In 1997, Arie de Geus published a book about why some companies live for centuries while most die in their forties. He was Shell’s head of strategic planning. The board had asked him a simple question: how do we last? De Geus went looking for answers and came back with a list of companies that had been operating for 200, 400, 700 years. What he found wasn’t a strategy. It was a structure.
In March 2026, Jack Dorsey and Roelof Botha published a paper called “From Hierarchy to Intelligence.” It describes how Block, a $40 billion company, is replacing the org chart with three roles and an AI-mediated information system. The paper is being read as a thesis about AI. It’s also, whether the authors realize it or not, a working implementation of de Geus’s argument from twenty-nine years ago.
Less discussed is what this does to the people working inside the new structure. When intelligence moves into the system and humans move to the edge, the capabilities those humans need aren’t the ones most companies are building.
What de Geus actually found
De Geus studied companies that had outlived their peers by orders of magnitude. Stora Enso, the Swedish forestry firm, had been operating since 1288. Sumitomo had been mining copper in Japan since 1590. Most of the 1955 Fortune 500 were gone within four decades. The average corporate lifespan had been compressing since the industrial revolution. De Geus wanted to know what the survivors had in common.
He identified four traits.
- Sensitivity to the environment. The long-lived companies paid attention to what was changing around them and adapted before they had to. Learning was a survival mechanism. Most companies still treat it as a budget line.
- Cohesion and identity. Their people knew who they were as a company. Identity was independent of any specific product, market, or leader. When a business unit died, the company did not.
- Tolerance and decentralization. They tolerated experiments at the edges. They let many branches grow rather than optimizing one. Power sat with the people closest to the work.
- Conservative financing. They held cash. They retained optionality and didn’t bet the company on growth they couldn’t absorb.
De Geus called these companies “living.” The point wasn’t metaphorical. He was describing biological selection. The companies that survived behaved like organisms in a changing environment, rather than machines optimized for one.
Stora Enso has been responding to environmental pressure for 738 years. It started as a copper mine. Today it makes paper, packaging, biomaterials, and wood products. The product changed five times but the company didn’t.
Why the four traits worked together
The four traits look like a list, but they’re actually a system. Take any one away and the system collapses.
Adaptations rests on the four traits
Sensitivity to the environment is the input. The company sees what’s changing. Tolerance for experiments is what the company does with that input. Identity is what holds the company steady while the experiments run. Conservative financing buys the time to let the experiments mature without panic.
Take any one trait away and the system collapses. A company that senses change without tolerating experiments freezes. A company that tolerates experiments without identity drifts. A company that has identity without conservative financing dies in the next downturn before it can adapt. A company with all the financing in the world but no sensitivity to environment will spend it on the wrong thing.
That’s what de Geus meant by “living.” A living organism doesn’t have one survival mechanism. It has a coupled system of them. The long-lived companies weren’t lucky. They had the full system and ran it long enough for compound adaptation to work.
The mechanism was slow. Stora Enso didn’t pivot from copper to forestry in a quarter. It happened over decades. The pressure that produced the four traits was generational.
Until 2023, that is.
AI compressed the timeline
AI isn’t a new business problem. It’s the same selection pressure de Geus described, running at ten times the speed.
Companies that already practiced the four traits get a head start. Companies that didn’t are now being churned in quarters instead of decades. When you map the four traits onto the AI environment, three things happen. Some scale up. Some break. Some new requirements appear that de Geus couldn’t have anticipated.
Where they scale up
Sensitivity to the environment becomes the company’s biggest advantage. The companies that already practiced collective learning, what de Geus called the “birds that flock” effect, get amplified by AI. Information moves faster, signals propagate faster, adaptation cycles shorten. Companies that already distributed power push decisions further out. Block is the structural proof. The Dorsey and Botha paper describes a company where, in their words, “the intelligence lives in the system. The people are on the edge.” That’s de Geus’s argument translated into engineering.
Tolerance and decentralization matter more when the cycle compresses. Companies running many small AI experiments at the edges learn faster than companies centralizing AI as an IT function. The latter are pruning too early.
Where they break
Conservative financing was a survival trait when adaptation cycles were generational. When the cycles compress to quarters, hoarding cash can mean missing the window. The instinct that protected living companies for centuries can now be a liability.
Cohesion and identity get harder. In a 700-year company, identity came from the institution’s continuity. AI is dissolving that continuity faster than most companies can rebuild it. People can no longer rely on the org’s longevity to anchor who they are. Identity has to become personal and portable, or it doesn’t survive the next restructure.
What’s new
Two requirements appear that de Geus couldn’t have anticipated.
The first is a “company world model.” Block has one. It’s the transaction data of millions of businesses, structured so AI can read it and act on it. Most companies don’t have this. They have files in folders that no AI can interpret as a coherent picture of how the business actually runs. Without that substrate, AI can’t replace the connective tissue that hierarchy used to provide.
The second is that workflows have to be agent-native. The processes a company runs need to be designed for humans and AI to operate inside them together. Failure signals in the system should generate development priorities directly, without a committee in the middle. This is a structural property of the org. No tool selection fixes it.
What this does to the people on the edge
When intelligence moves into the system and humans move to the edge, the work the humans do changes. AI handles the scripted situations. The humans handle what’s left, and what’s left is the hard part.
- Judgment in unscripted situations. AI gives confident, fast, often wrong answers when the problem isn’t framed correctly. The human’s job is to frame the problem and catch the wrong answer.
- Trust-building without hierarchical signals. When the org chart no longer signals authority, you have to build trust the slow way. Through reputation. Through showing up. Through being someone other people want to work with.
- Context fluency. Knowing what to give AI and what to keep human. This is a judgment call that compounds. People who get it right pull ahead. People who don’t get exposed.
- Comfort with iteration and imperfect output. AI produces drafts. People who treat the draft as an answer will be wrong, often. People who treat it as a starting point will move faster than anyone working without AI at all.
- Prompt craft as a thinking skill. The act of writing the prompt is the act of clarifying the question. People who can’t write a clear prompt can’t think a clear thought. Prompt craft is a thinking skill that happens to use a tool.
None of these capabilities can be built through information transfer. None of them are course content. They develop the way de Geus said learning developed in living companies: through psychological safety to not-know, through participation in real decisions, through collective learning at the edges, and through bottom-up practice that no one can mandate from the top.
De Geus said it in 1997 about long-lived companies. It’s more true now than it was then.
Why most L&D is solving the wrong problem
Most L&D programs were designed for the org chart that’s being dismantled. Role-specific tracks. Manager pipelines. Leadership levels. They were built for hierarchy as an information-routing mechanism. AI just removed the need for this hierarchy.
What gets built instead of judgment is a course library that fits the old structure. Onboarding modules tied to job families. Manager certifications tied to reporting depth. Leadership tracks tied to title progression.
There’s a parallel trap: AI training programs that focus on tool fluency are repeating the same mistake at higher cost. They treat AI as content to be transferred. Module 1: what is an LLM. Module 2: how to write a prompt. Module 3: AI ethics and safety. These programs are useful for a quarter but then they expire. People who finish them know how to use AI tools. They don’t know how to think with AI in unscripted situations.
The answer is to design for the conditions de Geus described. Distributed practice. Peer learning. Judgment built through reflection. Capability built through cycles of real work with real feedback.
That’s a different design problem than most L&D teams are set up to solve.
Seek-Sense-Share as the development model
If the people conditions de Geus identified are the substrate, the practice loop sits on top of it.
Our work at Curious Lion is built around a framework called Seek-Sense-Share.
We developed it about 20 months ago, working with a high-growth e-commerce company in San Francisco. They had manager development figured out. What they needed was for their individual contributors to drive more impact on their own. We didn’t go in with this framework. We went in to build a program that would make ICs better consultants. Teach them to research, gather perspectives, form a point of view, and pitch it inside their organization.
The pattern and the framework emerged from the work. Once it had a shape, the lineage was easy to trace back through Peter Senge, Chris Argyris, Donald Schön, and David Bohm. Harold Jarche’s work on PKM inspired the language. But the framework itself came from the room, with real ICs trying to move real decisions, and the loop is what kept working.
The loop is short.
- Seek. Curiosity at the edges. Going looking before AI gives you the easy answer. The ability to ask questions in a way that invites stories from people. Stories carry details, and details are what meaning is made of. A summary from an AI doesn’t contain those details. A real conversation does.
- Sense. Making meaning of what you found. A lot of this work can now be done with AI. Synthesizing information, organizing it, sifting it through different perspectives. What doesn’t move to AI is the so-what. Why does this matter? What’s the point of view that this set of details actually supports? That judgment is the human contribution, and it’s the part that gets sharper or weaker depending on how seriously people practice it.
- Share. Adapting your communication so your point of view drives action. The work is to take the so-what you arrived at in Sense and shape it for the specific person whose decision you’re trying to move. Different audiences need different versions of the same point. People who can do this well move things inside their organizations. People who can’t are stuck, no matter how good their thinking was.
The loop is intentionally simple. The work is in running it often, in real situations, with real feedback. That’s how de Geus said capability developed in living companies. It’s also how it develops in humans on the edge of an AI-mediated organization. The mechanism didn’t change, but the pace sure did.
Close
The 700-year companies practiced this slowly, by accident, over centuries. Selection pressure was generational. Most companies died because the pressure was real and adaptation was slow. The ones that survived had structures that let them adapt without noticing they were adapting.
AI removes the slow part.
The traits that used to be a competitive edge for the long-lived become table stakes for anyone who wants to survive the next cycle. The companies that already practice them are going to look like geniuses in five years. The ones that didn’t are going to look like the companies de Geus left out of his book.
