The Redesign Dividend: Why Most Organizations Still Aren't Getting Value From AI, and How Leaders Can Redesign Work for a Human-Agent Workforce

[00:00:00] So let me ask you a question. Are you using AI at work right now? I mean, I definitely am. Right. And chances are, you listening right now probably are too. Oh, absolutely. Everyone is. But here's the real question we need to talk about. Has it actually changed your company's bottom line? Ooh, yeah, that is the big one. It really is. And that's exactly the mission of our deep dive today. We are unpacking this really fascinating piece from the Human Capital Leadership Review by Dr. Jonathan Westover.

[00:00:27] Yeah, it's called the Redesign Dividend, and it is so good. It is. Because he highlights this massive, just this huge disconnect in the data. He pulls from Gartner's 2026 and 2027 surveys, and the numbers are just grim. Very grim. Like everyone is buying AI, obviously. But only 13% of organizations say they've realized any significant or, you know, transformational value from it. 13%. That's it. I mean, that is practically a rounding error when you think about the billions being spent.

[00:00:57] Right. And 44% of CEOs are just openly admitting that they're struggling to show any financial return to their stakeholders. Which is wild, to be honest. Okay. Let's unpack this. Because Dr. Westover's core argument here completely reframes the whole issue. He really does. Yeah. He says this massive AI value gap is not a technology problem at all. It is a work design problem. Exactly. And that term he uses, the redesigned dividend, is just a brilliant way to frame it. Totally.

[00:01:25] Because the fundamental mistake companies are making right now is they treat AI like a light switch. You know, you buy the license, you flip the switch for your team, and you just sit back and wait for the profits to roll in. Like it's magic. Right. But AI does not produce value just by existing. You only get that redesigned dividend when the work itself, like the daily tasks, the actual workflows, who approves what, when all of that is deliberately dismantled and reorganized. So humans and AI agents are actually complementing each other.

[00:01:55] Exactly. Okay. You have to design it that way. Well, I think to really grasp how to redesign work, we first need to look at why the default approach is failing so hard. Like, why can't I just give my team a chat bot and walk away? Yeah. The whole bolt-on approach. To understand why that fails, we have to look at this economic concept from researchers like Eric Brynjolfsson. It's called the productivity J-curve. The J-curve. So it dips down before it goes up.

[00:02:22] Right. History shows us that whenever a massive general-purpose technology comes along, electricity, computers, and now AI, it actually depresses measured productivity at first. But wait, why though? If I give my team a faster tool, shouldn't they just instantly be faster? You'd think so, right? Right. But it's because of what economists call intangible complementary assets, which is a mouthful, but... Yeah, what does that actually mean?

[00:02:48] Okay. Think about electricity. When factories switched from steam power to electric motors, they didn't just instantly speed up. Oh, because they were still set up for steam. Exactly. Steam engines needed this massive central driveshaft, so factories had to be built vertically, multiple stories high, just to connect to it. Oh, wow. I didn't know that.

[00:03:09] Yeah. So when electricity came along, you could put a small motor anywhere. But to actually get the efficiency gains, they had to literally tear down those vertical factories and build the sprawling single-story assembly lines we know today. Oh, I see. So they had to completely redesign the physical space to match the new tech. Exactly. And with AI, those intangible complementary assets are basically the modern factory floor. It's your training protocols, your reporting structures, the whole workflow.

[00:03:35] So if you just drop AI into a steam-era workflow, you're spending a fortune, but you're stuck at the bottom of that J-curve. Right, because you haven't built the structure to support it. That makes the Boston Consulting Group experiment make so much more sense to me now. Oh, the GPT-4 study? Yeah. Yeah, that one is classic. Yeah. So they gave GPT-4 to their consultants just to see what would happen. And on tasks that were really well within the AI's capabilities, like summarizing documents or brainstorming, the consultants worked, I think, 25% faster.

[00:04:05] Yep, 25%. But, and this is the crazy part, on tasks that were just slightly outside what the AI was good at, the consultants using AI actually got the wrong answer more often than the control group who didn't use AI at all. Right. The researchers call that the jagged technological frontier. The jagged frontier. Yeah, and it's probably one of the most important concepts for any leader to understand right now, because AI is fundamentally just a pattern-matching engine, right? It doesn't have a clean, obvious boundary of what it knows.

[00:04:35] Like, it doesn't say, hey, I'm great at math, but I suck at history. Exactly. It can write this brilliant, super-nuanced marketing strategy and then completely bomb a basic arithmetic problem in the very next paragraph. And it sounds so confident doing it, so it just tricks you. Exactly. It always defaults to the most statistically plausible answer, which is not always the true answer. So if you just bolt AI onto a job without figuring out where that jagged frontier is, you're basically trading visible speed for invisible errors.

[00:05:05] Which perfectly explains that anthropic data from 2026. Oh, yeah. Even an AI company fell for this. Exactly. They gave their own engineers top-tier AI assistants, and coding speech shot up by five times. 500%. But then human attention became the bottleneck. Right, because nobody redesigned the workflow. If you generate five times as much code, someone still has to review it for those invisible errors. Someone has to test it. Someone has to integrate it.

[00:05:31] So the engineers ended up spending 20% more time trapped in meetings arguing about the code, while their AI agents were sitting totally idle like 80% of the day. They automated the creation of the work, but not the management of it. Wait, let me push back on this a little bit. Sure, go for it. Because if I'm just an individual worker, and I use an AI tool that literally saves me, say, three hours a week writing up my weekly reports, I can actually see that time back on my clock. Right, local time savings. Yeah.

[00:05:59] So you're telling me that time just evaporates? I mean, how does saving three hours not help the company's bottom line? It vanishes because saving time locally doesn't change the global throughput of the company if the bottleneck is somewhere else. Okay. Give me a real-world example of that. Okay, let's say you finish that report three hours early, right? And you send it to your manager. But your manager isn't using AI. They are still reviewing everything at the old, slower human speed. Oh, I see.

[00:06:27] Yeah, so your report just sits in their inbox for three days. The company itself didn't move any faster. And meanwhile, what are you actually doing with those three hours? I mean, realistically? Probably taking a longer lunch or, you know, browsing the internet. Exactly. It evaporates into slack time. Or, even worse, your manager knows you use AI, so instead of just accepting your first draft, they ask you for six different versions of the report just because it's so easy for you to generate them. Oh, man.

[00:06:56] So I'm doing more work but zero actual values being created. Yep. Unless the whole overarching system changes, those individual time savings are really just an illusion. Okay, so bolting AI onto a broken process just gives you a faster bottleneck. So we need to look at the companies that are actually capturing this redesigned dividend, right? The ones completely dismantling how they operate. Yeah. Moderna is a fantastic example from the research. What did they do?

[00:07:24] They realized you can't plan human work and AI work in separate silos. Usually, you know, IT buys the tech and HR deals with the people. Right. Totally separate. Well, Moderna literally merged their HR and IT departments together. They created a single unit called People and Digital Technology. Oh, wow. That is a massive structural change. It is. But it lets them design the tech, the talent, and the processes all at the exact same time. They aren't just handing out generic chatbots.

[00:07:51] They're building targeted tools for specific workflows. Kind of like what Morgan Stanley did, right? Yes, exactly like Morgan Stanley. Because instead of just giving their financial advisors chat GPT, they looked at the workflow and found the bottleneck was post-meeting admin work. All those follow-up emails and notes. Yeah. So they build a custom agent called Debrief just for that. It listens to the client meeting, takes the notes, drafts the email. It saves them like 30 minutes of meeting.

[00:08:18] But the key is that the workflow was redesigned. So those 30 minutes don't become slack time. They are explicitly redirected into getting new client leads. Right. And here's where it gets really interesting, I think. Because to capture the dividend, you don't just redesign the workflow. You have to redesign the roles themselves. Yes. The actual job descriptions. Let's talk about IKEA. This blew my mind. They brought in an AI customer service bot named Billy. Billy is a great case study. Yeah.

[00:08:46] And Billy eventually took over 47% of all their call center queries. Almost half. Now, the center corporate move there is obvious, right? Oh, yeah. Fire half the staff and pocket the cash. Exactly. Just lay off thousands of call center workers. But IKEA didn't do that. They took those 8,500 workers and they completely retrained them to be remote interior design advisors. They upskilled them. Yeah. They didn't just give them a catalog. They taught them spatial planning and sales.

[00:09:13] They pointed all that freed up human capacity at a high margin, highly human service. And it paid off big time. Huge. It generated 1.3 billion euros in revenue. It's just the ultimate example of the redesigned dividend. They didn't shrink their company. They elevated their workforce. Okay. Let me play devil's advocate here for a second. For the listener who is maybe getting a little nervous hearing this. Fair enough. This sounds amazing for IKEA's bottom line.

[00:09:40] But if you take away all the easy, mindless tasks from my day, the password resets, the basic FAQs, my brain is going to be sprinting at 100% capacity all day long, dealing with only the most complex, angry customer meltdowns. Right. All the hard stuff. Yeah. That sounds exhausting. Aren't we just completely overloading human workers? It's a very real concern and psychologists actually have a name for it. It's the job demands resources model.

[00:10:09] The demands resources model. Yeah. Basically, every job has demands, the mental or emotional effort it takes, and resources, which are the tools, the support, and the autonomy you get to handle those demands. Okay. Makes sense. Well, if you automate away all the easy tasks, you dramatically spike the cognitive demands on the worker. And if you don't give them new resources and more autonomy to match, they will absolutely burn out. So how do companies actually prevent that?

[00:10:38] This is where economists Duran Asimoglu and Pasquale Restrepo talk about the reinstatement effect. The idea is that tech only really benefits workers if we deliberately create entirely new, high-value tasks for them to do. You can't just leave them with the stressful leftovers the AI couldn't handle. Oh, I see. Like Walmart, for example. When they automated a bunch of merchant tasks, they didn't just leave their employees to deal with the messy exceptions. They created these brand-new roles called agent builders. Agent builders?

[00:11:08] Yeah. They gave their employees the training and the authority to actually build their own AI tools. They gave them new resources to match the new demands. Okay. So you have to build the new house before you tear down the old one. Exactly. But Walmart and Ikea are the success stories. What happens when a company does all this, builds a new workflow, announces it, and then the AI pilot just completely fails? Oh, it happens a lot. And the data shows we are historically terrible at pulling the plug. Oh, really? Yeah.

[00:11:37] Gartner's data on this is actually pretty scary. Only 31% of chief information officers have any explicit guidelines for when an AI pilot should not go into production. Wait. Only 31%? Yeah. And it gets worse. Only 17% have rules for when to kill an AI system that's already live but failing. That means over 80% of companies are just letting these zombie AI programs run and drain money. And they have no plan to stop them. Why wouldn't you just turn off a tool that doesn't work?

[00:12:06] Because of a psychological trap called the escalation of commitment. Barry Staw studied this back in 1976. When leaders make these big public announcements about an AI initiative, they tie their personal reputations to it. And probably their bonuses, too. Exactly. So when the data shows it's failing, admitting defeat is just too politically dangerous for them. It's like keeping a gym membership for three years because you went once, took a selfie, and now you're too embarrassed to admit to your friends that you hate working out.

[00:12:35] Yes, that is the perfect analogy. They throw good money after bad because the optics of failing are just too painful. And we saw this exact thing play out super publicly with Klarna, right? Oh, yes. Klarna is the textbook example now. Yeah. So early 2024, they were on every single tech blog bragging that their AI assistant was doing the work of 700 human customer service agents. Right. They completely paused hiring.

[00:13:00] Their headcount dropped, and everyone was pointing to them as proof that AI was going to replace humans overnight. But look at the timeline. By May 2025, just a little over a year later, they had to quietly reverse course. Ouch. Yeah. The CEO had to admit that, sure, the AI was cheaper, but the quality of service was just way too low. They hit that jagged frontier. Exactly. Customers were furious, problems weren't getting solved, and Klarna had to start actively hiring human agents all over again.

[00:13:28] What's fascinating here is that Dr. Westover points out this wasn't even really a failure of AI technology. It was a failure of governance. Yes. A lack of stopping rules. Right. So a stopping rule is like a predefined metric you set before the project even launches, right? Exactly. You agree ahead of time, hey, if customer satisfaction drops below this specific number, or if repeat calls go up by 10%, we automatically pull the plug.

[00:13:53] Because Klarna only measured the cost savings, they didn't have stopping rules tied to quality, so they just let a broken system run for months. And their customers paid the price. But, you know, a delayed stopping decision doesn't just hurt the customers. It actually wreaks havoc on the internal workforce, too. Which brings us to the human impact of all this. Because we hear so much about AI taking jobs, but the text points to a much sneakier crisis happening with the talent pipeline. Yes, the headcount fallacy.

[00:14:22] So many executives just assume AI's main goal is mass layoffs. But Workday put out data showing that while 39% of leaders expected AI to reduce headcount, only 14% actually saw any decline. Okay, so mass layoffs aren't happening as much as we thought. Right. The real threat isn't senior people getting fired. It's the quiet disappearance of the entry-level job. Yeah, the Anthropic internal data is honestly chilling on this.

[00:14:49] They found that the share of junior employees among their new technical hires dropped from 27% down to just 9%. They basically just stopped hiring beginners. Right. And that creates a catastrophic gap in how an organization learns. There's a seminal paper from 1983 by Lisanne Bainbridge called The Ironies of Automation. 1983, wow. So this isn't a new concept. Not at all.

[00:15:13] She observed that if you automate all the routine, baseline work, human operators lose the daily hands-on practice they need to actually develop expert judgment. So it's like if the autopilot flies the plane 99% of the time and then a massive storm hits and the system disengages, the human pilot hasn't manually flown a plane in months. Precisely. They've lost that tacit knowledge, the gut instinct and the muscle memory to handle those complex edge cases.

[00:15:40] When the AI fails, the human is suddenly called in to fix it, but they haven't been practicing the basics. They just don't have the intuition for it anymore. Right. And honestly, this is what terrifies me. Because if AI eats all the entry-level coding or the basic legal research or the starter financial modeling, where do we get our senior experts five years from now? It's a massive problem. If we don't let junior employees practice on the easy stuff, aren't we just mortgaging our entire future expertise just to save a few bucks today?

[00:16:09] It's a huge risk. But there's actually a profound paradox here in the research that offers a way out. Oh, tell me. An MIT study by Eric Brynjolfsson found that AI is actually incredibly effective at helping novices learn because it captures the hidden knowledge of the top performers and it makes it accessible to the beginners. Oh, wow. So novices using AI actually level up faster? Exactly.

[00:16:35] The data shows they reach proficiency much quicker than those without AI. So AI doesn't have to be the enemy of the junior employee. It could be like the ultimate set of training wheels. Yes. But, and this is a massive but, that only works if the company actively protects those learning tasks. You can't just automate them away completely. Right. If you remove the junior employee from the loop entirely, you break the pipeline.

[00:17:00] You have to design roles where junior staff work alongside the AI to accelerate their learning instead of just automating them out of the building. Okay, so if you are a leader listening to this, your to-do list is looking pretty daunting right now. Very daunting. You've got to protect your junior employees. You need strict stopping rules to kill failing pilots. And you have to redesign workflows from the ground up just to actually make money. It's a lot of work. But how do we even ensure any of this is happening?

[00:17:28] Because the metrics most companies use to track AI seem completely detached from reality. Oh, they are. The current metrics are almost entirely vanity metrics. Like what? Well, McKinsey found that 80% of people claimed AI improved their personal productivity. But, like we talked about earlier, only 37% of companies saw any actual impact on their earnings. Right, because measuring how many times someone logs into a chatbot or tracking prompts per day, that tells you absolutely nothing about actual business value.

[00:17:57] It just measures activity, not outcomes. And a perfect example of how this goes wrong is Shopify. Oh, right, the mandate. Yeah, their CEO made AI usage a baseline expectation for performance reviews. Teams even had to prove they tried using AI before they were allowed to ask for new human hires. Look, if my boss told me tomorrow that my year-end bonus depended on my AI usage stats, I would absolutely start using ChatGPT to write, like, one-sentence Slack messages just to hit my quota. Exactly.

[00:18:26] Even if it takes me longer to prompt the AI than to just type, sounds good, myself. Doesn't a mandate like that just create a huge activity trap? If we connect this to the bigger picture, you've hit on the exact danger. When you make usage the target, people will just game the system to hit the target. Of course they will. To capture the redesigned dividend, your governance has to be tied to actual outcome measures. You don't measure prompt volume. You measure the cost per resolved customer case. Right.

[00:18:55] You measure the cycle time for a software release. You measure customer satisfaction and error rates. So you have to measure the actual work, not the tool being used to do it. Precisely. And this requires integrated governance. You can't have IT buying tools in a vacuum, HR dealing with the fallout of the changing jobs, and finance wondering where all the money went. They have to be talking to each other. Yes. Yeah. They have to govern AI, digital labor, and workforce design as one single integrated ecosystem. Man.

[00:19:25] So what does this all really mean for us? We have covered a lot of ground today. We really have. The AI value gap is incredibly real. We are spending billions for a 13% success rate. But that redesigned dividend is sitting right there. It's waiting to be captured. Yeah. And it's available to any leader, manager, or honestly any employee who is willing to do the hard, unglamorous work of rethinking how the work gets done. It is not about bolting a generic chatbot onto an outdated process. Nope.

[00:19:54] And it's not about using AI as an excuse to just slash headcount and boost margins for a quarter. It's about meticulously mapping those workflows, designing new roles, and having the discipline to pull the plug when things aren't working. And I'd say this applies to you listening right now, no matter what your job title is. Absolutely. In a human agent workforce, every single person basically has to become a work designer. You have to understand how to shape your own role, what psychologists call job crafting. Job crafting. I like that. Yeah.

[00:20:23] You have to actively figure out which of your tasks should be handed to an agent and which ones actually require your unique human judgment, your empathy, and your oversight. Your well-being and your long-term career depend entirely on how well you design that partnership. Which leaves us with a pretty wild thought to chew on as we wrap up today. Let's hear it. Think about this.

[00:20:45] As AI agents become completely commoditized, as literally every company on Earth gets access to the exact same super smart AI models for pennies on the dollar, the actual doing of the work becomes cheap. It just becomes a commodity. Wow. Yeah. So what happens to your career when the most valuable, highly compensated skill in the global economy is no longer the ability to do the tasks, but the architectural vision to design the space where humans and machines meet?

[00:21:13] Are you ready to stop being the builder and start being the architect?