A Conversation to Enrich, Not Extract

[00:00:00] So what if I told you that the harder you push your employees to use artificial intelligence, the less productive your company will actually become? I mean, it sounds completely backward, right? Right, because we are living in this moment where businesses are just, you know, writing these massive, massive checks for enterprise AI software. Oh, absolutely. You buy it, you plug it in, and the natural assumption from the management team is that it's like upgrading the engine in a delivery truck.

[00:00:26] You put a bigger engine in, the truck drives faster, you make more deliveries. Right. It's supposed to be this clean, linear, totally predictable thing. Exactly. But that linear thinking is exactly why we are seeing such spectacular failures in deployment right now. Yeah. Leadership teams gravitate toward that engine analogy because, well, it fits neatly onto a quarterly spreadsheet. But introducing cognitive automation into a human workflow isn't a mechanical upgrade. It is a psychological earthquake. It really is.

[00:00:55] Not really. And if you are listening to this deep dive right now, you are probably feeling the tremors yourself. For sure. Maybe you're an individual contributor trying to figure out how to integrate these tools into your daily workflow without, you know, accidentally automating yourself right out of a job. Or maybe you're a team leader. You've been tasked with writing the launch memo for some new incredibly expensive AI suite. And you're just trying to figure out how to get your team to actually use the thing. Yeah.

[00:01:22] It is a universally high stress situation across the corporate landscape right now. The pressure to show a return on investment is immense. And it's coming straight from the top down. Okay. Let's unpack this. Our mission today is to take the standard corporate playbook for AI deployment, just completely tear it up, and look at the actual documented science of human motivation. Which is deeply needed right now. Definitely. And to do that, we are grounding our entire conversation in a single piece of research.

[00:01:50] It's a paper published in October 2026 in the Human Capital Leadership Review by Jonathan H. Westover. Yeah. And the title really gives away the punchline. It does. But it's worth stating up front. It's called Enrich, Don't Extract. Why selling AI as a productivity tool backfires and how leaders can introduce AI as a path to more meaningful work. What makes this paper so valuable is that it isn't just abstract theory. It's a synthesis of massive recent data sets from Microsoft, Workday, Gallup, and BCG.

[00:02:20] Real numbers. Exactly. Layered over controlled field research from Stanford and Dartmouth. And the data all points to one incredibly uncomfortable reality for corporate leadership. The central paradox. Right. If you want to get the maximum possible productivity out of artificial intelligence, you have to explicitly and intentionally stop introducing it as a productivity tool. It completely breaks the conventional wisdom.

[00:02:45] We are going to look at the sociology, the psychology, and some, frankly, unforgiving business metrics to explain why chasing efficiency directly is the fastest way to kill it. So let's start with what the paper calls the translation problem. The massive gulf between what the C-suite intends to say about AI and what the rank and file employees actually hear. Yeah, the disconnect is huge. Are we seeing this disconnect play out in the broad economic data yet? We absolutely are.

[00:03:12] And the numbers paint two entirely different realities coexisting inside the exact same companies. Take the Workday Global Workforce report from October 2026. Okay. They surveyed thousands of business leaders and 40% of them stated that their primary expectation for AI is to get more output from their current employees. 40%. Yeah. And even more aggressively, 28% explicitly expect AI to allow them to reduce overall head count. Wow.

[00:03:41] So the narrative from the top is pure extraction. Give me more work from the same number of people or give me the same work from fewer people. That is the dominant boardroom framing, period. But then, you know, you look at the employee experience. Right. The view from the break room. Exactly. The Upwork Research Institute released a massive data set in 2024 that perfectly captures this collision. They found near universal agreement in the C-suite. 96% of executives expected AI to generate higher productivity.

[00:04:11] Almost all of them. But when they surveyed the employees who were actually mandated to use these new AI tools, 77% of them reported that using the AI had actually added to their workload. Wait. I want to pause on that because it's so critical. 77% of people given a tool meant to save them time ended up with less time. Yeah. It's totally backwards. It's like it's like management is handing employees a state of the art bicycle, but they're mounting it on a treadmill. Oh, that's a great way to put it.

[00:04:40] Like the leaders are standing off to the side admiring the carbon fiber frame and the gears thinking, look at the speed potential. But the employee actually sitting on the saddle just feels like they have to pedal twice as fast just to avoid falling off. Yeah, they're sweating. They're exhausted. And they aren't moving an inch forward. The treadmill analogy captures the friction perfectly, particularly because the data shows employees are often peddling blind.

[00:05:03] In that same Upwork study, 47% of the employees openly admitted they didn't even know how to achieve the expected productivity gains their bosses were demanding. So you have a mandate from above to go faster using a tool you don't fully understand. And the act of trying to figure it out is eating up your whole day. And this leads us to what Microsoft identified in their 2025 work trend index as the AI experience gap. The experience gap. Right.

[00:05:31] The entire perception of AI fractures along hierarchical lines. According to Microsoft, 69% of leaders are using AI weekly compared to only 45% of employees. OK, so a big drop off there. And because leaders use it more, 79% of them view AI as a career accelerator. But only 67% of employees feel the same way. You know, I have to put myself in the shoes of a CEO for a second.

[00:05:58] If I'm running a company, I use an AI tool to summarize a dense 50 page financial PDF before a board meeting. It takes 10 seconds. I just saved myself an hour of reading. My personal experience with the tool is incredible. It's so easy for me to assume that same magic is happening at every single desk in my company. But that assumption is a massive blind spot. The CEO controls the surplus time they just generated.

[00:06:24] They can use that saved hour to strategize or rest or take a longer lunch. They own their time. Exactly. An entry-level analyst doesn't have that control. When an analyst saves an hour, the company absorbs it and just hands them more work. This raises an important question. What is the emotional baseline of the workforce receiving these tools? I mean, it has to be pure anxiety, right? It is.

[00:06:48] The Pew Research Center found that 52% of U.S. workers are actively worried about the impact of AI. And 32% expected to directly result in fewer personal opportunities. That's rough. And it gets worse. A massive global survey by Boston Consulting Group found that 41% of workers worried their specific job will completely disappear within a decade. Wow. So when leadership sends out like a cheerful Friday afternoon email saying, Great news team!

[00:07:15] We're rolling out a new AI suite to make us 30% more productive. Right. With all the exclamation points. Yeah. That message is landing in a minefield of existential dread. Exactly. But we have to look at the incentives driving the leaders too. Why do they default to this extraction framing when it causes so much panic? Well, I'd assume it's the cost. Enterprise software isn't cheap. It is astronomically expensive. And the financial pressure to justify that spin is immense.

[00:07:45] Gardner published a report in October, 2026 surveying chief information officers. Only 13% of CIOs said their enterprise had realized significant or transformational value from AI. Wait, just 13%? Just 13%. Wow. So you have executives staring down multi-million dollar software invoices. The transformational magic hasn't happened and the board is demanding answers. And in a state of panic, leaders revert to what they know how to measure.

[00:08:13] It is incredibly difficult to measure whether an employee's strategic thinking improved. Sure. But it is very easy to measure if they processed 20% more invoices today. So leaders default to raw, trackable efficiency. They frame the human being as an output variable to be maximized. Okay. Let's ground this in reality because talking about surveys is one thing, but how does this extraction framing actually alter behavior on the floor? I know the Westover paper highlights a really fascinating controlled study about this.

[00:08:42] Yes. The Stanford study published in October, 2026 by Karuna, Karan, Ravi Andrad and Kim. This is perhaps the most critical piece of evidence in the entire deep dive. Okay. Why is that? Because it perfectly isolates the variable of leadership framing. Set the scene for us. Where did this actually happen? It took place within a large corporate law firm. The firm purchased a generative AI drafting tool for their paralegals. And the researchers were able to study two distinct divisions within the firm.

[00:09:12] Like an A-B test. Exactly. These two divisions were essentially identical twins. Same demographics, same workload, same types of cases. And both groups were given access to the exact same AI tool at the exact same time. The only difference was the narrative the manager used to introduce it. Okay. Let's look at Division A. What did their manager say? The manager in Division A used a classic productivity frame, the extraction narrative.

[00:09:38] They told the paralegals, this tool is going to help you draft documents faster and it's going to allow us to cut our turnaround time significantly. Which, if we're being honest, sounds completely reasonable. I mean, we bill by the hour or we work on margins. My job as a manager is efficiency. Are you telling me I shouldn't want my paralegals to draft faster? From a purely mechanical standpoint, sure, it makes sense. From a psychological standpoint, it's a disaster. Really? Yeah. The results in Division A were terrible.

[00:10:08] Adoption was shallow. The paralegals only used the tool for the very specific narrow tasks the manager mentioned. They didn't experiment at all. And most revealingly, when interviewed, the paralegals reported feeling like their expertise was being commodified. They felt like cogs in a machine. I want to dig into why that happened. Why does being told to go faster make a highly educated professional just shut down? To understand that, we have to look at a foundational piece of organizational psychology called the job characteristics.

[00:10:38] Developed by Hackman and Oldham. Okay. What's the core of that? They proved that motivation isn't just about a paycheck. A job has to have specific internal characteristics to keep a human engaged. Two of the biggest ones are skill variety, the opportunity to use different talents, and task identity, the feeling of completing a whole identifiable piece of work from start to finish. So as a paralegal, I get a messy pile of case files. I have to read them, synthesize the arguments, structure a draft, refine the language, and then produce a final brief.

[00:11:08] I'm using analytical skills, writing skills, and organizational skills. Precisely. Now what happens when an AI is introduced purely for speed? The AI takes over the synthesis and the drafting. It takes the cognitively demanding, interesting parts of the job. So what is left for the human? Proofreading the robot. Exactly. You are reduced to monitoring the machine for hallucinations and fixing formatting errors. The skill variety is gone. The task identity is broken.

[00:11:35] You went from being a creator to an editor. Yeah, you're stripping the intellectual joy right out of the work. Which brings in self-determination theory by Desai and Orion. They found humans need three things at work. Autonomy, competence, and relatedness. Autonomy, competence, relatedness. Okay. When management says, use this machine to go faster, they destroy autonomy because the machine is now pacing the work. They destroy competence because the machine is doing the smart part.

[00:12:05] When you frustrate those psychological needs, internal motivation just crashes. Which brings us to the contrast. Let's look at Division B. The manager there took a totally different route, right? Yeah. The manager in Division B used the enrichment frame. They didn't start the meeting by talking about the software or turnaround times at all. They started by asking the team a question. What was it? They asked, what tasks in your daily workflow do you find incredibly tedious?

[00:12:33] What do you wish you had more time to actually focus on? Wow. That completely flips the dynamic. You aren't imposing a mandate. You are offering a solution to a problem they already have. And the data from Division B is astonishing. Compared to the group that was told to be productive, the enrichment group used the AI tool 58% more overall. That's a huge jump. And they engaged in 70% more experimentation. Because they weren't boxed into the single task of drafting faster. Right.

[00:13:03] And what's truly remarkable is how their usage evolved. The paralegals in Division B started using the AI for complex legal research. They used it to model different arguments for case analysis. They found highly valuable applications for the software that the management team hadn't even conceived of. There's an academic term for that, isn't there? When employees reshape their own work? Yes, job crafting. When employees feel secure and motivated, they proactively expand the boundaries of their roles.

[00:13:31] They automate their own boring tasks so they can elevate their value to the organization. That makes total sense. But they only do this if they aren't terrified of the consequences of their own efficiency. Division A felt threatened, so they did the bare minimum. Division B felt empowered, so they innovated. Here's where it gets really interesting and, frankly, pretty alarming. We've seen that pushing productivity stalls innovation. But the research shows it goes much deeper than that. It does.

[00:14:00] When you enforce an extraction frame on a fearful workforce, you don't just get stagnation. You actively damage the company and the psychological well-being of the employees. Absolutely. We are seeing the rise of a highly defensive corporate culture. Let's look at the phenomenon of concealed use. Concealed use, like hiding it. Yes. Because AI has been framed as a threat to headcount, employees are taking their usage completely underground.

[00:14:26] Slack's 2024 Workforce Index found that 48% of all desk workers are highly uncomfortable telling their managers that they use AI for common tasks. Wait, nearly half the workforce is lying about how they get their work done. If a tool is supposed to be this massive corporate priority, why are they hiding it? Because they are terrified of the reward for efficiency. If an employee admits that an AI helped them condense a four-hour report into a one-hour task,

[00:14:53] they don't believe the manager will say, great, take the afternoon to do some deep thinking. Right. They believe the manager will say, great, I need three more reports by 5 p.m. It's the treadmill again. If I show you I can pedal faster, you're just going to turn up the speed of the belt. Exactly. And a joint study by KPMG and the University of Melbourne shows how deeply entrenched this behavior is. They found 57% of employees actively hide their AI use, deliberately presenting AI-generated work as their own original human effort.

[00:15:23] So we have a shadow IT infrastructure where people are literally plagiarizing chatbots. But the most dangerous finding in that KPMG study is that 66% of employees admit they rely on AI output without checking its accuracy. Two-thirds of workers are blindly passing unverified information up the chain of command. Why would any professional risk their reputation doing that? Because of the productivity mandate.

[00:15:47] If management demands a 30% increase in output volume, quality control is the first thing that gets sacrificed to hit the metric. The employee figures, well, if you want it fast, you get what the robot gives you. And this leads directly to a new term that the Westover paper highlights, which I think is going to define the next decade of corporate life, work slop. It's a brilliant, if depressing, term. It comes from research conducted by BetterUp Labs and the Stanford Social Media Lab.

[00:16:13] They define work slop as AI-generated content that is grammatically perfect, highly polished, utilizes all the correct corporate jargon, and contains absolutely zero substantive value. Yes. It's the beautiful, empty shell of an email. A five-paragraph update that says absolutely nothing. And it is completely flooding enterprise communication channels right now. The Stanford study found that 40% of U.S. desk workers received work slop from a colleague in just the previous month.

[00:16:42] And here is the tragic irony of work slop. It doesn't actually save the company any time. The study noted that when someone receives a piece of work slop, it takes them nearly two hours to fix it, untangle it, or clarify it. Because you have to read the five paragraphs, realize it makes no sense, message the sender, and ask them what they actually meant. The sender saved 10 minutes writing the email and cost the recipient two hours trying to decipher it. The net productivity of the organization drops. If we connect this to the bigger picture,

[00:17:12] this feels like an inevitable outcome of treating human beings like assembly-line robots. Sociologists call this algorithmic control. When a company uses technology to aggressively direct, pace, and evaluate human work, the humans don't just passively accept it. They find defensive, often counterproductive ways to survive the system. They game it. Right. They optimize for whatever metric management is tracking, even if hitting that metric destroys the actual value of the work.

[00:17:39] And there is a massive psychological hangover to this kind of work, isn't there? Severe. A study published in Scientific Reports looked at 3,500 participants. They had them collaborate with a generative AI on a specific task. And predictably, the AI helped them complete that initial task faster. But then, the researchers gave them a secondary non-AI task to complete immediately afterwards. What happened?

[00:18:03] Intrinsic motivation for the secondary task plummeted by 11% and reported boredom spiked by 20%. Because the AI had hollowed out their sense of agency on the first task, their brain simply disengaged. When you move on to the next task, you are already checked out. And if you want to see the ultimate financial consequence of that hollowing out, look at the macroeconomic data.

[00:18:27] The Westover paper cites a massive Danish study tracking 25,000 workers across multiple industries to measure the actual real-world impact of AI on labor. The result? The AI saved an average of 2.8% of work hours. It is essentially a rounding error in a 40-hour work week. A rounding error. And it had zero detectable effects on earnings or wages. So the extraction frame is a comprehensive failure.

[00:18:56] It terrifies employees, breathes shadow IT, creates a tidal wave of error-prone work slop, tanks human motivation, and totally fails to deliver the promised financial returns. But the tragedy is that it doesn't have to be this way. When organizations choose the enrichment frame, the results are radically different. I want to look at some massive real-world deployments to prove this isn't just an academic theory.

[00:19:21] Let's contrast some highly publicized corporate failures in AI framing with some incredible successes. And ironically, the tech sector seems to be providing the worst examples. Yes, let's look at Klarna, the massive fintech company. In early 2024, their CEO made headlines by bragging that their new AI assistant was doing the work of 700 human customer service agents. They framed it as the ultimate triumph of productivity and cost-cutting.

[00:19:49] And I'm sure Wall Street rewarded them for it, for a quarter or two at least. They did. But by May 2025, the reality of algorithmic control set in. The CEO had to publicly acknowledge that their relentless focus on AI extraction had severely degraded customer service quality. Shocker. Customers were furious because they couldn't reach a human to resolve nuanced, complex financial disputes.

[00:20:15] Klarna actually had to reverse course and start aggressively hiring back human support agents. It's a completely unforced error. You degrade your own product to hit a metric. And then you have Salesforce. In 2025, they cut 4,000 jobs, explicitly citing AI efficiencies and stating they needed fewer heads. Now, whether or not that was a mathematically sound financial move for Salesforce in the short term, the collateral damage to the global workforce is immense. Oh, totally.

[00:20:44] These highly publicized corporate actions act as a warning signal. When a bank teller in Ohio or a marketing associate in London reads that Salesforce fired 4,000 people because of AI, they immediately view the AI tool on their own laptop as an existential threat. They will not job craft. They will not innovate. They will resist. So where is this actually working? The paper highlights an industry that frankly surprised me. Healthcare.

[00:21:12] What's fascinating here is that the healthcare sector essentially stumbled into the perfect enrichment framing, largely out of sheer desperation to keep their workforce from completely burning out. Let's look at the Permanente Medical Group. They rolled out ambient AI scribes to over 7,000 physicians, covering roughly 2.6 million patient encounters. Just massive scale. Huge.

[00:21:33] And for context, an ambient scribe is an AI tool that securely listens to the conversation between the doctor and the patient and automatically generates the formal medical notes for the patient's chart. And the results were phenomenal. The AI saved an estimated 15,791 hours of documentation time. But the time saved isn't the most important metric. What is? 84% of the physicians reported that using the AI resulted in better communication with their patients.

[00:22:01] And 82% reported much higher overall job satisfaction. We saw the exact same trajectory at Mass General Brigham. They piloted ambient documentation. And within just 84 days, the reported rate of clinician burnout plummeted by 21 points. Wow. 21 points in under three months. That is unheard of in medical administration. So we have to ask, why did the doctors embrace this technology so enthusiastically while the paralegals and the tech workers actively resisted it?

[00:22:31] Because the hospital administrators didn't walk in and say, great news, doctors. We bought an AI. Now we expect you to see 20% more patients every single day. If they had done that, the doctors would have rioted. Exactly. The AI targeted the exact tasks the physicians desperately wanted to shed. In medicine, they call it pajama time. The agonizing hours doctors spend sitting at home at 9 p.m. typing up clinical notes from the day. Ugh, the worst part of the job.

[00:22:57] By automating the drudgery, the AI returned the physician's attention to the deeply human, cognitively complex element of their job, diagnosing the patient sitting in front of them. The AI wasn't a threat. It was a rescue operation. And this massive divergence in how we use tools is captured perfectly in the Anthropic Economic Index. They looked at how people use Claude, their AI assistant.

[00:23:22] When individuals use the tool voluntarily for their own personal workflows, 57% of the time they use it for augmentation and collaboration, treating it like a brainstorming partner. But when enterprise companies mandate its use, 77% of the time it is used for pure automation, trying to blindly replace a human step in a process. Individuals naturally seek a partner to enrich their work. Corporations default to seeking a machine to extract labor.

[00:23:51] And the data overwhelmingly shows that the individuals have the right approach. So if you are listening to this and you manage a team, your primary directive is to find the equivalent of after-hours charting for your specific department. You have to find the pajama time. Which brings us to the operational core of Dr. Westover's research. He doesn't just diagnose this translation problem. He provides a highly specific, evidence-based playbook for leaders to introduce AI correctly.

[00:24:20] Let's dive deep into this playbook. It starts with co-diagnosing the work. What does that actually look like on a random Tuesday morning when a manager needs to roll out a new tool? It requires a massive shift in leadership ego. You do not walk into the conference room, plug in your laptop, and pitch the features of the software. You start by asking the team to map their own workflows. Okay, map it how? You ask them to categorize their daily tasks into three buckets. Tasks they want to automate, tasks they want to augment with a co-pilot,

[00:24:50] and most importantly, tasks they want to aggressively protect as strictly human endeavors. Just like the successful manager in the Stanford law firm study. Precisely. And the data shows that workers are actually eager for this conversation. The Stanford Work Bank study surveyed employees across 104 different occupations. They found that workers actually want automation for roughly 46% of their daily tasks. Almost half. They actively want to hand half their job to a robot.

[00:25:18] But only if they have a say in which half. They want to automate the rote administrative work to free up time for the high-value complex work. If you ask them what to automate, they will guide you perfectly. If you dictate from the top down what is being automated, they will fight you every step of the way. The next crucial concept in the playbook is backing your message with actual learning infrastructure and protected time. Because sending a lovely email about enriching our workflow is meaningless if you don't provide the resources to make it happen.

[00:25:48] This is where the hypocrisy of many corporate deployments is really exposed. A company will mandate that employees learn a new AI tool, but they won't adjust the employee's daily quotas or deadlines. Typical. So the only way the employee can learn the tool is by sacrificing their own personal time, working late to figure out how to prompt the chatbot. That instantly breeds resentment. Let's look at some contrasting approaches to training. JPMorgan Chase took a fascinating route. Yes.

[00:26:16] JPMorgan took a completely opt-in, curiosity-driven approach to their internal AI platform. They didn't make it mandatory. Instead, they offered a prompt of the week. They established peer-to-peer learning channels. And they gave employees space to experiment. And the result? Over 250,000 users adopted the platform voluntarily. Moderna is another great example. They focused entirely on employee-driven creativity.

[00:26:41] They gave their staff access to the tools and, crucially, protected time to experiment without penalty. Within just two months, Moderna employees had independently built 750 custom AI tools, tailored to their own highly specific scientific and administrative workflows. 750 custom tools in eight weeks. You cannot mandate that kind of innovation. It only comes from intrinsic motivation. Now, contrast that with how Citi approached it.

[00:27:10] Citi went with the traditional top-down corporate hammer. They made AI prompt training strictly mandatory for roughly 175,000 employees. Which already signals a lack of trust. But worse, the training they deployed was designed to be completed in just 10 minutes. 10 minutes. I mean, I can't even figure out how to format a spreadsheet in 10 minutes. What are you possibly teaching about generative AI architecture in 10 minutes? You aren't teaching anything.

[00:27:37] You are teaching employees how to click next to check a compliance box for the HR department. Right. It signals to the entire workforce that leadership doesn't actually value deep, meaningful exploration of the tool. They just want the legal cover of saying everyone was trained. To truly back the enrichment message, you must provide protected time. A manager has to look at their team and say, For two hours every Friday afternoon, I expect your output to drop to zero because I want you experimenting with this tool.

[00:28:06] This brings us to the moment of truth in the playbook. Reinvesting freed capacity. Let's say you do everything right. You co-diagnose the work. You give them protected time. The AI is deployed. And it works beautifully. Suddenly, your team has 15% more free time in their week. The drudgery is gone. What happens to that time? This is the inflection point. If management looks at that 15% surplus and says, Excellent, we can fire 15% of the team.

[00:28:34] You have permanently destroyed trust. No one in that company will ever innovate again. Yeah, you've shown your hand. If management says, Excellent, everyone's sales quota just went up 15%. You have turned the AI into a treadmill. So how do you reinvest it properly? I know the paper highlights IKEA, which is an incredible case study in doing this right. IKEA's parent company, the Inca Group, deployed a highly efficient AI customer service bot named Billy.

[00:29:01] It was incredibly effective at handling routine, massive volume customer queries, things like checking store hours or tracking a missing delivery. This is the exact scenario where a company usually issues a press release announcing a massive layoff in their call centers. But IKEA didn't lay them off. They took 8,500 call center workers who were actively being replaced by the bot. And they fundamentally redesigned their jobs. They retrained them to become remote interior design advisors.

[00:29:30] I really want to pause and appreciate the logistics of that. You are taking someone whose job was answering complaints about missing screws. And you are trusting them, training them, and elevating them to actually design a customer's living room. And it wasn't just a charitable public relations stunt. It was brilliant, hard-nosed business strategy. By reinvesting that saved capacity to human creativity and complex relationship building,

[00:29:55] that newly created remote interior design division generated 1.3 billion euros in revenue for IKEA in its 2022 financial year. 1.3 billion. They didn't extract the savings. They enriched the roles. And it paid off massively. Morgan Stanley did something very similar on a smaller scale. They rolled out a tool called AI at Morgan Stanley Debrief. It automatically summarizes client meetings, drafts follow-up emails, logs the action items.

[00:30:22] But they explicitly pitched this to their financial advisors as a tool to reclaim time, specifically for building deeper human relationships with their clients. Yes, shifting the focus completely. And the narrative was, let the AI handle the paperwork, so you have more time to look your clients in the eye and understand their financial fears. And if you are going to change the nature of the work like that, you have to fundamentally redesign your metrics. This is the fourth major concept in Westover's playbook.

[00:30:50] You have to measure value, not volume. The metrics trap. This is where so many smart organizations completely lose the plot. It all comes down to an economic principle called Goodhart's Law. It states that when a measure becomes a target, it ceases to be a good measure. Right. The classic example. If you tell a factory they will be paid by the number of nails they produce, they will produce a million microscopic, completely useless nails. If you pay them by the weight of the nails,

[00:31:18] they will produce one giant, unusable iron block. People optimize for the metric, not the goal. Exactly. And in the AI space right now, the most toxic metric being deployed is the token. Organizations are trying to track how many AI tokens an employee consumes in a week to judge their productivity and adoption rates. Which is absurd. I remember reading about Meta attempting this. They apparently set up an internal leaderboard that publicly ranked employees based on their sheer volume of AI token usage.

[00:31:48] It completely ignores the quality or utility of the output. It just rewards raw mechanical interaction with the chatbot. I read a report that Amazon staff started running internal AI tools to summarize massive, entirely irrelevant technical documents, things that had nothing to do with their actual jobs, just to inflate their personal usage stats so they wouldn't fall behind on the metrics dashboard. That's incredible. You literally have some of the smartest software engineers in the world burning through server farm compute power

[00:32:17] to summarize the corporate cafeteria menu, just to appease a blind algorithm. And Shopify took a similarly restrictive approach by tying AI usage directly to the performance review process. They essentially demanded the teams prove a task couldn't be done by an AI before they were allowed to request additional human headcount. Which guarantees that a manager will spend three days fighting with a chatbot to do a task it isn't suited for, failing, and documenting the failure,

[00:32:44] rather than just assigning the task to a human who could have done it in two hours. They waste time just to prove they obeyed the metric. Which is why leaders must ruthlessly retire volume metrics. You have to evaluate outcomes. You measure the depth of the research, the reduction in error rates, the quality of the client interaction. If an employee uses an AI to conduct incredibly deep, nuanced research on three complex cases, rather than shallow, hallucination-filled research on ten cases,

[00:33:13] the metric must reward the depth, not punish the lack of volume. Which brings us to the final piece of the immediate deployment playbook, safe disclosure and worker voice. We talked earlier about how 48% of workers are hiding their AI use. How do you create an environment where people feel safe bringing their innovations out of the shadows? If we connect this to the bigger picture, governance of artificial intelligence cannot just be a PDF list of prohibitions handed down from the legal department. Definitely not.

[00:33:42] When you impose rigid rules without context or employee input, you don't stop the behavior, you just drive it underground. You create a shadow IT environment where employees are pasting highly sensitive company financial data into random, unvetted web browsers just to get their work done secretly. We need governance that includes the voice of the worker. And a truly fascinating example of this is the 2023 Writers Guild of America strike and their eventual agreement with the major Hollywood studios.

[00:34:11] It remains a landmark agreement for the AI era because it prioritized human agency. Right. Because the writers didn't demand a total blanket ban on AI. They negotiated a highly nuanced agreement. It stated that AI cannot write or rewrite literary material and AI output cannot be considered original source material. It explicitly preserved and protected human authorship and human compensation.

[00:34:37] But crucially, the agreement allowed for the voluntary use of AI by the writers themselves as a brainstorming or structuring tool provided the studio consented. The workers maintained agency over their own craft. They had a structural voice in how the technology would be integrated. Contrast that collaborative, negotiated approach with the total public relations disaster at Duolingo. In April 2025, the CEO of Duolingo sent out an aggressive AI-first memo to the staff.

[00:35:07] It essentially announced that contractors were being rapidly phased out in favor of AI generation and that new internal headcount would be severely restricted unless automation completely failed. The backlash was severe and immediate, both internally from terrified staff and externally from the public. It painted the company as cold, extractive, and entirely focused on replacing humans. It was so disastrous that by August 2025, just four months later,

[00:35:34] the CEO had to do a massive public mea culpa. They had to clarify that they had never laid off full-time employees and they frantically started emphasizing their internal AI exploration time and human creativity. They had to aggressively backpedal to an enrichment narrative because the extraction narrative caused a total revolt. So if you follow these principles, diagnosing work together, providing real learning time, reinvesting saved capacity, redesigning your metrics,

[00:36:04] and ensuring safe disclosure, you can survive the initial shock of deployment. But Dr. Westover's research concludes by looking further out on the horizon. How does an organization build long-term structural capability to thrive in this new era? Because this isn't just a software update that you install and forget about. This is a permanent structural shift in how human beings interact with their labor. The paper points to three foundational pillars a company must build.

[00:36:32] The first pillar is recognizing that work design is now a fundamental daily leadership discipline. Historically, designing a job role was something the HR department did when you were hired. They wrote a job description and it rarely changed for five years. But with AI rapidly automating different unpredictable slices of a job, roles are becoming incredibly fluid. The boundaries of what constitutes my job are blurring every single week as the models get smarter. Exactly.

[00:37:02] So every single manager, from the frontline to the C-suite, must be trained to constantly analyze how AI is affecting their specific team's psychological needs. A manager has to look at an employee and ask, is this new tool decreasing their autonomy? Is it isolating them and reducing their human contact? Is it squandering their unique analytical skills? If the answer is yes, the manager has to redesign the workflow immediately to restore that meaning. It is an ongoing daily discipline.

[00:37:31] The second pillar is building learning systems that compound. And this is fascinating to me because the skills we actually need are changing faster than any university curriculum or corporate training module can possibly adapt. The 2026 Workday data proves this beautifully. They tracked employer demand for specific AI skills and job postings. They found that demand for basic AI prompting, just knowing how to type a conversational question into a chatbot,

[00:37:59] actually fell by 25% after peaking in early 2026. Wow, already. Yeah, the novelty wore off, the models got much better at understanding natural human language, and prompt engineering stopped being a unique superpower. So what replaced it? What are companies actually looking for now? Demand shifted rapidly to hands-on architectural skills. The demand for employees who know how to build custom AI tools, integrate APIs, and automate complex multi-step workflows rose by 51%.

[00:38:28] So it's not just about using the tool, it's about building the infrastructure around the tool to solve highly specific business problems. And you simply cannot teach those architectural skills in a single afternoon seminar. Organizations have to build compounding learning systems. They need internal peer-led networks, shared prompt libraries, internal Reddit-style forums where an analyst in marketing can post a weird, highly specific automation trick they discovered, and an analyst in finance can instantly adopt it.

[00:38:58] You have to let the workforce teach itself. So what does this all mean when we bring it together? It all rests on the final pillar, shared gains and durable trust. It means that if the company wins, the employee has to win too. It has to be a tangible reality, not just a slogan. Leaders have to prove repeatedly through their actions and their budgets that AI is a path to better work. If the AI rollout saves the company $10 million in operational costs,

[00:39:27] the employees who generated those savings need to see a share of those gains. Exactly. That might mean direct financial bonuses, but also might mean transitioning to a four-day work week without a pay cut. It might mean funding massive, personalized professional development budgets. It means providing fundamentally more interesting, engaging work. Because if the gains are not shared, if the company absorbs all the surplus and gives the employee nothing but a higher quota, the trust dissolves instantly. And without trust,

[00:39:57] as we saw with the hiding, the paranoia, and the work slop, the technology becomes a weapon that the employees will actively, if subtly, sabotage. It is the ultimate self-fulfilling prophecy. If you sell artificial intelligence as a ruthless productivity tool, you breed fear. You breed shadow IT. And you get terrible, hollow output. But if you lead with enrichment, if you genuinely, structurally try to make the jobs of your employees more meaningful,

[00:40:25] you actually unlock the massive transformational productivity gains that the boardrooms are so desperate for. The irony is profound. The only way to get the math to work on the spreadsheet is to stop obsessing over the spreadsheet and focus entirely on the psychology of the human being using the tool. Which brings this whole conversation directly back to you, the listener. Whether you are an individual contributor trying to figure out how to integrate these models into your own daily tasks, or a C-suite executive

[00:40:55] drafting a launch memo for a massive new enterprise suite, I want you to remember the Stanford paralegals. A perfect example. Ask yourself or ask your team, what is the drudgery you desperately want to shed? Find it. Automate it. And then fiercely, unapologetically, protect the tasks that make your work uniquely human. Because the goal of AI integration is not to turn a human into a faster machine. The goal is to free the human to be more effective, more creative, and more engaged.

[00:41:24] I want to leave you with one final, slightly provocative thought to mull over. We spend all our time in the business world debating how many hours AI will save us, or what kind of marketing copy it can generate, or how it will impact the quarterly margins. But what if the ultimate, long-term value of artificial intelligence isn't found in its output at all? What if its true value is the way it is forcing a crisis in management? By being so incredibly, ruthlessly good at automating the robotic, soulless,

[00:41:53] tedious parts of our jobs, AI is holding up a mirror to how terrible modern work design had become. It is forcing us to tear down the factory floor mindset of the 20th century. AI might be the exact painful catalyst we needed to completely redesign our workplaces, to force us to remember what human capital actually means. We started this deep dive talking about buying a faster engine, expecting a clean, linear transaction. The truth is, AI isn't an engine.

[00:42:24] It's a mirror. And right now, it's reflecting exactly how much we value the humans in our organizations. That is a profound way to look at it. Mm-hmm. It demands a totally different kind of leadership. So until next time, keep asking the hard questions, protect your human skills, and whatever you do, don't let anyone put your bicycle on a treadmill.