After the AI Layoffs: Three Steps to Rebuild Work, Not Just the Org Chart

[00:00:00] What if I told you that firing 20% of your staff and replacing them with artificial intelligence is basically guaranteed to make your company slower, more error prone, and less productive for the entire next year? I mean, it completely shatters the narrative we've been fed, doesn't it? It really does. Because the prevailing assumption out there is just, you know, the moment you automate a fraction of the work, you just subtract that exact fraction of your workforce. Right, just simple math.

[00:00:28] Exactly. And everything just hums along perfectly. But that's not what happens at all. And if you're sitting in a management meeting right now, you know, and someone proposes a massive head count cut because you just bought a shiny new AI platform, well, this deep dive is exactly the data you need to slide across the conference table. Absolutely. Because today we're unpacking an incredibly timely stack of research from Dr. Jonathan H. Westover. We're looking at what actually happens after an AI related layoff.

[00:00:57] It's such an important topic. Right. Because Wall Street always asks, how much money did we just save? But the author argues that if you're only looking at the balance sheet, you are completely missing the structural reality of your own company. You're missing the bigger picture. Yeah. The real question you should be asking is, what can this organization now do better and what has it become less able to do? Which is such a critical shift in mindset. Yeah.

[00:01:26] Because a smaller organization is rarely a better designed one by default. Right. It's just a comforting illusion for executives. So today we are walking through a really practical evidence led framework to rebuild work itself. Not just shuffling boxes on an org chart. Exactly. We're actually focusing on organizational capability, not just head count. Okay. Let's unpack this because before we can even talk about rebuilding capability, we need to like dismantle a massive illusion about how AI impacts productivity in the first place. We really do.

[00:01:55] I was reading through the source article and I realized that the way companies approach these AI layoffs is, well, it's like taking the engine out of a car to make it lighter. Right. Right. Oh, that's a great way to put it. And then expecting it to go faster just because you painted racing stripes on it, it just doesn't work. It doesn't work at all. And that analogy gets right to the heart of the invisible structural damage that layoffs cause. Yeah. Decades of research actually back this up.

[00:02:20] There was a comprehensive 2010 study that reviewed years of downsizing data and they found zero straightforward relationship between workforce reductions and improved organizational outcomes. Wait, really? Zero relationship? None. A sudden drop in headcount simply does not predict future performance.

[00:02:40] Which is wild considering how often we hear CEOs pitch layoffs as this, you know, necessary, almost cleansing reset that's going to unlock future growth. It's probably one of the most persistent myths in modern business. Yeah. And when you introduce AI into that mix, people just assume that technology immediately fills the void left by the people who walked out the door with their cardboard boxes. But it doesn't.

[00:03:03] No, it really doesn't. And researchers from MIT specifically looking at what happens when general purpose technologies enter the workplace. They proved this. And general purpose technologies, that's stuff like electricity, the Internet and now AI. Exactly. Technologies that change everything. And what they identified is this phenomenon called the productivity J curve. The J curve. Okay, let's break that down for everyone listening.

[00:03:29] The J curve basically means productivity actually drops down before it eventually climbs up. Yeah. The shape of the letter J is the perfect visual for this. Okay. Because buying AI technology and actually realizing its value, those are two entirely separate events. Right. Buying it is easy. Exactly. The purchase is instantaneous. You just sign a contract. Yeah. But integrating it, that requires massive complementary investment. So that's the dip in the J. That's the dip. You dip into the bottom of the J because you have to completely redesign your internal processes.

[00:03:58] You have to teach new skills to your workforce. Right. And you have to create entirely new organizational arrangements just to support the tech. So during that dip, everyone is basically scrambling to figure out how the new tools actually work. Workflows are totally disrupted and output just plummets. Plummets. Plummets. And on top of that, you're dealing with the loss of institutional memory from anyone you just laid off. Oh, wow. Yeah.

[00:04:23] So the value realization, the upward swing of that J is a long, steep climb that only happens well after that initial adjustment period. So if an AI doesn't instantly replace human output, we really need to completely redefine what we mean when we say an AI, quote unquote, does a job. We have to. Because the headlines make it sound like a robot is just rolling into a laid off employee's cubicle, sitting at their desk and doing literally everything they did.

[00:04:51] Yeah. The robot in the cubicle visual is the core misunderstanding here. Right. So let's look at the numbers driving this panic. The International Labor Organization estimated that one in four global workers are in occupations exposed to generative AI. One in four. One in four. But I have to push back on how that gets interpreted in the media, you know, because exposure doesn't mean replacement, right? Not at all.

[00:05:12] Like just because an AI can beautifully summarize my meeting notes and pull out action items doesn't mean it can handle my angry clients when a massive project goes completely off the rails. Exactly. And that distinction is exactly why the author of our source material separates work into four specific units of analysis. OK. We really have to stop talking about whole jobs and start looking at the granular components. Right. Break it down.

[00:05:36] So first you have tasks. A task is a single specific activity like drafting a status update email or verifying an invoice number. Simple stuff. Right. Second, you have workflows. A workflow strings those activities together to reach a broader outcome. Like resolving that customer complaint from start to finish. Exactly. Third, you have roles which bundle a bunch of workflows, decisions and relationships together. OK. And finally, the organizational structure which dictates how all those roles interact.

[00:06:04] So AI is really just attacking the bottom of that pyramid. It's automating the tasks, not the whole role. Right. It primarily affects tasks, which then, of course, transforms the role because your daily mix of activities changes. Makes sense. But it rarely replaces entire workflows from end to end without human intervention. Like the angry client example. Exactly. Think about that account manager role.

[00:06:29] It involves preparing meeting notes, researching the client's industry, negotiating a sensitive contract renewal, and, you know, reading the emotional temperatures of the room to repair a damaged relationship. Right. You can't outsource reading the room. No. An AI might handle the industry research task in seconds, but it is not going to independently own the emotional relationship repair workflow.

[00:06:52] Here's where it gets really interesting, though, because when you look at the data on how AI handles those individual tasks, the results are wildly split. They are. The sources highlight two studies that paint a really confusing picture if you just glance at them. First, there's a study looking at over 5,000 customer support agents. Right. And it found that giving those agents a generative AI assistant increased their issue resolution by 15% per hour. Which is huge. On paper, that is a massive immediate win.

[00:07:20] But then you pivot to this recent Boston Consulting Group study examining over 700 highly educated consultants. Yeah. And those BCG numbers are the ultimate cautionary tale. Tell me about it. They illustrate a concept researchers call the jagged technological frontier of AI. The jagged frontier. Right. Imagine AI capability as a coastline. Some areas jut out really far into the ocean. Those are the tasks AI is phenomenal at. Like summarizing data. Exactly.

[00:07:49] But other areas form deep recessed bays. Those are tasks where AI completely fails. Okay. I like that visual. So when those BTG consultants used AI for tasks that were firmly on the solid ground of that frontier within the system's tested capabilities, the results were stellar. They crushed it. They did. They completed 12.2% more tasks and worked 25.1% faster. Working 25% faster in a high stakes consulting firm is a massive competitive advantage. It's incredible.

[00:08:16] But here is where the jagged frontier gets treacherous. When researchers tested those exact same consultants on complex tasks that were just slightly outside the AI's capabilities. Huh. So, stepping off the edge of the cliff, the consultants using AI were approximately 19 percentage points less likely to reach the correct solution than the control group. Wait. 19% worse than the people not using AI at all? Yes. Explain the mechanism there.

[00:08:45] Why did the AI make highly educated consultants 19% worse at their jobs? Was it just like generating bad data? Well, it generated bad data, yes, but with extreme confidence. Oh. AI hallucination is part of it, but human psychology is the real culprit here. What do you mean? Because the AI was so incredibly capable on that first set of tasks, the consultants let their guard down. They got lazy. Basically, yeah. They cognitively offloaded their critical thinking to the machine.

[00:09:12] The AI's confident tone just lulled them into this false sense of security, so they accepted wildly incorrect answers without verifying them. Oh, man. They essentially fell asleep at the wheel while driving 25% faster. That is a phenomenal takeaway for anyone listening to this deep dive right now. When you bring AI into a workflow, you absolutely must measure both acceleration, A and D correction. Yes. If you only measure how fast your team is working now, you might just be producing catastrophic mistakes at a record pace.

[00:09:42] Speed without correction is just organizational sabotage. Exactly. And because AI's effectiveness is so jagged, you know, amazing at data processing but actively harmful at complex reasoning, leaders cannot just guess at the Newark chart. They really can't. You can't just cross off five names, buy a software license, and assume the remaining three people will just magically figure out how to catch the machine's mistakes. No, they need a rigid, evidence-based framework to rebuild. Right.

[00:10:10] And the source material lays out a three-step framework for this. Baseline, model, and equip. So let's start with step one. Establish the baseline. Okay. This sounds fundamental, but most companies actually fail right here. Yep. You have to start by mapping outcomes before you even glance at the reporting lines. So look at the goal, not the people. Exactly. Understand what your customers or internal stakeholders actually need delivered and then work backward to the specific granular activities required to deliver it. Okay.

[00:10:38] And crucially, you have to identify where independent human review is genuinely needed versus what is merely inefficient duplication. Break that down for me. Independent review versus duplication. Think about it this way. If you have two different people manually entering the exact same client address into two different sales databases, that is mere duplication. Right. That's just busy work. It's an inefficient, broken process. Let the AI automate that entirely. Sure.

[00:11:06] But if an AI drafts a highly sensitive, multimillion-dollar legal contract and a human lawyer reads every line before sending it to the client, that is an independent review. That's quality control. Yes. That human check is a vital safety mechanism. You cannot eliminate the lawyer just because the AI successfully generated the first draft. That makes total sense. You can't just consolidate roles because the tasks look similar on a spreadsheet.

[00:11:30] And there's a brilliant real-world example of this baseline mapping in the sources from Schneider Electric. Oh, I love this example. Right. In their 2023 human resources report, they detailed how they built an AI-supported open talent market. Yeah. They used AI to map all the invisible skills across their workforce and then connected employees with internal projects and mentorships. And the results were amazing. Yeah.

[00:11:55] In just one year, they had 3,000 employees working on internal projects and 4,000 mentorships formed. That is hard evidence of an organization actually making its internal capabilities visible. Yeah. The key insight for leaders here is to test internal mobility before recreating an old vacancy and conversely before making permanent cuts. Like, look at what you already have. Right.

[00:12:19] Ask yourself if the work actually requires a brand new hire or if it's just a temporary assignment that could be filled by a different combination of skills that already exist three floors down. Which brings us to the bottleneck of all this planning. Step two, continuous modeling. This is where things get tricky. Because the math of AI savings is incredibly deceptive. Let me give you a scenario. Okay. Say you have a department of 100 employees. You roll out an AI tool that saves every single one of them two hours a week. Sounds great.

[00:12:49] On a spreadsheet, 200 hours of saved time looks like you can comfortably cut five full-time jobs. Right. The math adds up. But you can't just fire five people and somehow tape those spare two-hour slivers together into a functional workflow. No. It completely defies the laws of operational physics. Right. You can't sweep up 200 hours of scattered minutes from across 100 different desks in different time zones with totally different skill sets and hand it all to one person. But it doesn't work. That released capacity is fragmented.

[00:13:19] So how do you actually model that fragmented capacity into something useful without just firing people? IKEA's Inca Group is the prime example of doing this right. Okay. Tell me about IKEA. So in 2023, they reported that their AI chatbot, they call it Billy, was independently resolving about 47% of all customer inquiries. Almost half. Almost half. I mean, in most boardrooms, resolving half your calls with a bot triggers an immediate celebratory 47% cut in the call center staff. Oh, absolutely.

[00:13:49] But instead of looking at that saved time and cutting headcount, they looked at the human potential they just unlocked. Okay. They took 8,500 call center workers and actually re-skilled them for complex, higher-value tasks. Wow. They trained them in remote interior design, digital retail sales, and complex relationship building. So they took people answering basic password reset questions and turned them into remote interior designers. Exactly.

[00:14:15] They used the release capacity for service improvement and revenue generation, not just to lure their employment costs. That is such a better approach. It is. When you model the future of your organization, you have to explicitly compare growth, risk reduction, and service improvement against simply making your workforce cheaper. I love that. Using AI to make your people more valuable to the customer, not more disposable to the company. Perfectly said.

[00:14:38] But all of this mapping and modeling on a whiteboard is completely useless if the managers on the floor don't know how to execute it, right? Right. So how do we equip them? Because just giving a middle manager a shiny new data dashboard isn't going to cut it. No. Visibility without time or authority is not enablement. It's just micromanagement. It's just a new flavor of micromanagement, yeah. Fair. The research references the NIST AI risk management framework to explain this. NIST?

[00:15:06] That's the National Institute of Standards and Technology, right? Right. And they provide benchmarks for how to safely deploy these systems. Okay. Their framework organizes work around governance, measurement, and management. For middle managers, that means they need the actual explicit authority to reject an AI's recommendation. Give me a scenario where a manager actually needs to override the machine. Okay. Let's say your internal AI system suggests that a specific employee is a high flight risk. Okay.

[00:15:35] Or that a certain worker is the statistically perfect skill match for a new project. Right. The AI says this is the person. Exactly. The manager needs the power to look at the human context, the nuance the AI didn't catch, like the fact that the employee just had a baby and is sleep deprived, not actually disengaged. Oh, that's a huge difference. Massive. And the manager needs to be able to say, no, this AI is wrong, and here is the recorded reason why. Right.

[00:16:03] And they need to be able to do that without being penalized by upper management for ignoring the multi-million dollar software. Right. They can't be afraid to push back. And they need the protected time to actually do that kind of nuanced review. Protected time to learn alongside their team is totally non-negotiable. Yeah. If you expect managers to coach more people, carefully review AI-assisted outputs for those jagged frontier errors we talked about, and manage internal redeployments like IKEA did. That's a lot of work. It is.

[00:16:32] You have to explicitly tell them what administrative work or individual production targets will be reduced to make room for that. You cannot just pile AI supervision on top of their existing 50-hour work week. You really can't. Let's pivot to the human element here, though, because equipping managers relies on the workers actually cooperating. That is the tricky part. Right. If you are a frontline worker and you are terrified an AI is going to take your job, how does a leader

[00:16:58] get you to honestly engage in this whole mapping and modeling process? If you challenge. Won't employees just hide their actual workflows like, nope, my job is a complete mystery and AI could never possibly do what I do? I mean, it is a completely rational survival mechanism. Totally. Why on earth would you willingly train your own replacement? Exactly. And this brings in some really famous research on psychological safety. Basically, studies of manufacturing teams have shown that psychological safety, which is the

[00:17:27] shared belief that a team is safe for interpersonal risk-taking, is directly associated with learning behavior. Okay. So for any of this AI redesigned to work, employees must feel structurally safe questioning assumptions. Meaning they won't get fired if they admit the AI can do 30% of their job? Exactly. Or if they need to disclose that the AI is making massive mistakes. They have to be able to challenge an unreliable output without leadership viewing it as disloyalty

[00:17:56] or stubborn resistance to change. Which is hard for some leaders to accept. It is. But the OECD, they ran surveys recently across manufacturing and finance sectors. And they found a direct measurable link between worker consultation and better AI outcomes. So talking to your employees actually works. It really does. You have to commit to your employees that they will have opportunities to correct inaccurate assumptions about how their work actually happens on the ground. All right.

[00:18:25] So to make this incredibly actionable for the leaders listening to this deep dive right now, the research outlines a 90-day test and learn operating rhythm. Yes. Let's walk through this because pacing is everything. It really is. So days 1 through 10 are all about stabilizing. If layoffs have already happened, you have to reconfirm ownership of critical obligations. Right. Don't let important customer commitments fall through the cracks just because Gary got laid off and no one knows who owns his inbox anymore. Stabilizing stops the immediate bleeding.

[00:18:55] That is crucial. Okay. Then, days 11 to 30 focus on establishing the baseline. That's mapping those workflows. Separating the inefficient duplication from the deliberate necessary checks. But realistically, can you map an entire company's workflow in just 20 days? I mean, that sounds impossible. Oh, you absolutely cannot map everything. Okay. Good. The framework explicitly says to focus on priority workflows. Got it.

[00:19:23] You find the critical paths that deliver the most value to the customer. You validate the real skills of the team working on those specific paths. Okay. That makes it feasible. Then, days 31 to 60, test alternatives. Right. This is where you run bounded pilots with the AI, begin targeted development, and give managers the authority they need. What does a bounded pilot actually look like in practice, though? It means you don't roll the AI out to the entire company on day one. That sounds like a recipe for disaster anyway. Exactly.

[00:19:52] You pick one team, one specific workflow, and you test the AI's capabilities there. Just a small sample. Right. You measure the acceleration, and crucially, you measure the correction rate. Like we talked about before. Yes, you figure out exactly where the jagged frontier is before you expose your whole client base to it. Smart. Then finally, days 61 to 90, decide. The decision phase. Based on the data from that bounded pilot, do you expand the AI rollout, revise it, pause

[00:20:19] it, or just reverse it entirely based on the quality of the work and the employee experience? And this is where leaders need to be very, very careful with their messaging. How so? At the end of those 90 days, you should never stand up in a company town hall and claim the restructuring is complete. Because technology is going to keep changing anyway, right? Exactly. The environment is constantly shifting. Right. What leaders should do is clearly explain to the team what was learned during the pilot, what assumptions they had to change based on the evidence, and where resources will move next.

[00:20:49] It's an ongoing process. It's a continuous cycle of hypothesis and testing. It is definitely not a one-time, cut-and-paste organization chart event. So what does this all mean? We have covered a massive amount of ground today, from the J-curve to the jagged frontier. We really have. And to synthesize the core of this research, AI layoffs are just not a magic wand for productivity. They really aren't.

[00:21:14] You don't just shrink the org chart, buy a chatbot license, and watch your profit margins soar. Right. Rebuilding successfully requires truly understanding the granular tasks humans actually do. Yes. Planning for the inevitable, sometimes invisible errors the AI is going to make when it hits the edge of its capabilities. That jagged frontier. Exactly. And focusing your continuous modeling on creating a more capable organization, not just a cheaper one. That is the ultimate goal.

[00:21:43] And as we wrap up, I know there's a final lingering thought you wanted to leave everyone with today, something that builds on the framework's mention of supervised practice. Yes. This is something I've been thinking a lot about. Okay. The framework stresses that less experienced employees desperately need supervised practice, right? And structured exposure to build their human judgment. Right. They have to learn the ropes. Exactly. So consider the long-term alternative.

[00:22:08] If an organization successfully automates all of its entry-level routine workflows today, just to save a few dollars on the balance sheet, where will the next generation of experts come from tomorrow? Oh, wow. If the ropes of human judgment are entirely outsourced to machines, how does the human workforce of the future ever learn the ropes in the first place? That is chilling, honestly. But it's the exact question every leader needs to be asking right now before they decide to

[00:22:36] knock down that load-bearing wall just to make the room look a little bigger. Absolutely. Well, thank you all for joining us on this deep dive. Keep questioning the headlines. Keep looking beyond the org chart. And we'll catch you next time.