A Conversation about Work Design and AI Enterprise Value

[00:00:00] Right now, I mean, companies are just pouring literally billions of dollars into artificial intelligence. Billions. It's wild. It is. Like, if you're sitting in a corporate office today, you are probably just bombarded by, you know, announcements about new enterprise AI licenses or generative tools. Right. The new digital assistant that's going to, you know, revolutionize everything you do. Exactly. It supposedly revolutionized the way you work. And I mean, almost 90 percent of global companies are using it daily at this point.

[00:00:29] Yep. Ubiquitous adoption. But here is the crazy part. Almost half of the CEOs who are actually authorizing these massive checks are walking into their boardrooms completely empty handed, like zero financial return to show for it. None. Why is this? I mean, this is the greatest technological leap of our generation, right? Why is it seemingly just setting corporate capital on fire? Well, I mean, that is literally the defining management puzzle of 2026. Right.

[00:00:59] The disconnect between the technological capability of these systems and the actual, you know, measurable impact on the bottom line. Staggering. It really is. We're looking at this landscape where, like I said, ubiquitous adoption has somehow resulted in phantom returns. And that's exactly what we're going to unpack in this deep dive today. So for you listening, we're looking at this incredibly insightful 2026 paper by Dr. Jonathan H. Westover. Right. From the Human Capital Leadership Review.

[00:01:28] Yeah, that's the one. It's titled The Redesign Dividend. Why most organizations still aren't getting value from AI and how leaders can redesign work for a human agent workforce. It's a long title, but it really nails the issue. It does. And our mission here is to figure out what actually separates the companies that are succeeding with this technology from the ones that are just, you know, buying expensive software and hoping for a miracle. Because, spoiler alert, hoping for a miracle is a terrible business strategy.

[00:01:55] It really is. Especially when you look at the actual data. Oh, the statistics are just sobering. Like Gartner did this 2027 CIO and Technology Executive Agenda survey, right? Okay, yeah. They looked at over 2,000 tech executives. And out of that massive pool, only 13% reported that their enterprise had realized significant or, you know, transformational value from their AI investments. Wait, only 13%? Yeah, 13%.

[00:02:23] Which means 87% of companies are either seeing like marginal incremental tweaks or just absolutely nothing at all. Wow. And it gets worse. They estimate the likelihood of any individual AI use case actually succeeding is roughly one in five. A 20% success rate. I mean, if you're a venture capitalist, you might accept those odds, right? Sure. Yeah. In VC, that's fine.

[00:02:47] But if you're a CIO trying to run a stable enterprise, a 20% success rate on mission critical software deployments is, well, it's a fast track to getting fired. It absolutely is. And it's not just the CIOs feeling the heat either. Gartner also found that 44% of CEOs are struggling to demonstrate any AI driven financial returns. So almost half of the CEOs. Yeah. And McKinsey's global survey from the same year reinforces this perfectly.

[00:03:13] They surveyed over 1,700 respondents across like 100 countries. Okay. Almost everyone has integrated AI into at least one business function. Yet only 37% can attribute any EBIT impact, any earnings before interest in taxes to the technology. So they're using it, but it's not making them any money. Exactly. And if you filter for what McKinsey calls high performers, those attributing at least 5% of their EBIT to AI, you're looking at a mere 6% of companies.

[00:03:42] That's, that's insane. So we really have to ask why. Right. Because if the technology is inherently powerful and we know it is, right? We've seen it write code. We've seen it draft legal briefs, analyze data. Well, the raw capability is amazing. Yeah. So why is it failing to translate into corporate value? You know, the paper argues that leaders are treating this like traditional software, like, I don't know, an upgrade to a spreadsheet application or something. Which is a huge mistake. Right.

[00:04:08] Because we are undergoing this transition from generative AI to agentic AI. Can you, can you break down the mechanics of why that distinction matters so much to the bottom line? Yeah, absolutely. So generative AI, which kind of dominated the conversation for the last few years, it basically operates as a sophisticated content producer. Right. Like writing a poem or something. Yeah.

[00:04:29] You give it a prompt and it generates text or a block of code or an image, but it requires a human to initiate the action, evaluate the immediate output, and then, you know, manually plug that output into the next step of whatever workflow they're doing. Okay. So it's a tool. I use the tool, then I take the result and go do something else with it. Exactly. But agentic AI, which is what all these enterprises are desperately attempting to scale right now, is fundamentally different. How so?

[00:04:56] It's designed to plan and execute sequences of actions across different systems to achieve a broader goal. I see. So instead of me asking a bot to like draft a polite email apologizing for a delayed shipment, an agentic system is actually reading the customer support ticket itself. Right. It reads the ticket. And then it's querying the internal inventory database to find the delayed item, identifying the new shipping date, drafting the resolution, emailing the customer, and then logging the whole interaction and closing the ticket in the CRM. Precisely.

[00:05:26] It's doing the whole chain. And it does this with, you know, varying degrees of human oversight. But the reason the ROI is failing so hard is that you cannot just drop an agentic system into a workflow that was designed for human to human handoffs and expect it to function smoothly. It's like, the visual I keep coming back to is buying a state of the art Formula One racing engine and just bolting it directly onto the front of a wooden horse-drawn carriage. That's a great analogy. Right?

[00:05:54] Like, you have all this power, thousands of horsepower. The capability is undeniable. But you haven't redesigned the chassis. Nope. You still have wooden wheels. Exactly. A carriage built to move at 10 miles an hour. When you hit the gas, you don't win a race. The sheer torque of the engine just violently rips the wooden carriage apart. You shake the whole organization to pieces. Yeah, because the structural integrity of the carriage wasn't designed for the power of that engine. Right.

[00:06:19] And Dr. Westover identifies five specific drivers causing this massive failure rate. Basically, five ways organizations are essentially throttling the engine while flat out refusing to upgrade the carriage. Okay, let's get into those. What's the first one? The first driver is task substitution without workflow change. Okay. This happens when a company inserts AI into a single isolated step of an existing process, but leaves the entire surrounding environment completely untouched.

[00:06:46] Okay, let's follow the thread on that because I actually want to challenge the premise a little bit. Sure. Because, okay, if I have an employee who normally spends 45 minutes researching and drafting a weekly status report. Right. And an AI tool allows them to do it in five minutes, I've just saved 40 minutes of labor. Mathematically, yes. Right. From a purely mathematical standpoint, that is a massive productivity gain. So, why does the paper argue that this doesn't translate into actual organizational value?

[00:07:16] Like, where do those 40 minutes actually go? Well, that is the illusion of localized productivity. Okay. Think about it. If the intake process for that employee's work, the approval chain above them, and the overall staffing model of their department all remain totally static. Right. Those saved 40 minutes do not hit the P&L statement. They just vanish into the structural slack of the organization. So, wait, you mean the employee just takes a longer lunch? I mean, sometimes yes. Or they read the news.

[00:07:46] But more often than not, Parkinson's law takes over. Oh, remind me what that is? It's the adage that work expands to fill the time allotted for its completion. Oh, right. Yeah. So, the employee gets pulled into an unnecessary meeting. Or, you know, they spend that extra 40 minutes endlessly polishing a different presentation that doesn't actually require polishing. Just making things look busy. Exactly.

[00:08:08] Unless leadership structurally redesigns the workflow to capture that newly freed capacity, maybe by increasing the volume of intake or redirecting that employee to a new, higher value, revenue-generating task. The company has paid for the AI license, but gained absolutely no systemic advantage. Wow.

[00:08:28] In fact, McKinsey's data shows that nearly three-quarters of their high performers have fundamentally redesigned their workflows, compared to only a quarter of the organizations that are struggling with ROI. That makes perfect sense. It's like you're buying the speed, but the surrounding traffic jam prevents you from actually moving any faster. Exactly. You're just idling faster. Idling faster. Okay, so what is the second driver? Pilot proliferation. Meaning too many pilots?

[00:08:57] Way too many. Because AI tools are relatively accessible now, and there is this immense political pressure within organizations for, like, middle managers to be seen as innovative or championing AI. Right. Everyone wants to be the AI guy. Yes. So, companies are launching test pilots at a reckless pace. Everyone wants a shiny pet project to present to the board.

[00:09:20] But wait, testing new things isn't inherently bad, is it? Like, a culture of experimentation is usually praised in business literature. Sure, but experimentation without discipline is just chaos. Fair. Gardner found that only 31% of CIOs have established guidelines for deciding when a pilot should be prevented from moving into production. Only 31%? Yeah. And it gets more alarming. Only 17% of criteria for abandoning initiatives that are already in production, but are clearly failing.

[00:09:49] So, they have a gas pedal but no brakes. Exactly. But if a project is failing, why wouldn't a manager just quietly kill it? Like, why keep funding a broken pilot? Because of a well-documented psychological phenomenon known as the escalation of commitment. Okay. It was originally researched by Barry Stahl back in 1976.

[00:10:09] Stahl basically demonstrated that decision makers have this powerful tendency to continue investing heavily in a failing course of action, particularly when they feel personally responsible for the initial decision. Because their ego is attached to the outcome. Oh, 100%. When an executive stands up at an all-hands meeting and publicly touts this flashy new AI pilot, their professional narrative becomes intertwined with its success. Right. They can't back down. They can't.

[00:10:36] So, when the data later shows the pilot is underperforming or, you know, creating massive workflow friction, the executive doesn't kill it. They rationalize. It just needs more data. Exactly. They tell the board, oh, the model just needs more fine-tuning. Or, we just haven't trained the staff properly yet. They throw more budget and engineering hours at a fundamentally flawed use case just to avoid admitting defeat. Which bleeds right into the third driver, right? Cost uncertainty. Yes. Because, I mean, software used to be beautifully predictable.

[00:11:05] You negotiate a contract for 1,000 seats of a CRM platform, you lock in a per-user license fee, and the CFO knows exactly what the expenditure will be for the next 36 months. Oh, the good old days. Yeah. Generative and agentic AI operate on a fundamentally different economic model. The costs are highly probabilistic. Meaning they change based on usage. Right. They're determined at runtime by compute usage, typically measured in tokens.

[00:11:31] So, every time an agent queries a database, reads a document, or generates a response, the meter is running. So, if you have this culture of pilot proliferation, dozens of pet projects running across the company, and you can't kill them because of executive ego. You are suddenly hit with massive, unpredictable cloud computing bills. Wow, yeah, that's a nightmare for a CFO. Which is exactly why 51% of CIOs now expect AI to actually increase the total cost of ownership of their technology stack. Increase it. Yes.

[00:12:00] They are paying more for the infrastructure while realizing less measurable value. And this dynamic creates an immense amount of panic in the C-suite, which leads directly to the fourth driver. The headcount first mindset. The ultimate corporate anxiety. Right. The board sees these spiraling compute costs, they see the millions spent on AI licenses, and they demand immediate financial justification. Yep. And the easiest, fastest way to show a return on a balance sheet is to fire people. It is.

[00:12:28] The pressure from Wall Street to use AI as an immediate payroll guillotine is just intense. McKinsey found that 39% of respondents fully expected AI to reduce their organization's headcount over the following year. Wow. Leaders literally billed their entire business case for AI adoption around the premise that they will be able to drastically slash labor costs. But, but the reality doesn't match the expectation, does it? Not at all.

[00:12:53] Only 14% of organizations actually reported that AI had contributed to an overall workforce decline in the past year. So 39% expected it, but only 14% actually did it. Right. That is a massive delta between the expectation of labor reduction and the reality on the ground. Why the discrepancy though? If the AI is ostensibly doing the work, why aren't the companies able to make the cuts they promised the board?

[00:13:18] Well, largely because of the jagged nature of the technology, which we will discuss deeply in a moment. But the larger strategic failure here is the opportunity cost. Oh, so? When leadership teams design their entire AI rollout around defensive cost cutting and layoffs that just never materialize, they miss the offensive play entirely. Oh, I see. They failed to look at how they could strategically redeploy their existing talent to higher value revenue generating work.

[00:13:46] They get obsessed with shrinking the pie to save a few dollars rather than using their new human agent capacity to make a massively larger pie. That requires a total paradigm shift. Going from how many people can I replace to what can my people achieve now that they aren't bogged down in administrative heavy lifting? Exactly. But making that shift requires leadership that actually understands how to design work. Which brings us to the fifth and final driver, the readiness gap at the top. Right.

[00:14:16] The disconnect between vision and execution here is staggering. In Gartner's CEO survey, 77% of chief executives expect AI to have a high or medium impact on their strategic objectives. Okay, so they know it's important. Mm-hmm. And 63% of them verbally recognize that radically redesigning their operating model is necessary to achieve this. So the CEOs theoretically understand that they need to rebuild the carriage to handle the Formula One engine. Theoretically.

[00:14:45] But practically, only 20% of those CEOs report feeling adequately prepared to actually execute that redesign. Wow. And when you look down the org chart to the CIOs, you know, the executives tasked with the actual implementation, the numbers just fall off a cliff. How bad is it? Only 4% of CIOs report having explicit, documented plans to redesign employee roles. Wait, 4%? 4%. 4%.

[00:15:12] So 96% of organizations are just deploying these incredibly disruptive agents into the wild and hoping the workforce organically figures out how to adapt without breaking the company? They are relying on hope as a strategy. That's terrifying. It is. But here's the thing. This specific failure, you know, deploying transformative technology while completely ignoring the social and structural realities of the human workforce, it's not unique to AI. We have made this exact mistake before. Right. You were telling me about this.

[00:15:42] Yeah. If we look at the history of industrial automation, we find a perfect chilling mirror to our current crisis. I really want to dig into that because the historical precedents are just fascinating. It kind of proves that the fundamental nature of work and human psychology doesn't change, even if the tools do. Take us back to the Tavistock Institute study. Okay. So we have to go back to 1951. 1951. Okay. Researchers Eric Trist and Ken Bamforth conducted this landmark study on the British coal mining industry.

[00:16:10] The industry was attempting to modernize, right, moving away from manual extraction to a highly mechanized process known as the long wall method. The long wall method. Mm-hmm. So this was like heavy machinery, conveyor belts, automated cutters? Exactly. The cutting edge technology of the mid 20th century designed to drastically increase output per miner. Makes sense. And it was a mechanical triumph. But Trist and Bamforth discovered a massive problem.

[00:16:38] See, before the long wall method, miners worked in small, tight knit, autonomous teams. Okay. They self-organized, they communicated constantly, and they relied on each other intimately for both productivity and physical safety in a highly, highly dangerous environment. Right. Because if someone messes up, everybody's in danger. Exactly. They had high task identity and deep social cohesion. So what did the machinery do to that social structure? It obliterated it. Just wiped it out. Yeah.

[00:17:07] The long wall method required the work to be fragmented into narrow, repetitive, highly isolated tasks spread across different shifts. So a miner might spend their entire shift just clearing debris in isolation, never even seeing the miners who operated the cutter on the previous shift. Oh, wow. The technology mechanized the physical work but completely fractured the social organization of the mine. And what was the business result? Well, because if the machinery was faster, output should have skyrocketed, right? Uh.

[00:17:36] You'd think so, but the exact opposite happened. Morale plummeted, absenteeism spiked to unprecedented levels, workers experienced severe psychological stress, and overall productivity was just catastrophically bad. Wow. So the tech made things worse. Yes. Because the mechanization deployed without any regard for redesigning the social structure actually destroyed the efficiency of the mind.

[00:18:01] Because they treated the human beings as just another interchangeable part of the machine rather than, you know, a complex social network. Exactly. And this realization birthed a foundational concept in organizational psychology called sociotechnical systems theory. Sociotechnical systems. Right. It posits that an organization is comprised of two intertwined systems. The technical system, the tools, software, and equipment, and the social system, the human roles, relationships, and norms.

[00:18:30] And the core principle is that organizational performance depends entirely on the joint optimization of both systems. So if you hyper-optimize the technical system but ignore the social system… The whole structure collapses. Wow. Now, overlay that 1951 coal mine onto a 2026 corporate office. The environments look different, obviously. But the organizational dynamics are identical. Because people are still people. Exactly.

[00:18:56] If you fracture a modern knowledge worker's identity, strip away their human collaboration, and isolate them in front of a monitor with an AI agent executing all the complex, interesting parts of their job, you are recreating the long-wall method. You're just doing it with software instead of cool. Yes. You are isolating the worker, degrading their task identity, and you will inevitably trigger the exact same collapse in morale and productivity that Tristan Bamforth documented 75 years ago. The parallels are just undeniable.

[00:19:25] I mean, we are digitally isolating workers the exact way they physically isolated the miners. But surely there is a historical precedent for technology eventually paying off, right? Like the steam engine, electricity, desktop computers. They all eventually drove massive economic growth. They did. But they all followed a specific, often very painful trajectory.

[00:19:48] Economists Eric Bridgolson, Daniel Rock, and Chad Syverson actually formalized this in a 2021 paper detailing what they call the productivity J curve. The J curve. Yeah. They studied the historical adoption of general-purpose technologies and found that these innovations typically depress measured productivity in their early years, so the line on the graph dips down before it violently swings up into the shape of a J. Wait, I don't get that. If I replace my abacus with a calculator, my productivity goes up instantly.

[00:20:15] Why does a general-purpose technology cause a productivity dip? Because a general-purpose technology like electricity or AI is not a simple tool substitution. Okay. Think about it. When factories transitioned from steam power to electric motors in the early 20th century, they didn't just swap the power source. Steam engines required the entire factory to be built around a single massive central drive shaft with belts running everywhere. Right.

[00:20:41] To actually get the benefits of electricity, companies had to physically tear down their factories and rebuild them to accommodate individual electric motors at each workstation. Oh, which takes years of capital expenditure and, you know, halted production. Exactly. The J curve exists because organizations must invest massive amounts of time, capital, and attention into complementary intangible assets before the technology pays off. Like training and restructuring? Yes.

[00:21:09] They have to invent entirely new workflows, develop new training programs, and restructure their management hierarchies. During this investment phase, the old systems are disrupted and the new systems aren't fully operational yet, which causes a systemic drag on productivity. So leaders who demand instant ROI from AI are just, they're ignoring a century of economic history. Totally ignoring it. They are demanding the upward swing of the J without being willing to endure the dip. Okay.

[00:21:39] So we have to endure the dip and basically rebuild the factory. But let's look at the technology itself for a second. We tend to view AI as this universally brilliant oracle, right? Like if it can pass the bar exam, we assume it can do basic data entry perfectly. But that isn't the case, is it? It absolutely is not. And this is perhaps the most dangerous misconception in the enterprise today. We have to understand the concept of the jagged technological frontier. The jagged frontier.

[00:22:08] Explain the mechanics of this because it sounds like a trap. Oh, it is a massive trap. The term comes from a landmark 2023 field experiment conducted by researchers from Harvard Business School and the Boston Consulting Group. Okay. They took 758 highly skilled management consultants and divided them into groups. Some were given access to GPT-4 and others worked without it. The researchers then assigned them tasks that simulated their daily consulting work. Let me guess the baseline here.

[00:22:37] On standard tasks, the consultants with the AI vastly outperformed the control group. They decimated the control group. On tasks that fell inside the AI's actual capabilities, the consultants using the tool completed 12.2% more subtasks. They worked roughly 25% faster. And their final output was rated significantly higher in quality by independent human graders. Wow. I mean, that is the utopian vision every CEO wants.

[00:23:05] But the researchers were clever. They deliberately designed a second set of tasks that looked incredibly similar on the surface to the first set. Uh-oh. However, these new tasks were engineered to fall just outside the true reasoning capabilities of the AI model. Okay. Edge cases. Right. They were edge cases. Tasks that required a nuanced, highly specific type of logic that the model struggled with. Even though it could generate text that sounded highly plausible. So what happened when the consultants used AI on the edge case?

[00:23:35] Disaster. The consultants using the AI were markedly less likely to arrive at the correct answer than the consultants working completely unassisted. Wait, really? Yes. Less likely. Yes. Because the AI had been so brilliant on the previous tasks, the humans over-trusted it. The AI confidently presented a mathematically flawed or logically unsound answer. And the highly trained consultants simply accepted it and passed it along as their own work. It just lulled them into complacency. Exactly.

[00:24:04] AI capabilities do not form a clean, expanding circle where it is equally good at everything within a certain difficulty level. The frontier of its capability is jagged. Ah, jagged. I get it? It might be able to write complex Python code flawlessly but fail spectacularly at basic chronological reasoning or simple arithmetic embedded in a text prompt.

[00:24:26] If I'm a manager trying to redesign my team's workflow, how on earth do I know if a specific subtask is inside or outside that jagged frontier before I hand it over to an agent? Well, that is the defining operational challenge. If an organization just uniformly mandates AI use across a department without painstakingly mapping where the tool is reliable and where it hallucinates, they are making a dangerous trade.

[00:24:52] They are trading highly visible speed gains on the front end for invisible systemic errors deeply embedded in their data and customer interactions. And by the time those invisible errors compound and blow up a major client account, the damage is catastrophic. Exactly. Which leads perfectly into the final ghost of technology past you wanted to highlight. Because if we are trusting these systems and they have a jagged frontier of competence, we are relying on humans to monitor them and catch the errors.

[00:25:20] And that reliance brings us to Lizanne Bainbridge's 1983 paper, which introduced the ironies of automation. 1983. So we were studying automation long before large language models. What is the core irony Bainbridge identified? Now, Bainbridge pointed out a fundamental paradox in systems engineering. The goal of automation is usually to remove the human from the loop because, you know, humans are error probes. Right.

[00:25:46] But the more complex and highly automated a system becomes, the more critical the human supervisor's role becomes. When the automated system inevitably encounters a situation it wasn't programmed to handle. An edge case outside the jagged frontier. It fails. It crashes. And at that exact moment, the system relies entirely on the human supervisor to step in, diagnose the unprecedented error, and manually resolve it. So the human has to be the ultimate safety net. But here's the irony.

[00:26:15] Because the human is now just passively monitoring a screen instead of actively doing the work day in and day out, they lose the hands-on practice that builds and maintains their expertise. Oh. Their skills atrophy. Oh, that is a terrifying dynamic. It is. We expect the human to intervene in the most complex high-stakes failure states. But we have systematically stripped them of the daily practice required to actually understand the system.

[00:26:40] We are creating a workforce of passive monitors who are slowly forgetting how to actually do the jobs they are supposed to be supervising. Yes. Exactly. Diagnosing the illness here is vital, but it's really only half the battle. We know the carriage is shaking to pieces. We know it's driven by cost panics, jagged capability frontiers, and the fracturing of socio-technical systems. Now let's pivot to the cure. Let's do it.

[00:27:04] The paper outlines the redesign dividend, basically the five evidence-based responses from companies that are successfully navigating this transition. Let's start with solutions one and two, which deal directly with workflows and roles. Right. So solution number one is the most critical. In the example, organizations must redesign workflows end to end. As we established with the localized productivity trap, you cannot just speed up a single subtask. Right. You just create a traffic jam. Exactly.

[00:27:33] You have to map the entire life cycle of the work. You have to actively redesign how work is taken in, how it is approved, and how it is handed off between human teams and AI agents. I want to see what this looks like in practice. Give me an example of a company that structurally redesigned itself to do this. Okay. Look at Moderna, the biotech company. Okay. They realized that you cannot separate the design of the technology from the design of the human organization.

[00:27:59] So they did something radically disruptive to traditional corporate structures. They took their human resources department and their information technology department and completely merged them. Wait, merging HR and IT? Yep. In most corporations, those two departments barely speak the same language. I mean, IT views humans as security risks and HR views IT as a roadblock to buying the software they want. That sounds like a recipe for a massive internal turf war.

[00:28:28] Oh, it requires immense leadership. Tracy Franklin took over this combined entity as the chief people and digital technology officer. She explicitly stated that the goal was to eliminate siloed workforce planning on one side and isolated technology procurement on the other. Okay. Instead, they moved to a unified model she calls work planning. Work planning. That shifts the focus from managing people or managing servers to managing the actual flow of value.

[00:28:58] Exactly. Moderna had developed thousands of custom internal GPTs. By merging the departments, every time they looked at deploying a new agent, the people who understand human psychology, training, and change management were in the exact same room with the exact same budget as the software engineers. That's brilliant. They put the socio-technical system at the very center of their strategic planning. They aren't just buying tools. They are architecting the flow of the entire enterprise.

[00:29:26] That is a massive structural commitment. What about a company doing it at the workflow level? Morgan Stanley Wealth Management is a great example here. Okay. They rolled out an internal generative AI assistant for their financial advisors. It was highly successful in terms of adoption. They hit 98% usage. But they didn't stop at just providing a chat window. In 2024, they launched a highly integrated workflow tool called Debrief. How does Debrief function within the advisor's day?

[00:29:56] Like what does it actually do? Well, with explicit client consent, Debrief actively joins the advisor's client meetings. It listens, takes comprehensive notes, highlights specific action items discussed, drafts a customized follow-up email that the advisor can review and edit. And then, this is key, it automatically logs the entire interaction directly into the firm's CRM database. Wow. So it's completely embedded. It takes the most tedious administrative friction points out of the whole process.

[00:30:26] Exactly. One advisor actually reported saving 30 minutes of administrative work per meeting. Per meeting. Yeah. But here is the critical part. Morgan Stanley didn't use this tool to justify laying off financial advisors. Which is what everyone fears. Right. Instead, they explicitly framed Debrief's purpose as freeing up the advisors to spend significantly more time on actual client relationship building.

[00:30:52] They recognized that the deeply human element, the empathy, the trust building, is the core value of the wealth management role. They redesigned the workflow to eliminate the administrative burden so the human could hyper-focus on the tasks that only a human can do. Precisely. Which perfectly transitions into the second solution. Leaders must write explicit role redesign plans. Right.

[00:31:16] The Westover paper notes that organizations that actually document how roles will change are 2.4 times more likely to be vanguard performers. Yet, as we mentioned earlier, only 4% of CIOs are doing this. If you want to survive the jagged frontier, you literally have to write down exactly how the human's job description is going to evolve. The contrast between the companies doing this and the ones defaulting to layoffs is stark. Look at IKEA. The furniture company. Yeah.

[00:31:45] They deployed an AI customer service bot internally named Billy. Billy was highly capable. Over a two-year period, it learned the system so well that it was successfully handling 47% of all customer queries flowing into their global call centers. Almost half the call volume automated away. I mean, the reflex for 90% of executives would be to instantly fire half the call center staff and declare a massive win for the quarterly earnings report. But IKEA chose a different path.

[00:32:14] Beginning in 2021, they took roughly 8,500 call center workers whose jobs were being eaten by Billy, and they actively retrained them to become remote interior design advisors. Wait, really? That is a massive pivot, going from fielding complaints about missing hardware or tracking delayed shipments to actively consulting on living room aesthetics and upselling furniture collections. And look at the financial result of that role redesign.

[00:32:41] These new remote design services delivered by phone or video consultation by retrained call center staff brought in 1.3 billion euros in their 2022 financial year. 1.3 billion in new revenue. That accounted for roughly 3.3% of their total global revenue. Wow. And Inca Group, their parent company, saw the success and set a strategic goal to raise that share to 10% by 2028.

[00:33:07] When an executive was point blank asked by the press if AI would reduce their overall head count, they firmly said no. They viewed the chatbot not as a mechanism for cost cutting, but as a mechanism for capacity creation. Exactly. They deliberately redesigned those roles to fuel a massive, highly profitable new growth engine. Walmart is doing something similar on a massive scale, aren't they? Yeah. Walmart employs over 2 million people.

[00:33:33] They have publicly stated they expect their global head count to remain roughly flat over the next three years, even as AI fundamentally rewires the mix of jobs across the company. How so? Well, they are actively tracking which traditional functions are shrinking, and they are proactively creating entirely new career paths. They've created roles like agent builders. Agent builders. Yeah, domain experts who are trained to build and refine custom AI tools to help the merchandising teams.

[00:34:03] They are actively managing the transition of their human capital rather than just letting it erode. To understand the economic theory behind why IKEA and Walmart strategies work, we should look at the research of economists Darren Asimoglu and Pascual Restrepo. In 2019, they formalized how technology impacts labor markets. Yeah. Asimoglu and Restrepo argue that automation exerts two opposing forces on labor. The first is the displacement effect. This is the dynamic everyone fears.

[00:34:32] The machine attains the capability to take over specific tasks previously performed by humans, reducing the demand for human labor in that specific area. The chat bot replacing the call center agent. Exactly. But the second force is the reinstatement effect. This occurs when technology or the broader economic shifts caused by it creates entirely new tasks and new industries where human labor holds a comparative advantage.

[00:34:59] Like the creation of the remote interior designer role or the agent builder. Precisely. But here's the vital insight for business leaders. While macroeconomic forces might eventually drive reinstatement on a global scale over decades, an individual organization cannot afford to wait. They need it now. Right. Within the walls of a specific company, deliberate role redesign is the only way a leader can force reinstatement to happen in real time.

[00:35:25] If you displace the tasks without intentionally building the new tasks for the humans to shift into, you just get unemployment and a hollowed out company. And beyond the economics, we have to consider the psychological reality of the work itself. If we are redesigning jobs, how do we ensure we aren't just creating miserable jobs? We have to apply Hackman and Oldham's 1976 job characteristics model. It is the gold standard for understanding work motivation. They identified five core dimensions that make a job fulfilling.

[00:35:53] Skill variety, task identity, task significance, autonomy, and feedback. Let's focus on task identity, the feeling that you own a complete coherent piece of work from start to finish rather than just tightening a single bolt on a massive assembly line. Right. When AI absorbs the routine, predictable tasks in a workflow, the remaining human job can fracture in one of two directions. If leaders are intentional, the job can become richer.

[00:36:19] The human is elevated to focus on complex judgment, relationship building, and strategy, increasing their autonomy and task significance. But if leaders are negligent? The job becomes terribly thin. The human is reduced to a passive monitor, just staring at a screen waiting for the AI to hit an edge case, and then dealing exclusively with the frustrating exceptions the machine can't handle. That sounds awful. It is. Work design is not a soft HR initiative. It is a hard strategic capability.

[00:36:49] It dictates whether your post-AI workforce is highly motivated and driving growth, or alienated, bored, and desperate to quit. I actually want to push back on this idea of deliberate planning. Sure. We've talked about the bolt-on deployment pitfall where companies just buy the license and tell employees to figure it out. Let's say a company does that. Won't roles just naturally evolve anyway? Employees are smart. They will figure out how to use the tool.

[00:37:15] If role change is going to happen organically without a top-down plan, what is the actual harm? Like, what happens to the employee stuck in a role that is slowly eroding through organic attrition rather than deliberate design? We call that redesigned by attrition. And it is toxic to an organization. What happens is the employee's coherent job description slowly dissolves. The AI takes the easy 50% of their tasks.

[00:37:41] The employee is left holding a random, disconnected residue of complex edge cases and interpersonal friction points that the AI can't quite handle yet. So their job becomes a Frankenstein's monster of leftover chores. Yes. They no longer have a coherent task identity. Furthermore, their manager has no idea how to properly evaluate their performance anymore because the original job description is totally obsolete. Oh, that's a good point.

[00:38:07] The employee loses all intrinsic motivation because they are just plugging holes in a leaky digital system. Eventually, your most capable, ambitious people realize their career path has turned into a meaningless scrap heap of edge cases and they take their talents to a competitor that actually offers a coherent, meaningful role. Wow. Okay. To truly grasp the friction of this emerging human agent workforce, we don't have to rely entirely on theory.

[00:38:34] We can look at internal data from one of the most AI-intensive, bleeding edge workplaces on the planet. I want to spend some time on the Anthropic case study. Oh, this internal study is a fascinating, almost prophetic glimpse into the future of enterprise work. Tell us about it. So Anthropic, the highly prominent AI research company that created the Claude family of large language models, had their internal economics team conduct a rigorous study on their own workforce.

[00:39:00] They wanted to analyze exactly how work and employment dynamics changed at their company between 2025 and 2026 as their internal tools became exponentially more powerful. So you have a company building the absolute frontier of AI technology, turning the microscope on their own incredibly brilliant employees to see how they use it. Exactly. Let's start with the positives. What did the data show? The raw productivity gains were staggering, as you would expect.

[00:39:28] In a 2026 internal survey, 27% of Anthropic employees reported experiencing major paradigm shifting productivity gains. But the objective metrics for the engineering teams were even more striking. The median number of code commits per engineer rose to roughly five times their level just six months prior. A 5x increase in code output per human. That is the kind of metric that makes a CEO's heart skip a beat. It is a massive scaling of human leverage.

[00:39:56] Furthermore, 20% of the employees surveyed described a fundamental, qualitative shift in their daily work routines. They transitioned from manually doing tasks directly to orchestrating and managing multiple instances of CLAWD executing the tasks on their behalf. On the surface, that sounds like a technological utopia. You elevate from being the worker bee to being the general commanding an army of hyper-efficient digital workers. But there is a massive hidden cost to that shift.

[00:40:25] Where does the system break down? The breakdown occurs in the immense psychological and organizational strain this orchestration creates. Anthropic's internal calendar data revealed a shocking reality. What was it? Despite the 5x increase in individual output, the median employees spent 20% more time in meetings than they did a year earlier. For the technical individual contributors, the engineers actually writing the code, their meeting load increased by 16%.

[00:40:54] That feels entirely counterintuitive. If the AI is making them 5x faster at writing code, why are they spending 20% more of their day trapped in conference rooms? Because orchestrating a swarm of autonomous agents requires a massive, unprecedented level of human coordination. When an engineer is operating 5 agents simultaneously, the output of those agents intersects with the output of the agents being run by the engineer sitting next to them. Oh, I see.

[00:41:22] The systems conflict. The code bases merge unpredictably. And the humans have to constantly meet to untangle the chaos generated by their digital proxies. And the study found something fascinating about the AI's downtime, right? Yes. The study logged the activity of the clawed AI sessions being used by the employees. They found that the AI sessions were sitting idle roughly 80% of the time. Wait, the AI was just waiting? 80% of the time.

[00:41:51] 80% of the time. He was waiting for human input, human review, or human approval. The ultimate conclusion of the study was profound. In a highly advanced human agent workforce, machine capacity is abundant. It is practically infinite. Human attention and human coordination have become the absolute non-negotiable binding constraint on the productivity of the company. This perfectly aligns with an analogy I've been thinking about.

[00:42:20] Imagine you are a commercial airline pilot. Okay. For your entire career, your job has been to fly one single airplane. It requires intense focus. It is highly demanding. But you are physically sitting in the cockpit. You have your hands on the yoke. You can feel the vibration of the engines. You sense the turbulence and you know exactly what the machine is doing at every microsecond. You have ultimate task identity. You are operating a single, coherent, technical system. Exactly.

[00:42:49] But now with the shift to agentic AI, you are pulled out of the cockpit. You are placed in a glass tower. And instead of flying one plane, you are tasked with simultaneously orchestrating 50 autonomous cargo drones. The nature of the work has completely inverted. You are constantly scanning radar screens. You are frantically checking for edge case errors. You are issuing rapid fire course corrections to drone number 12 while trying to ensure drone number 47 doesn't crash into a mountain. That sounds stressful.

[00:43:18] And on top of that, you are constantly on the radio communicating with all the other air traffic controllers to make sure your 50 drones don't collide with their 50 drones. No wonder the engineers at Anthropic are stuck in meetings 20% more of the time. Right. The cognitive load of orchestrating a swarm of drones is exhausting in a completely different, much more frantic way than flying a single plane. That analogy perfectly illustrates the psychological strain.

[00:43:43] In occupational psychology, we use the job demands resources model developed by Bakker and Demiruti in 2007. It dictates that employee strain and burnout arise when job demands like extreme workload, frantic coordination, and the relentless mental attacks of monitoring agents vastly outpace the available job resources like autonomy, peer support, clear feedback, and skill development.

[00:44:09] And managing 50 drones spikes the demands while stripping away the tangible resource of hands-on practice. Exactly. Supervising autonomous AI agents creates extremely high cognitive load and attention fragmentation.

[00:44:22] If organizations just dump these agents onto their workforce without deliberately providing new resources, redesigning the workflows, and adjusting expectations around human capacity, they are guaranteeing massive workforce burnout, even if the raw output technically rises for a short period. But the Anthropic study highlighted another perhaps more existential problem for the broader economy. It wasn't just about the strain on the current engineers. It was about who they were hiring. Ah, the pipeline problem.

[00:44:51] This is the dark side of the reinstatement effect. Anthropic analyzed their hiring data, and they found that the share of junior employees among their new technical hires plummeted. Over a short period, it fell from 27% of new hires down to just 9%. Wow. The overall representation of junior staff and their technical workforce fell from 21% to 14%.

[00:45:15] Because the AI agents are now highly capable of performing the entry-level coding tasks, the basic bug fixes and the routine documentation that junior engineers traditionally cut their teeth on. From a short-term cost perspective, why pay a junior engineer a salary when an instance of Claude can do the work instantly for fractions of a penny? That is the short-term financial logic. But extrapolate that logic out five or ten years, across an entire industry. This creates a catastrophic talent pipeline crisis.

[00:45:44] If you systematically eliminate the entry-level roles because AI can perform them, where do your future senior experts come from? Right. You don't just birth a senior engineer with ten years of intuitive judgment, deep system architecture knowledge, and crisis management skills out of thin air. A senior engineer is just a junior engineer who survived a decade of making mistakes and learning from them. We circle back to Bainbridge's ironies of automation.

[00:46:08] True expertise is built almost exclusively through hands-on practice and gradual exposure to complexity. Now, it is true that emerging research shows AI can actually accelerate novice learning. It can help junior staff absorb tacit knowledge faster by acting as a highly personalized tutor. But that tutor dynamic only works if the novice is actually employed and inside the building. Exactly.

[00:46:33] If you stop hiring them to save a few dollars on the entry-level payroll, you are actively starving your own future talent pool. In a decade, when the complex age cases arise that the AI still can't handle, you will have no mid-level or senior human experts left to step in and fix the system. So if human attention is the ultimate bottleneck and we are dangerously close to starving our own talent pipelines, how do organizations course correct?

[00:47:01] How do they stop wasting their constrained human attention on doomed AI experiments and properly invest their newfound capacity? This brings us to the final set of strategies. Let's look at solutions three and four, installing stopping rules and reinvesting capacity. Solution three addresses the pilot proliferation we discussed earlier. To combat the escalation of commitment and executive ego, organizations must install rigid stopping rules.

[00:47:29] How does a company operationalize a stopping rule so it isn't just a vague concept? It requires extreme discipline before the project ever launches. You must define the expected business value and the specific metrics of success before any budget is approved. Simultaneously, you must set explicit, mathematically defined criteria for killing the project if it fails to hit those marks within a specific timeframe. You build the abort sequence into the launch code. Yes.

[00:47:58] But more importantly, there has to be a cultural shift. Leadership has to actively, publicly celebrate well-reasoned stopping decisions. Killing a bad pilot should not be viewed as a career-limiting failure for the manager who championed it. It should be praised and rewarded as an act of saving the company precious capital and, more importantly, human attention. Let's talk about a real-world example of what happens when a company lacks those stopping rules and lets the desire for cost-cutting override quality control.

[00:48:27] Klarna, the buy-now-pay-later fintech company. Oh, the Klarna customer service saga is a textbook case study in the dangers of delayed stopping decisions. Set the stage for us. In early 2024, Klarna generated massive global headlines. They aggressively deployed an AI assistant to handle their customer service, proudly announcing that the AI was doing the equivalent work of 700 full-time human agents.

[00:48:53] They paused all hiring and their overall headcount dropped by roughly 22%, largely through attrition and not replacing staff. Wall Street and tech evangelists lauded them as the vanguard of the AI revolution. The short-term optics were exactly what investors wanted to see. But the reality on the ground was deteriorating. Fast forward to May 2025. The CEO of Klarna, Sebastian Semyakovsky, had to reverse course in a very public way. What happened?

[00:49:24] He had to admit that while the AI system was undeniably cheaper and faster, it was producing a noticeably lower quality of service. The jagged frontier had caught up with them. Customers with complex, nuanced issues were trapped in algorithmic loops and were becoming deeply frustrated. So they hit the edge case and there were no humans left to act as the safety net. Exactly.

[00:49:46] To repair the damage, Klarna had to pivot aggressively and start actively hiring remote human customer service staff again, guaranteeing that a frustrated customer could always bypass the AI and reach a real person. So the initial deployment wasn't a total failure of the AI technology itself. It was a failure of governance, measurement and stopping rules. Precisely.

[00:50:06] If Klarna had defined a strict, non-negotiable quality baseline before they launched the agent and rigorously measured customer resolution outcomes alongside the immediate cost savings, they would have caught the degradation much earlier. A stopping rule would have forced them to adjust course, retain a buffer of human agents and avoid taking a public hit to the reputation. Which perfectly sets up the fourth solution from Dr. Westover's paper.

[00:50:31] When AI does successfully automate tasks, leaders must choose to reinvest that freed capacity rather than defaulting to immediate headcount cuts. This is where the truly vanguard companies separate themselves. When AI frees up 20% of a department's time, the default reflexive executive instinct is to cut the bottom 20% of the staff. But the companies realizing massive EBIT growth choose to reinvest. Look at Salesforce.

[00:50:58] In 2025, they went through a major restructuring and cut roughly 4,000 support jobs because their AI agents were successfully handling over half of all routine customer interactions. And that layoff is the headline everyone remembers. It reinforces the fear. It does. But the crucial, less publicized part of the Salesforce strategy is that they didn't just bank all the savings.

[00:51:21] They explicitly and deliberately redeployed hundreds of those affected employees into new professional services, technical sales, and customer success roles. They took employees who understood the product deeply and shifted them from being a pure cost center support into revenue generating roles. Exactly. They captured the reinstatement effect internally. But executing this requires incredibly careful communication because the workforce is already terrified. We saw this play out with Duolingo.

[00:51:51] The language learning app. What happened there? In early 2025, Duolingo's CEO, Luis Fanon, shared an internal memo outlining the company's aggressive new AI-first strategy. The memo implied that the company would heavily reduce their reliance on contractor work and strictly limit future headcount growth because they expected AI to handle the expansion. And the internet immediately lost its collective mind.

[00:52:17] The narrative instantly became Duolingo is firing everyone and replacing them with robots. It was a massive PR backlash. It escalated so quickly that much later, Von Ahn had to do a major media tour and a lengthy interview to clarify the situation. He had to publicly insist that he hadn't provided enough context in the original memo. He had to unequivocally state that they had never laid off full-time employees due to AI and had absolutely no plans to do so.

[00:52:45] The strategy was always to expect their existing salaried staff to accomplish more and scale the business using AI, not to replace them. It demonstrates how incredibly sensitive and volatile the topic of job security is right now. You cannot communicate about AI purely in terms of efficiency without terrifying your people. And this touches on a vital organizational concept established by Amy Edmondson in her 1999 research on psychological safety.

[00:53:12] Edmondson proved that for any organization to learn, adapt and innovate, employees must feel a baseline of psychological safety. They must feel secure enough to experiment, to report errors without fear of retribution, and to admit when a new process isn't working. Let's apply that to a human agent workforce. If your workforce fundamentally believes that every single time they correct an AI agent, every time they provide feedback to fine tune the model,

[00:53:40] they are literally teaching their replacement how to fire them, what are they going to do? They are going to actively sabotage the system. They will passively hide the AI's errors. They will refuse to provide the critical feedback loop that these machine learning systems desperately need to improve. A workforce paralyzed by the fear of replacement is fundamentally incapable of capturing the redesigned dividend.

[00:54:04] Interestingly, research from MIT has shown that the most significant financial returns from AI often come from using it to reduce external agency spend, software licensing costs, or outsource business processes, not by firing your core internal staff. Okay, I have to push back hard on this. It sounds incredibly enlightened, humane, and academically sound to say, reinvest capacity, keep your people, and retrain them.

[00:54:30] But we operate in the brutal reality of modern corporate capitalism. Wall Street, activist investors, and corporate boards do not generally reward long-term redeployment strategies. They reward companies with immediate stock price bumps for decisive, brutal headcount cuts. It looks like strong management to them. How does a CEO actually stand up in a hostile boardroom and say, we are going to absorb the cost of these massive AI software licenses,

[00:54:58] and we are explicitly not going to lay anyone off to pay for it because we are retraining them? If they say that, they'll be replaced by a CEO who will make the cuts. It is the hardest argument an executive has to make. To win that argument with a board, you cannot rely on appeals to humanity. You have to fundamentally change what you are measuring and how you present the definition of value.

[00:55:21] You have to mathematically prove that retaining and redeploying human capital generates higher long-term earnings before interest in taxes than a short-term payroll slash. Which brings us directly to the fifth and final solution for long-term survival. Section seven, governance and measurement. How do we measure this so the board actually buys in? Solution five is leaders must govern AI economics exclusively through outcome measures.

[00:55:48] You cannot justify the investment by presenting a dashboard showing how many employees logged into a chatbot this week, or how many millions of tokens were generated. You have to ruthlessly track unit economics, end-to-end cycle time, direct customer outcomes, and even human experience metrics, like the meeting loads we saw at Anthropic. Give me an example of a leader who is trying to force this structural shift in how value is perceived. Toby Lutka, the CEO of Shopify, is an interesting case study.

[00:56:18] In early 2025, he sent a company-wide memo effectively mandating the reflexive use of AI for all employees. But he tied it to growth. He instituted a policy where managers had to explicitly prove why they couldn't accomplish a new objective using AI tools before they were allowed to request any new human headcount. Wow. He put up a hard gate. You want to hire a new analyst? Prove to me that an AI agent can't do the analysis first. Exactly.

[00:56:46] And he went a step further, adding AI usage and proficiency questions directly into the performance and peer review processes. Now, this is a very aggressive leadership signal. It leaves no ambiguity that AI is an absolute non-negotiable priority. But relying on usage metrics also highlights the danger of what organizational experts call the activity trap. Break down the activity trap. Why is measuring usage dangerous?

[00:57:12] If an organization, whether it's Shopify or anyone else, measures an employee's success merely by how many times they interact with the AI or how many prompts they write per day, employees will act rationally. Employees are smart. They will optimize their behavior for the metric they are being judged on. So if my bonus depends on my AI utilization score, I'm just going to ask the enterprise chatbot to summarize my personal grocery list ten times a day so my dashboard looks green? Exactly.

[00:57:41] You will generate massive amounts of completely empty activity just to satisfy the review process. Usage absolutely does not equal value. Dr. Westover's paper highlights a glaring statistic. 80% of employees self-report experiencing personal productivity gains from using AI. But as we discussed, only 37% of companies report any actual EBIT impact.

[00:58:06] That 43% gap between personal feeling and corporate reality is the activity trap in numerical form. Your metrics must tie exclusively to business value, not keystrokes. So let's make this actionable for the listener. If you are a director or a VP evaluating an AI initiative today and the vendor or your IT team shows you a dashboard, what is the single most important metric you should demand to see to prove it's actually working?

[00:58:34] You demand to see the cost per resolved outcome, and it must be measured alongside a strict independent quality score. I don't care how fast the agent drafted the email. What did it cost the organization in total compute tokens and human review time to fully resolve a customer issue or deliver a finished product feature? And did the quality of that resolution meet or exceed our historical standard?

[00:58:58] If you aren't tracking both the true cost of resolution and the quality simultaneously, you are flying blind into the jagged frontier. And to manage all of this over the next decade, Dr. Westover argues that companies need to build three core long-term capabilities. What are they? First, they need work-designed literacy distributed across the entire leadership bench. For the last 50 years, most managers were trained simply to manage people within static roles that someone else in HR designed decades ago.

[00:59:26] Today, every single frontline manager has to become a work designer. They have to know how to tear a process down and dynamically divide tasks between agents and humans. And structurally, how does the organization support that? That is the second capability. Integrated people and technology governance. You cannot have IT buying software in a silo, HR planning headcount and training in a silo, and finance setting arbitrary budgets in a silo. They must be forced to plan together.

[00:59:55] No AI initiative should ever be approved without a concurrent, fully funded plan for how it will affect the human workflows. And the third capability brings us back to the pipeline problem. Yes. Learning systems and talent pipelines. Organizations have to actively protect junior development time. They have to intentionally design apprenticeships where junior staff can learn the systems.

[01:00:19] And they must cultivate environments of deep psychological safety where employees are rewarded, not punished, for reporting when the AI is failing. But looking at the broader economic data, the corporate world is actually running in the exact opposite direction, aren't they? They are. And this is perhaps the most alarming data point in the entire paper. Workday released a massive data set in 2026, analyzing global job postings.

[01:00:45] Unsurprisingly, employer demand for hands-on technical AI skills skyrocketed, rising by 51%. Everyone wants prompt engineers and AI integrators. That makes sense. But simultaneously, employer demand for broad management and leadership skills fell by 7%. And demand for training and development skills fell by a staggering 13%. Oh, that is a massive flashing red flag. It is. Catastrophic.

[01:01:13] At the exact historical moment when organizations desperately need skilled managers who know how to redesign work, coordinate complex human-agent interactions, and safely train a fragile workforce, they are aggressively deprioritizing those exact skills in favor of pure technical capability. They are spending millions buying the Formula 1 engine. And they are actively firing all the mechanics who know how to attach it to the car. That is the perfect synthesis of the AI value gap.

[01:01:42] We have covered a massive amount of ground today, diving deep into the realities of the modern enterprise. Let's synthesize this journey. The core truth is that artificial intelligence is not a magic wand that you wave over a P&L statement to make costs disappear. It is a general purpose technology, much like the steam engine, electricity, or the early internet. And like all general purpose technologies, it demands a fundamental, often painful redesign of how we actually structure human work.

[01:02:11] The lesson of history remains entirely unchanged. From the isolated coal miners of 1951, grappling with the long wall method, to the exhausted anthropic engineers of 2026, trapped in endless coordination meetings, the rule is absolute. If you optimize the technology while ignoring the human social system, you will fail. You will shake the corporate carriage to pieces.

[01:02:36] To actually capture the redesigned dividend, we have to remember the five evidence-based steps. 1. Do not just substitute tasks. Redesign workflows end-to-end to capture freed capacity. 2. Explicitly plan and document how human roles will evolve. 3. Combat executive ego by setting firm, predefined stopping roles for bad pilots.

[01:02:59] 4. Reinvest your freed human capacity into revenue-generating roles rather than defaulting to the illusion of head count cuts. 5. Govern by measuring real business outcomes and resolved costs, not just empty activity metrics. 5. For you, the listener, whether you are a CEO trying to steer a global enterprise or a junior manager just trying to figure out how to integrate an agent into your team's daily stand-up, you can champion these principles today.

[01:03:27] 6. Don't just let AI erode your tasks into a meaningless scrap heap. 7. Use it to force a redesign that elevates your role. 7. I want to leave you with one final forward-looking thought to ponder. 8. We have spent an hour talking extensively about the jagged frontier of AI capability and how humans must act as the ultimate air traffic controllers, monitoring the systems for edge case errors, providing quality control, and stepping in when the agent inevitably fails.

[01:03:53] 8. But, AI models are improving at an exponential compounding rate. 8. What happens in 3, 5, or 10 years when the AI gets so incredibly sophisticated at evaluating its own output that the jagged frontier smooths out entirely? 8. When the models can perfectly self-correct. 9. When the machine no longer needs a human supervisor acting as a safety net. 9. Exactly.

[01:04:16] 10. When the air traffic controller is no longer required because the 50 autonomous drones can flawlessly manage themselves without ever crashing. 10. When your ability to catch an edge case error is no longer needed. 10. How will you redesign your own value to the organization and to the world when your human expertise is no longer the ultimate fail-safe? 10. That is the existential question we will all have to answer. A huge thank you for joining us on this deep dive.

[01:04:43] 10. Keep redesigning your work, keep questioning the metrics, and we will see you next time. 10.