You know, if you are running a mid market business right now, you are probably feeling this this very specific, very frustrating kind of tension. Oh, absolutely. It's everywhere. Right. Like you're spending real money on A.I. tools. You've got the enterprise subscriptions. Your team is using them daily. But but when you look at your bottom line or, you know, your overall operational leverage. Nothing has fundamentally changed. Exactly. Nothing has changed. You're waiting for this massive structural transformation and instead you're just getting, I don't know, slightly faster at doing the exact same things you've always done. Yeah, it is the defining frustration for business operators today. I mean, the software spend is climbing, but the underlying structure of the business remains, you know, totally static. So today our mission for this deep dive is to completely dismantle that tension. We've pulled a pretty massive stack of sources today. Yeah, we've got some great stuff. We really do. Ranging from Jack Dorsey's internal organizational restructuring memos at Block to Y Combinator's newest operational frameworks. Plus, we have some enterprise architecture data from Etlon and some highly sobering research out of MIT, RAN and S&P Global. Very sobering. Yeah. The goal here is to change your entire frame of reference regarding artificial intelligence. And to set the thesis for this, I want to read a direct quote from Y Combinator partner Diana Hu. Oh, this is the perfect quote to start with. It really is. She said, quote, AI should not be a tool your company just uses. It should be the operating system your company runs on. Every workflow, every decision and every process should flow through an intelligent layer that is constantly learning and improving. Right. So we are breaking down what we can call the tool frame. The tool frame, yeah. Which is, you know, the habit of using AI to do single isolated tasks slightly faster. And we want to move you into the operating system frame. This is about building a company that is fundamentally at its core an intelligence. To visualize this, I like to think about buying a brand new top of the line smartphone. OK. And then you only ever use the calculator app. Right, right. Like you are completely missing the fact that the operating system is designed to connect your calendar, your maps, your email, your contacts. It fundamentally changes how you navigate your entire day. If you only use the calculator, I mean, sure, you're definitely doing math faster. But you haven't changed the way you operate at all. Exactly. And the urgency to make this shift from the calculator to the full operating system, it really can't be overstated. Because the data we're seeing proves that simply bolting AI tools onto legacy workflows, it's a financial dead end. Let's get into those stats because they are rough. They are. Let's look at three major failure statistics from the sources that really anchor this reality. First, you've got this 2025 research from the MIT Manda Initiative. OK. They found that a staggering 95 percent of generative AI pilots fail to scale to production. Wait, 95 percent. That's I mean, that isn't just a tech glitch. A 95 percent failure rate indicates a structural flaw in how these pilots are even being integrated. Exactly. It points to a massive architectural mismatch. And honestly, it gets worse. Great. Yeah. The RAND Corporation analyzed the enterprise market and found that 80.3 percent of AI projects fail to deliver their intended business value. And then S&P Global tracked the share of companies that are abandoning most of their AI initiatives entirely. That number jumped from 17 percent to 42 percent in a single year. Over 40 percent just walking away. I mean, if the underlying technology is as powerful as we know it is, this level of abandonment requires a diagnosis. Right. What is actually breaking down inside these companies? Well, the breakdown happens because of the the probabilistic nature of treating AI as an isolated tool. What do you mean by probabilistic? So say you buy a generative AI license for your marketing team to write copy. They get a temporary like 15 to 30 percent productivity bump. But the output is probabilistic. The tool is just guessing the next best word based on a localized prompt. Right. It doesn't actually know your business. Exactly. It doesn't have a deep structural understanding of your actual company. Furthermore, your company's information remains entirely fragmented across different departments. Just siloed away. Yeah. You remain what systems engineers call an open loop. Open loop. OK. You execute a task, but there is no automatic feedback mechanism feeding the results of that action back into the central brain of the company. Information just bleeds out everywhere. OK. I want to push back on that a little bit. Sure. Go ahead. Because if I am an operator listening to this, I might be thinking, well, if my marketing team is writing their campaigns 30 percent faster or my developers are clearing their backlog quicker, isn't that still a massive win? It feels like one. Yeah. Right. So why should I care if my business is technically an open loop as long as I'm saving my employees hours of time every single week? I get that. But saving time on a fundamentally fragmented process doesn't actually improve the business. It just gets you to the bottleneck faster. Oh, just rushing to the red light. Precisely. If marketing is churning out campaigns 30 percent faster, but the sales team still doesn't know what marketing is promising and the product team is building something completely disconnected from both of them, you haven't solved the core operational drag. You've just accelerated the creation of siloed work. That makes total sense. Yeah. A tool makes a localized task faster, but an operating system makes the entire organization smarter. So we are essentially taking ultra fast modern AI and layering it over an incredibly slow ancient human routing system. That is exactly what we're doing. Which brings us to this fascinating piece of historical analysis from the sources, specifically from Jack Dorsey and Roloff Botha, regarding why Block is totally restructuring. Oh, this part is so interesting. It's wild. They trace the modern corporate organizational chart all the way back to the Roman army's contubernium. Yeah, the contubernium. It was a unit of eight soldiers who shared a single tent led by one decanus. And then 10 of those groups formed a century of 80 men led by a centurion, scaling all the way up to a legion. But this military structure was, at its core, an information routing protocol. Because they didn't have radios. Exactly. It was built around the biological limitation known as the span of control. Span of control. Right. The fact that one human brain can only effectively manage and route information for about three to eight other people. And that biological limitation literally became the blueprint for corporate America. Like the sources night, the Prussian general staff adopted it in the 1800s to support generals, essentially inventing middle management. And then in the 1850s, the American railroads copied the military because they had trains literally crashing into each other across 500 miles of track. They needed a way to pass messages safely. Right. So they needed a structured hierarchy for information to flow from the station master all the way up to the executives. So for 2000 years, middle management has served as a human A.P.I. A human A.P.I. Let's pause on that phrase. Yeah, because in software, an A.P.I. allows two different programs to communicate and share data instantly. You are saying that in a traditional company, middle managers are functioning as biological software connections. That's exactly what they're doing. They aggregate context from their direct reports, summarize it, pass it up the chain, get a decision and pass back down. And human beings are incredibly lossy A.P.I.'s. Highly lossy. Every time we pass information, the context degrades. It's slow. It is subject to office politics and it creates massive latency. Like a bad game of telephone. Exactly. We've relied on this human routing system for two millennia just because we had no other choice. But the operating system frame suggests we finally do. Which is the solution block is building with their company world model. Yes. And it aligns perfectly with the architecture Elan provides with their enterprise A.I. memory layer. In an O.S. framed company, the artificial intelligence replaces the information routing that middle managers use to perform. OK, so how does that system actually work under the hood? I mean, how does an A.I. comprehend an entire mid-market manufacturing or sauce business? Well, it requires transforming your company's daily exhaust into machine readable data. Daily exhaust. Yeah. Like every project management ticket you close, every Zoom call transcript, every customer support resolution. It all becomes a captured artifact. OK. The A.I. system ingests these artifacts using technologies like vector databases and semantic mapping. Yeah. It doesn't just store the text, though. It turns those conversations into mathematical relationships. So it's connecting the dots. Right. It connects a complaint from a client call directly to a specific code push from the engineering team. It builds a centralized, multidimensional graph of your business in real time. And this creates that closed loop we talked about, where the intelligence layer instantly feeds the necessary context to whoever needs it without requiring a manager to hold a status meeting. But wait, there is a massive red flag here that the Atlan research highlights. If a centralized A.I. is reading every single artifact across the company, we run into immense security and privacy walls. Oh, absolutely. It's a huge risk. Right. If you just let an A.I. agent hallucinate based on secure financial data or H.R. records, the liability is staggering. This introduces what the sources call the governance imperative. Yeah. The operating system model completely collapses without a governed, centralized context layer. OK. Break that down for me. So what many companies do today is give every individual A.I. agent its own isolated memory. Now, that is a compliance nightmare. Because they get confused. Yeah. If Agent A defines recognized revenue differently than Agent B, they will contradict each other. But more severely, ungoverned A.I. creates direct legal exposure. Like the EU AI Act. Exactly. The looming European Union AI Act carries penalties for compliance failures reaching 35 million euros or 7 percent of global turnover. If your system makes a decision based on regulated data, you must have lineage. You have to be able to audit exactly why it made that decision. And the sources also note that 38 percent of executives recently surveyed reported making incorrect business decisions based on A.I. hallucinations. Yes. Almost 40 percent. Almost 40 percent of leaders are making bad calls because the tool simply made something up. But here's the thing. A hallucination in a business context usually isn't the A.I. just dreaming things up. It is the A.I. lacking the correct governed business context. Oh, it just doesn't have the full picture. Right. A centralized A.I. memory layer uses role based access controls to ensure the A.I. only sees what it is legally allowed to see, guaranteeing that every agent across the organization is drawing from a single compliant source of truth. OK, pulling this together. We are migrating from an architecture where intelligence is spread out across hundreds of people and a human hierarchy routes that intelligence to an architecture where the intelligence lives inside the system itself and the people operate on the edges of that system. Beautifully put, the system holds the context, allowing the people to take immediate action. But if the A.I. operating system is handling the routing and holding the context, the traditional org chart essentially collapses. The people don't need to be routers anymore. They really don't. So what exactly do the human roles look like in this new environment? We see the organizational chart undergo a profound compression. Both Jack Dorsey and Y Combinator's Diana Hu describe human roles compressing into three highly distinct archetypes. Three archetypes. OK, what's the first one? First, we have the individual contributors or the builders. These are deep specialists making the actual product. OK. But their workflow shifts radically. Builders no longer bring slide decks to alignment meetings. They bring working prototypes generated by A.I. software factories. The software factory concept. This completely changes the baseline of productivity. Completely. Instead of a human writing line after line of code or manually drafting 100 variations of a marketing asset, the human builder writes the specifications and the tests. They define the parameters of success. And then the A.I. agents generate the code or the asset, test it against the human's parameters and iterate automatically until the tests pass. The human defines the what and the A.I. executes the how. So if your builders are churning out product using A.I. factories, someone still has to own the actual business outcomes. Which brings us to the second role. Right. The directly responsible individual or the D.R.I. You can think of them as the owner. They own specific cross-cutting customer problems. Like what? Like their job title isn't director of retention anymore. Their job is fix merchant churn for the next 90 days. Wow. One person, one metric. Nowhere to hide. Nowhere to hide. And the defining characteristic of the D.R.I. is that they do not manage people. They don't? No. Because the A.I. operating system holds the context of the entire business, the D.R.I. can instantly pull insights, data and resources from the system to solve their specific problem. They completely bypass the need to coordinate through five different department heads. That is wild. Okay. So that leaves the third role. Yes. The player coach or the founder. This is the listener. In this compressed structure, you are not sitting in endless status roll up meetings. You are actively building alongside your team, coaching them on how to leverage the system and modeling what frontier A.I. capability looks like. It's a completely different way of managing. It is. And this compression also completely upends traditional business economics. The sources describe a shift from wanting to be headcount maxing to token maxing. Token maxing, meaning optimizing for compute power instead of payroll. Right. You are measuring your leverage by how many A.P.I. calls you are making to the A.I. model. Exactly. You should actively want an uncomfortably high A.P.I. bill. Uncomfortably high. Yes. Because a massive A.P.I. bill means your intelligence layer is successfully doing the heavy lifting of routing, coding and analyzing. That compute spend is replacing what would have been an incredibly slow, inflated and expensive human middle management layer. Okay. But there is a very real risk here that we need to confront. We are talking about stripping away layers of management and handing the overarching context of the business to a machine. But business is incredibly messy. What happens to human intuition? That's the big question. Right. What about reading the tension in a boardroom negotiation or sensing the cultural dynamics of a team or, you know, making an existential call where the math says one thing but your gut says another? An A.P.I. call cannot do that. It's a vital concern. And frankly, this transition is going to break a lot of legacy company cultures because it is not a clean, seamless swap. Yeah, I'd imagine not. But Jack Dorsey's framework for block directly addresses this messy reality. He talks about operating at the edge. The centralized system holds the intelligence, but the humans operate at the edge, the precise point where your company makes contact with reality. Oh, where the map meets the territory. Exactly. Humans can sense dynamics the model simply cannot perceive. Trust between partners, nuanced ethics, high stakes, existential risks where the cost of being wrong is catastrophic. The A.I. operating system does not exist to make those judgment calls. The OS exists to give the human at the edge the perfect, intently synthesized context they need so that the human can apply their intuition without waiting two weeks for a data team to run a report. So you elevate the human to do only what the human is uniquely qualified to do. Precisely. So how is a mid-market operator actually executing this without halting their current revenue engine? I mean, you cannot just pause a functioning business to rebuild your entire data infrastructure in midair. You can't, which is why mid-market companies actually have a massive hidden advantage over legacy enterprise giants in this transition. How so? Well, a Fortune 500 company has decades of entrenched bureaucracy, siloed data systems and thousands of employees. Shifting them to an A.I. operating system is like trying to turn a battleship in a canal. Good luck with that. Right. And startups have the luxury of building this from day one, but they lack data. The mid-market operator is in the Goldilocks zone. You have the agility to make fast, sweeping decisions combined with real, established operational data to optimize. So what's the playbook? What are we seeing work best for operators in this space right now? It's an incredibly targeted approach. The sources lay out a three-step start this week framework. Step one, do not attempt to install an overarching A.I. brain on a Monday. OK, good to know. Instead, you isolate a single high-stakes workflow, might be the sales pipeline or the hiring process. You map out the human decisions happening within that specific workflow and you figure out exactly what context is present in the room when those decisions are made. OK, so step two is looking for context leaps. Exactly. Context leaks. The moments where partial or missing information directly leads to a bad call. Like, did sales promise a feature on a call that product doesn't know about, leaving customer success completely blindsided a month later? You find out exactly where the open loop is bleeding context. Yes. And notice step three isn't about automating anything. It's not. No. The goal is simply making the business legible to the machine. Oh, right. Capturing the artifacts. You have to capture the artifacts. You introduce A.I. note-takers to every meeting. You migrate siloed direct messages into searchable public channels and you build custom dashboards that unify your revenue and operational data. You are laying the plumbing. Because the A.I. operating system cannot learn from a conversation that was never recorded or a decision that was never documented. Exactly. Once you make the foundation of your company's world model legible, then you can apply the intelligence layer to start routing that context automatically. Let's bring all this together. We started by examining the tension of spending heavily on A.I. without seeing operational leverage. Right. We established that the tool frame, you know, bolting A.I. onto old processes to do isolated tasks faster. It's a trap. It caps your R.O.I., leaves you operating as a lossy open loop and directly contributes to the 95 percent failure rate of generative A.I. pilots. And the alternative is the operating system frame. By shifting your architecture, you compress the org chart. You eliminate the slow, lossy human routing of middle management and you replace it with a centralized, governed A.I. memory layer. Creating that closed loop system. Yes. A system where your company compounds in value and intelligence every single day. Leaving your human team at the edge, fully equipped with context to do the intuitive, high stakes work they were actually hired to do. Exactly. And I want to leave the listener with a final thought to ponder as you look at your own operations today. Let's hear it. Think about your fiercest competitor in the market right now. OK. If they adopt this operating system frame before you do, their company world model is going to start compounding its understanding of their business and the market every single second of every day. Wow. They will be building an intelligence that never sleeps, never forgets a customer preference and never drops a context leak. While they are building an autonomous intelligence layer, can you really afford to just keep paying for an A.I. tool that helps you write faster emails? That is a reality check that should keep every operator up at night. The Calculator app won't save you if your competitor is using the whole smartphone. Thank you so much for joining us on this deep dive into the operating system frame. Don't wait on this. Start mapping out your first workflow and plugging those context leaks tomorrow morning. I'll see you next time.