on a Zoom call from corporate headquarters.

[00:30:09] The messenger matters just as much as the curriculum. Completely. Okay. So let's imagine a company gets all of this right. They deploy it enterprise-wide like J.P. Morgan. They pay for extensive peer-led training like Walmart. Everyone has access and everyone knows how to prompt the model. Are we done? Is the AI dividend gap closed? Not by a long shot. Not by a long shot.

[00:30:30] Because you can have all the training in the world, but if the workforce fundamentally believes that this technology is being wielded against them, like to track their bathroom breaks and eventually eliminate their jobs, they will find a million invisible ways to sabotage its implementation. Yeah. That makes sense. The Gallup data showed that 52% of workers want more influence over how technology is adopted in their workplace. That brings us to response three. Build worker voice into technology decisions.

[00:30:58] This is where we shift from software procurement to actual power dynamics. What does worker voice actually look like when you're buying AI? It looks like participatory design. It means bringing the end users into the room to define the operational problem before management signs a vendor contract. Do we have a good example of this? The flagship example in the paper comes from the healthcare sector. The 2025 collective bargaining agreement between Kaiser Permanente and the Alliance of Healthcare Unions. Okay. What did they do differently?

[00:31:28] They didn't just hash out wages. They created a dedicated joint labor management AI subgroup, and they spent over 150 hours in deep deliberations solely focused on artificial intelligence. Let's really emphasize that. 150 hours of committee meetings about software deployment. I can practically hear the collective groan of startup founders and venture capitalists listening to this. Oh, I know they hate it. Because the prevailing ethos of Silicon Valley is move fast and break things.

[00:31:56] If you have to sit through 150 hours of union deliberations just to roll out a scheduling algorithm, isn't that just a recipe for bureaucratic paralysis? How can a company remain competitive if it takes them six months to deploy what a competitor deploys in a weekend? This raises an important question because it's the most common managerial counterargument. But Dr. Westover leans on the classic economic theory of Freeman and Madoff regarding the voice face of worker organizations. The voice face. Yeah.

[00:32:25] When management moves fast and breaks things in an enterprise environment, what they usually break is the core workflow that actually keeps the business running. Right. Doing consultation theater where you buy a multi-million dollar tool, force everyone to use it, and then send out a survey asking for feedback, that just destroys trust. But genuine upfront work of voice surfaces critical operational data that management is completely blind to. Give me a scenario where management is blind to the reality of the work.

[00:32:53] Imagine a hospital administration wants to buy an AI tool that automatically drafts discharge summaries for patients to save the nurses time. Management looks at the demo, sees it takes 30 seconds instead of 15 minutes, and they just love it. Looks great on a spreadsheet. Exactly. But if they bring a veteran floor nurse into the room before buying it, that nurse will say, Look, drafting the summary is actually when I mentally review the patient's entire stay and catch medication discrepancies. Oh, wow.

[00:33:23] Right. So if you automate the drafting, I lose my cognitive check and our medication error rate is going to spike. Management cannot know that because management doesn't administer the medication. So the 150 hours of deliberation isn't about slowing down innovation. It's about preventing a catastrophic million dollar mistake that harms patients. Exactly. Worker voice acts as the ultimate quality assurance mechanism. Dr. Westover also points to Germany's Works Council Modernization Act. How does that work?

[00:33:52] Under this law, if an employer wants to introduce AI, the Works Council has the legal right to hire an independent external AI expert to evaluate the system on their behalf. That is brilliant. The principle is that giving workers a seat at the table is meaningless if they don't have the technical expertise to actually interrogate the algorithms. You have to arm them with independent data. Which flows perfectly into response four. Bargaining over how gains and risks are shared.

[00:34:18] Once the workers are at the table, what are they actually negotiating for? We've seen some massive high profile examples of this recently. The most visible was certainly the Writers Guild of America strike in 2023. Oh yeah, that was huge. The writers didn't just ask for more money. They established hard existential guardrails around generative AI. They secured contract language explicitly stating that AI is not a writer. That AI generated output cannot be classified as literary material.

[00:34:45] And crucially that a studio cannot force a writer to use AI against their will. They protected the human definition of the craft. And then you look at SAG-AFTRA, the actors union. They were fighting over the use of digital replicas, right? Yes. They secured the requirement for informed consent before a studio can clone an actor's voice or face. And they established that actors must be compensated for every new performance generated by that digital clone. They directly negotiated a mechanism to capture their share of the AI dividend.

[00:35:15] Exactly. And in the service sector, you have the Culinary Workers Union in Las Vegas. They secured a contract stipulating that the massive casino resorts must provide six months of advance notice before introducing new AI or robotics that could affect jobs. Six months is a good amount of buffer. It is. And if jobs are automated, the company is on the hook to provide extensive retraining for those specific workers. Notice, consent, guardrails, and retraining. That's a solid framework.

[00:35:45] And the vital point Dr. Westover makes is that you do not need to be a unionized company to implement these frameworks. Progressive non-union organizations should voluntarily adopt these policies as internal governance, because it is the absolute only way to build the trust required for smooth adoption. We have finally arrived at the final stage of the journey, the culmination of the entire paper. Response five, reinvest the dividend in better jobs. We've talked about access, enablement and voice.

[00:36:14] Now we have to talk about the reward. The most important part. Let's assume the system works perfectly. The workers were consulted, the tool was deployed safely, and it actually ends up saving the workforce millions of hours of tedious labor. What does the corporation actually do with that massive surplus of time? Dr. Westover gives us two incredibly vivid, contrasting case studies here. Let's start with the gold standard, IKEA. The IKEA case study is just a master class in how to manage technological dividends.

[00:36:44] A few years ago, IKEA deployed an AI customer service bot named Billy. The bot was remarkably successful. How successful? It was so good that it ended up successfully handling 47% of the cost of money. The company's largest customer service bot named Billy. Almost half of the entire contact volume just vanishes overnight. Let's be honest, the instinct for 99% of chief financial officers in that exact moment is to look at a spreadsheet and say, incredible, fire 47% of our call center staff.

[00:37:13] Our margins will be spectacular this quarter. That is the standard reflex, yes. But IKEA took a completely different path. Instead of capturing that dividend as a reduction in headcount, they decided to redeploy and elevate that human capital. What did they do? They took 8,500 of those contact center workers, people who previously spent their days answering basic questions about missing screws or store hours, and put them through an extensive retraining program.

[00:37:39] They turned those 8,500 workers into remote interior design advisors. They moved them from a reactive cost center to a proactive revenue generator. Exactly. When customers were browsing online, they could connect with a human expert who deeply understood the entire IKEA catalog to help them design a kitchen or living room. And what was the financial result of that, of treating human beings as an asset to be elevated rather than a cost to be cut?

[00:38:04] Those remote interior design sales generated 1.3 billion euros in revenue in 2022. That is the ultimate realization of the AI dividend. The company unlocks a massive new billion dollar revenue stream and 8,500 human beings get upgraded from a repetitive low agency call center job to a creative, consultative, higher quality career path. Everyone wins. It's beautiful. But we have to talk about the dark mirror to this story.

[00:38:31] We have to talk about the cautionary tale of falling into the head count trap. Let's look at Klarna. Klarna is a major financial technology company, and their approach was the exact inverse of IKEA's. In early 2024, Klarna's leadership went on a massive PR blitz, loudly bragging that their new AI assistant was now doing the equivalent work of 700 full-time human customer service agents. 700 agents. Right.

[00:38:56] Based on this AI capability, they instituted a hiring freeze and allowed their overall head count to shrink significantly through natural attrition. They basically presented themselves as the vanguard of the new ultra lean, AI-driven corporation. I mean, it sounds like a Wall Street dream, but what actually happened to the business? Well, reality caught up with them. About a year later, the CEO had to give an interview to Bloomberg, admitting that the company was actively scrambling to hire human customer service staff again. Why? What broke?

[00:39:26] Because while the AI was great at basic queries, it completely collapsed when dealing with complex, emotionally charged, or highly nuanced financial disputes. The overall quality of customer service just plummeted. Wait times for complex issues skyrocketed. Customers were furious. They crashed straight into the jagged technological frontier. They assumed the AI was universally competent. They removed the human safety net to save a few bucks, and the machine hallucinated or failed when things got complicated.

[00:39:55] They captured the AI dividend entirely as short-term payroll savings. And in doing so, they eroded the very core of their service quality. The lesson from Dr. Westover is clear. Using AI to strip-mine your organization of human judgment is a path to systemic fragility. So what does this all mean for the person listening right now? If you are a business leader, a manager, or just a professional trying to navigate this landscape, how do you make sure your company behaves like IKEA and avoids becoming the next Klarna?

[00:40:25] How do you tangibly close this gap? Dr. Westover concludes by insisting that leaders must structurally tie AI adoption to job architecture from day one. You cannot just deploy the tool and see what happens. You must proactively decide where the saved time is going to go. Right. You need a plan for the time. If AI is going to automate 30% of a marketing team's workflow, leadership must actively design the new, higher-level analytical tasks those marketers will pivot to.

[00:40:54] If you don't design the new work, the time saved just defaults to the reward leak. Dumping more of the same grueling tasks on the most efficient workers until they break. Exactly. You have to build the escalation routes for the human in the loop. Yes. Furthermore, you have to fundamentally change how you measure success. Most companies measure AI adoption by looking at aggregate metrics. You know, total active users or the total number of API calls made to the server this month. That is useless. So what should they measure?

[00:41:22] You must measure the dividend by demographic segment. You need dashboards showing access, training hours, and self-reported job quality broken down by role level, by education, by age, and by gender. That makes total sense. If your dashboard shows that only your executives with master's degrees are experiencing productivity gains, you are failing. You have to prove with data that the benefits are actually reaching the front line. We have covered immense ground today.

[00:41:49] We started by looking at the towering expectations we all had for generative AI, that it would be the great equalizer, democratizing knowledge across the workforce. Yeah, we had such high hopes. But Dr. Westover's dive into the 2026 data revealed a much harsher reality. Right now, AI is trapped in a loop of compounding privilege. It is flowing to the people with the highest education, the most security, and the largest salaries.

[00:42:14] It's bypassing the front line workers who mathematically stand to gain the largest performance boosts from it. And for the young workers who do use it efficiently, they are often punished with job intensification. Yet the roadmap to fix it is entirely within our grasp. It requires treating AI as universal infrastructure, paying for peer-led training, embedding worker voice deeply into the procurement process, and committing to elevating human roles rather than just eliminating head count.

[00:42:42] The technology is genuinely capable of creating a renaissance in job quality. The failure is entirely a failure of imagination and management. Absolutely. And as we wrap up, I want to leave you with one final lingering thought to chew on. We spent a lot of time today discussing how AI acts as a delivery mechanism for tacit knowledge, how it transfers the hard-earned instincts of an elite expert directly into the hands of a novice, dramatically shrinking the performance gap between the two. Right, the Gordon Ramsay effect. Exactly.

[00:43:12] So if we actually succeed in closing the access gap, if managing and directing an AI assistant becomes the new basic universal literacy of the modern workplace, what happens to the entire concept of unskilled labor? It's a profound sociological question. If the machine holds the technical execution, the value of the human shifts entirely toward judgment, empathy, adaptability, and real-world problem solving. Exactly.

[00:43:38] The revolutionary shoes might allow anyone to run a four-minute mile without breaking a sweat, but you still need a human being to look at the map, evaluate the terrain, and decide which direction the race should actually go. Well said. Thank you so much for joining us on this deep dive. We highly encourage you to seek out and read Dr. Jonathan Westover's complete paper, The AI Dividend Gap in the Human Capital Leadership Review. More importantly, we hope you take this framework and look critically at your own workplace.

[00:44:07] Ask the hard questions. Who is getting the tools? Who is getting the training? And who is actually deciding where the save time goes? Until next time. I'll see you next time.