The New Era of Jobs: Organizational Singularity | EP #258

1h 1m 158K views 26 May 2026

Is it worth your time?

You get an operational framework for building AI-native digital twins, eliminating coordination drag, and redesigning corporate workflows around intelligence.

This conversation examines how artificial intelligence forces companies to abandon traditional hierarchies and restructure around intelligent agent workflows. Ismail explains the organizational mechanics required to survive this disruption over the next decade.

Death of the modern company. Analyzing corporate structures, Ismail argues that traditional companies built on Ronald Coase's theory of coordination costs have broken down because AI drives down execution costs. He warns that over 80 percent of corporate AI projects fail because legacy organizations merely automate human bottlenecks. He notes that coordination is more expensive than execution today, making human-to-human workflows obsolete.

ExO 3.0 intelligence architecture. Detailing the destination architecture, Ismail outlines an intelligence stack modeled on the OODA loop that acts as the core engine for AI agents. He explains that this architecture requires a govern and assure loop with a trusted eval architecture and human review queue to prevent agents from going rogue. He adds that humans rise to perform dashboard oversight and exception handling while agents execute tasks.

Digital twins at the edge. Addressing enterprise transformation, Ismail contends that trying to transform an existing company fails because of internal corporate immune systems, so organizations must build an AI-native digital twin at the edge. He expects middle management to absorb 60 percent of workforce reductions as coordination roles drop by 90 percent. He calculates that once the digital twin runs properly, performance improvement should reach 100 times or higher.

Rewrite methodology for workflows. Outlining operational steps, Ismail describes a six-step rewrite methodology starting with backcasting to envision an AI-centric company and scoring organizational drag. He points out that companies must map prescriptive workflows and migrate them one by one into the new digital twin structure. He estimates this full transition across the corporate sector will take 5 to 7 years.

Intro

It's Salim's sweet 16 birthday. We're going to talk about something we've been teasing on the Moonshots podcast: the organizational singularity. This is useful for everyone, whether you're a Fortune 500 CEO, an entrepreneur in a small company, or a parent advising your kid on where to work.

When we wrote the Exponential Organizations book, we didn't realize how prescient it would be, but over 10 to 12 years, we were dead on. Now that we see a gen ticket on the future of intelligence, we've worked with my entire community to determine what the organization looks like. I've been saying AI has killed the modern company. The Fortune 500s are out there, but they haven't gotten the memo yet. When the comet hit, the dinosaurs didn't vanish overnight; it took a few generations to die out. The key is what you do once you understand that everything has changed.

For 100 years, we've run organizations on a theory set coined by Ronald Coase in his 1937 Nobel Prize-winning paper, "The Nature of the Firm." He theorized that big companies get bigger because transaction and coordination costs inside a company are cheaper than outside; since everyone is on payroll, you can order them around to get better work done. For 80 years, we followed that. Deep thinkers like Simon discussed organizational boundaries, Clay Christensen's Innovator's Dilemma showed how smaller companies can deliver cheaper products, and Stanley McChrystal addressed coordination at scale without losing emotional connection.

EXO.1 used community, crowd, and AI to pull Coase sideways and extend our reach. Consider how X Prize coordinates external teams, or how Uber's mission-critical function—matching driver and passenger—happens out in the wild rather than inside the organization. When technology enables that, you can scale. We found ways of extending Coase's law, and Jack Dorsey did the same with Block, but we've concluded that the whole thing breaks in the face of gigantic AI. Coase's law no longer applies.

If you build a website inside a company, you face layers of meetings and approvals from branding, privacy, and IT. Today, you can step outside the company, use Vercel at home for five minutes, and get it done for free. It can know your brand guidelines and design taste, spinning up a dozen versions to try in the market. As one fantastic tweet put it: building the feature is now cheaper than having the meeting about the feature. Coordination is more expensive than execution today, especially as AI drives down the cost of execution.

The Fiduciary Wedge - A Gap Between Human Judgement and What AI Can Do

We have a term called the fiduciary wedge. Coordination and execution costs have become low—the primary reasons for organizations over the last 100 years—but you still need a purpose, fiduciary, and liability container, like SPVs for investments. Companies are increasingly becoming these containers that hold legal and fiduciary liability.

The fiduciary wedge is the gap between what AI can do and human judgment and liability. You still need an organizational structure and legal entity, but the question is what is inside that container: assets, IP, agents, and some humans. These agents make API calls, hack things, and call people, like when Alex Finn's AI called him. This evolves the ExO book from 1.0 to 2.0, and now to what we call the organizational singularity.

The Organizational Singularity

You can learn more at organizationalsingularity.com. We are releasing the book as an AI because a book is a static thing that would be out of date the minute it is published. We are launching a Claude skill because something changes the game every three days.

Today, over 80 percent of AI projects in companies are failing miserably because existing companies are geared towards human-to-human-to-human workflows. It is like when we first created television and simply put radio announcers on TV; these projects fail because you are moving AI into legacy organizations and automating legacy human bottlenecks. You need an AI native environment. We had to step back and realize the entire EXO model, Coast, and all previous thinkers would break.

To be clear, if you do not retool or restart your organization in this fashion, you will be disrupted by someone who will eat your lunch. If there is a high-margin line of your business that two guys with Open Cloak could replicate in 60 to 90 days, call us, because I guarantee those two guys are out there disrupting Dropbox. It is a whole new world and a whole new ballgame. By organizational singularity, we mean that instead of coordinating the company around hierarchy, you organize it around intelligence and the Massive Transformative Purpose (MTP) from the original book.

ExO 3.0 Model: The Destination Architecture for AI Agency

MTP becomes a protocol and guide for AI and human agents to act properly. Consider the early days of Uber; their MTP was that everybody should have a private driver. However, if you always ordered surge pricing, they would always charge you that rate even if a cheapskate standing next to you got a lower price. Pushing those ethical boundaries is now guided by this MTP architecture.

We have Drive, the intelligence scaffolding and engine, and Shape, which defines how the organization works. In the diagram, you'll see an architecture with six layers of that core intelligence engine. The best analogy is Boyd's OODA Loop—observe, orient, decide, act. This core flywheel is the center of the solve-everything framing; whatever you put into that loop creates a positive feedback loop on everything else. We created the intelligence stack to act like the OODA Loop for constant learning, wrapped in a critical harness of govern and assure to ensure agents don't go rogue, like the railway agent that deleted volumes of rental car data.

At the heart is this intelligence stack with a clear governance protocol consisting of a trusted eval architecture, a searchable log for every agent, granular rollback to previous versions, and a human review queue. Humans rise a level to perform dashboard oversight, monitoring, exception handling, problem solving, and efficiency increases. It is like in Germany, where nobody works on the flat concrete floors, yet unemployment hasn't dropped because everyone is doing more problem solving, design thinking, and efficiency work. This govern and assure loop combined with the OODA loop creates a tight core engine that prevents the system from flying off the rails.

When agents talk to other agents, they need clear mechanisms: a purpose layer, a sensing layer, an interpretation layer, a decision layer, an orchestration layer, and a learning layer for feedback.

Imagine you're a retail company and a competitor suddenly announces same-day delivery. Sensing agents detect this and bring the information back to a decision layer to determine the response: should you offer same-day delivery, buy a startup that does it, or ignore it? Normally, your strategic and marketing officers would meet to decide, which in a manual operation could take months; now, you do it in hours and days.

Historically, there has been an impedance mismatch between a Fortune 500 company and a startup. A big company has so much to lose that they are paralyzed; one of twenty people can say no and kill an idea. Conversely, a startup only needs one of twenty investors to say yes to be off to the races.

Beyond purpose, sensing, interpretation, and deciding agents, there is an orchestration agent. If the decision agent suggests buying a startup, the orchestration agent sets up functions to find startups, analyze them for M&A, and alert the corporate development team and legal agents. Finally, a learning loop asks if previous acquisitions worked, powered by a core engine of recursive learning.

The organizational singularity occurs when you have recursive self-improvement at the workflow level. For example, in invoice processing, you could replace human checkpoints—like verifying if goods arrived—with an agent that constantly asks how to make the process better every loop.

When AI Agents Talk to Other Agents - Cross-firm Operation Architecture

We recognize that agents will do crazy things, so we've developed a framing based on Web3 smart contracts and old web architecture. Every agent gets a passport with metadata defining its constraints—such as policy-controlled APIs and data object metadata specifying what can be exposed. This liability framework ensures agents don't do illegal things that would make lawyers go bananas. With these constraints and oversight, other agents in the governor's loop watch them; the minute something goes off the rails, a human is notified, and the agent is stopped, rolled back, and checked.

This works because agents are relatively free. Much like needing a thousand physical qubits for one logical qubit in the quantum world, you can have many agents doing things with many agents overseeing them.

Could a two or three person team with Hermes or Open Claw disrupt your major lines of business? Moats include proprietary data, regulatory capture—which we see in healthcare—and an intelligence moat. If you learn faster than everyone else, nobody catches you; this is why Claude and ChatGPT's learning loops are further ahead than Manas or Grok.

Finally, being deeply committed to a purpose and not wavering develops a depth in the relationship with the end customer. Brand is critical because it sits with MTP and that emotional connection; a strong brand should use these new agents and capabilities to reinforce that. Existent EXO 3.0 is the best name I have.

The Middle 60% Problem & The Four Phases of Becoming AI-Native

C-suite executives become accountability holders, dashboard oversight evaluators, and validators rather than doers. They use their wisdom and experience to guide agents, watching their actions and deciding whether to approve or redirect them.

Middle management is where the biggest change happens because their current role is almost completely coordination—repackaging data from the coal face for the C-suite. That function drops about 90%. We need to lift those human beings up to handle exceptions and problem solving. Meanwhile, the bottom 20% do more enabled work, as agents do almost everything while they provide oversight.

On Moonshots, we've discussed the idea that firm sizes will reduce by 80%, meaning you can run an average company with 20% to 25% of the previous workforce. You could view this as 75% unemployment, but our view is that we'll see five times more companies created, making them more innovative. We've already seen a Cambrian explosion of startups and an interesting increase in entry-level hiring.

I think 60% of workforce reductions come from middle management, 20% from the bottom, and 20% from the top. Middle management is hit hardest because you cannot outperform an agent at gathering and aggregating sales reports. We believe active, aggressive apprenticeship programs are necessary. If a middle manager is displaced, they should partner with the CFO to learn alternatives, returning to a guild-like apprentice model.

This creates a new intelligence core for the organization across the C-suite, middle management, and coal face. We have deep expertise here because when we built the EXO model, we solved the immune system problem where antibodies attack anything disruptive in a big company. If you are a hundred-million-dollar trucking company, two guys with an AI-centric organization can lease trucks and compete the hell out of you. Based on our experience, you cannot change and transform the existing company.

This goes back to Buckminster Fuller, who said you can't fix an existing system; you have to build a new system at the edge and let that become the new gravity center. John Hagel and John Seely Brown identified that disruptive things happen at the edge. The poster child is Nestlé, which created Nespresso in 1976. For ten years they tried to run it as a line of business inside the mothership, but it didn't fit because of the different brand, supply chain, delivery, and customer proposition. Nespresso is now one of their highest performing lines of business, and every hotel room in the world has one.

Other classics include Steve Jobs starting the Mac, IBM creating their PC, or Lockheed with their Skunk Works. Apple would take a small team, put them at the edge, keep them secret, and tell them to disrupt a different industry. There is a reason Amazon Web Services wasn't done in the core service; it just doesn't fit. I tried this in one of my companies with a hundred-person organization and still could not get it, so I had to start it as a separate organization. Richard Branson took this to an extreme, spinning off another company every time he hit 150 people to break through the Dunbar number problem.

I've seen the innovation process in detail in probably 250 of the Fortune 500, and I've never seen any method work other than doing disruptive things at the edge pointing into adjacent spaces. We've been helping companies from Procter & Gamble to Siemens Energy, Black & Decker, and HP do this. If you do anything else, you will fail.

If you do this on the edge, the organization must report to the CEO, and the board of directors must provide the CEO full support. If you're disrupting your own organization without the board's support, you're screwed. You do not touch the existing organization. Right now, people are trying to inject AI into places and it's just not working. Instead, you create an AI native digital twin at the edge of your organization. Once you set up that separate entity, take three to five of your crazy young people and partner with a builder—not a consulting company—to get forward deployed engineers. Pick a standardized, cookie-cutter workflow from the legacy organization, like invoice processing, and rebuild it in this new entity. You don't move it; you copy it. Then, you take the steps in that process and use our methodology to task breakdown and score each task.

You replicate it in this new system by forking the data and running it there. This de-risks the process because if something goes horribly wrong, you aren't risking the mothership. You run this in parallel until you hit that recursive self-improvement loop. Once you see the improvement loops here are way faster, you know you're in. Even then, give it another few weeks to quality check against the original. Once you've confirmed everything, you slowly deprecate the old and move the next workflow over—maybe receipt confirmation, then demand forecasting. Little by little, you grow a full digital twin at the edge until your AI native digital twin is fully running.

Our current estimates are that once that digital twin is running properly, your performance improvement should be 100X or higher. If it's processing one invoice now, it should process 100 next; if something took 100 days, it should take one. In this new model, you have fewer human beings who focus more on oversight, exception handling, and problem solving. What excites me is that once you've built this AI native digital twin at the edge, you can create adjacent companies. Many of my companies have amazing teams I don't want to push further for the sake of their quality of life, but with an automatic digital twin, that team can start building other products and services.

Two sectors have gone through this full loop. One is contact centers. We went from human business processes outsourcing to chatbot assisted customer service, and now we have AI native customer service, as Klarna has done. When I set up a new website for the Organizational Singularity, the AI on Cloudflare told me exactly how to run the exception rules for domain forwarding.

The second domain where this is fully happening is marketing and content generation. It used to be agency heavy, then AI assisted, and now it's AI native. We've seen this to the point that we're working a 7-day week. It took three years of hell to write the first book and two and a half years of hell to write the second, mostly because we had to rewrite it after generative AI came out. But this third book took three months because every contributor could use an AI to add more data and their methodology, and then you're off to the races.

Rewrite Methodology: The Six Sequenced Steps

We call this methodology rewrite. If you have a workflow like invoice processing that you want to move over, you must first do a backcasting exercise. Backcasting is a methodology in future studies and forecasting where you define the vision. For example, if Elon wants to get to Mars in seven years, he asks where he has to be in five years to achieve that.

Step one is to take your company—such as the trucking or retail company I mentioned earlier—and paint a picture of what that company looks like fulfilling its MTP and architecture in an AI-native centric way. This is one of the easiest things to do in conversation with a large language model, so take your C-suite and perform that exercise.

Step two is to score your company using several metrics. First, determine the organizational drag: does getting something done require five or six decision loops and approvals, or is it like Nvidia, where you go straight to the founder for a yes or no? Second, determine if AI is a first-class citizen. If it is a tool injected by IT, your score is low; if you have a chief AI officer and are building AI-native capability, your score is much higher on a scale of one to ten.

Next, take your most prescriptive workflows and map and document them to ensure clear knowledge. A major problem is tacit knowledge. For instance, in video production, there are steps a producer takes that are not obvious from the outside or documented anywhere. If you lose that person, an AI cannot do those tasks right away.

Companies are currently shattering roles with agents, but 44% of Gen Z workers are sabotaging the AI by giving it bad information so it cannot take their jobs later.

We have developed a 10-week process to hack culture at scale and break the immune system, which we have implemented a hundred times for big companies. The next steps are to cut organizational drag, build a digital twin to migrate workflows one by one, and rewire systems so everything flows into that new structure.

Currently, companies have a cloud provider and networking, a set of ERP systems like Oracle financials or SAP, and an application layer. People try to layer AI on top of this horrible 50-year-old architecture that cannot be easily unwound. New architectures feature a connectivity and cloud provider, a data lake with all data accessible in one spot with proper approval levels for each object, a custom-built application layer for workflows created by AI, and then AI and agents on top.

SaaS providers are freaked out because their model is incompatible with this. They are trying to maintain their place by staying wired into the legacy systems of organizations, but building a proper stack provides full agency and control at a much cheaper cost. Anyone who has tried to implement an ERP system knows the hell involved.

We believe this full transition for all companies will take about 5 to 7 years.

The 2036 AI-Native Firm - 100x More Performant

To survive the majority of companies over a five to seven-year period, you must transition through what we call the turbulent transition. We have to carefully architect societally how we get through this two to eight year period.

In our opinion, you should be able to run a company with 10% to 25% of your current staff. A marketing company might drop to 10% human beings, while a physical company, like one building a data center, might stay at 25%. For example, we estimated that Fermi America should be able to run a power plant with about 80 people instead of 800.

This creates a ratio of one manager to 20-plus high impact individual contributors, as Jack Dorsey calls them, instead of the current one to three or five. The manager uses AI to keep everyone connected. Take Cognition Labs: their ARR grew 73 times when they went fully AI native. Every data point we have gathered over the last few months points exactly to this trajectory.

If you are Unilever and Procter & Gamble automates everything with a recursive improving digital twin, you will not outcompete them. Elon has talked about triple-digit GDP growth from companies delivering 100x compared to before. By doing things 100x cheaper, revenue scale and profit go through the roof. This demonetizes things and leads toward universal high income as the cost of everything drops, which brings us to UBI, UHI, and universal basic services.

AI-Native Firm: What Survives and What Doesn’t

What survives is the MTP encoded as a protocol in the company, the accountability shell—the legal entity, fiduciary holder, and liability container—and critical proprietary intelligence in that stack. Coordination protocols and curatorial judgment become killer; when execution is nearly free, judgment and taste become vital.

The traditional org chart completely fails. David Rose famously said the structure that made you successful in the 20th century will make you fail in the 21st. Static planning and the five-year plan die completely because we are in the middle of the singularity. While traditional org structures only change during major events like M&A transactions or launching new business lines, the new world requires a dynamic structure that adapts like an amoeba. The organization itself becomes a protocol.

Middle management as a coordination layer, quarterly reviews as a unit of decision-making, and annual planning are gone. Inertia mode—where customers don't switch because it is annoying—and wasting assets in the aging economy are gone.

If your company is under 50 people, you can brute force this because you are on a first-name basis with everyone. If you are over 50, do not risk the existing company by trying to break its immune system; instead, create a digital twin at the edge. We are currently doing this with about four companies and will likely do 10 at a time.

To get started, email kevin@openexo.com. Kevin Allen is our head of community and his AI will help navigate the process. Alternatively, visit organizationalsingularity.com and fill out a form. We will tell you to fix your organizational drag first, as we believe it is a 90-day process to get the first 10 or 20 workflows working in this new way.

My EXO community of 50,000 people in 150 countries is being retrained for this. I will be personally involved in the first couple of batches, just as I was in the first sprints.

Sheikh Mohammed wants to run 50% of the Emirati government on this. Almost all government processes are prescriptive; for example, renewing a driver's license is extremely well understood. Minister Al Olama told me, "Salim, come and get a golden visa. You're going to be my poster child." They are now processing golden and resident visas in 5 hours. You can create an organizational design where you pick a domain like healthcare or education and set up a structure that allows that inner loop to move.

We have a platform and a playbook. We are launching the book as an AI, specifically as a Claude skill. Just as Claude mentioned having connectors to QuickBooks and other tools, we are downloading the entire contents of the EXO framework as a Claude skill because we learn new things every two or three days.

If you have a process that takes ten steps, brute force it and rethink it so it takes three or less. Once you achieve that, you're ready to move into the digital twin. You can also start setting up the legal framework and getting board approval, as there is a lot of scaffolding required. You may have legacy legal issues; for example, in Germany, workers councils decide how many employees a bigger company is allowed to have, which is not great for flexibility, but there is much else you can do to navigate this.

One of our folks, Patrick Sandino, suggested figuring out a way to retrain people whose work might be at risk, building a transition plan for society to solve the social contract along the way.

After about three months of this, I started writing the first version of the book, using three instances of Gemini, ChatGPT, and Claude to take cracks at different things. I sent it to the community for feedback, gathered lessons learned, and talked to cutting-edge AI practitioners. We have a dedicated team spending a huge amount of time keeping up with breakthroughs and headlines so we can constantly tweak the methodology for rebuilding.

At organizationalsingularity.com, I think we are teaching boards and founders how to survive the coming disruptions. As William Gibson said, "The future's here, it's not evenly distributed." If you're a five-person startup, you're building an AI-native way anyway, as seen with Alex Finn, the open claw stuff, and Hermes.

The central thing to think about is that our past organizational structures were organized around hierarchy and human-centric workflows. Now they need to be AI-native agentic workflows, a totally different model architected around intelligence, not hierarchy. I hope we'll be able to track this and you can report on companies that have made this transition.

Mentioned

People

Ronald Coase Clay Christiansen Stanley McChrystal Jack Dorsey Eric Schmidt Buckminster Fuller John Hagel John Seely Brown Richard Branson David Rose Kevin Allen Sheikh Mohammed Al Olama Patrick Sandino William Gibson

Companies

Uber X Prize Block Dropbox Nestlé IBM Lockheed Procter & Gamble Apple Siemens Energy Black & Decker HP Klarna Cloudflare NVIDIA Oracle SAP Fermi America Cognition Labs Unilever