The Great Splintering

There is a word that keeps resurfacing as I think about what the next decade actually looks like, and it’s not “collapse” or “disruption.” It’s “splinter.”

Over the last century, we built monoliths. Power consolidated at the center and extracted value from the edges. Software monopolized to Meta and ServiceNow. $1 trillion in real estate assets rolled up to Blackstone. Zoning laws concentrated wealth and stamped the logic of industrial production onto the physical world.

But those big, centralized systems are now cracking at the seams. From the splinters, new shapes driven by trust, reputation, and proximity are forming. This isn’t just the same old cycle of unbundling and rebundling. The scale of the reset feels bigger than previous cycles, and the stakes feel higher. Because we aren't just unbundling software this time. We're unbundling the human experience itself: separating out what is computational from what is irreducibly human, and deciding where value really lies.

The Great Splintering will reach far and wide, but there are three specific areas where I predict its expression will be most visible:

• Software moats will weaken, and advantage will shift to trust and expertise.
• Work will become less about cognitive execution and more about human-centric judgment and relationships.
• The built environment will reconfigure around the new technological backdrop.

I. No More Software Tollbooths

For decades, enterprise software companies have operated a remarkably durable business model: build a platform once, extract rent forever. ServiceNow, Salesforce, and Workday aren’t just software companies; they’re tollbooths sitting between buyers and the workflows they depend on. The genius of the model was that switching costs were so high, and the cost of replication so prohibitive, that no one seriously threatened the tollbooth.

That calculus is changing.

Because the tollbooth didn't just collect rent: it had an army of agents directing traffic toward it. Trusted experts referred, implemented, and entrenched them inside enterprises. Accenture or Deloitte would be hired to solve a business problem, and the solution was almost always someone else's software. The dynamic remained: the intermediary owned the relationship, but the platform owned the economics.

As the cost of software has begun to trend precipitously toward zero, these same intermediaries has discovered something powerful: they don’t need to sell on behalf of the big platforms anymore. They can build their own. For a modest upfront cost, they can spin up a near-replica of an enterprise platform and deploy it directly to the client relationships they already own. Near-replica becomes full-featured replica once agents can reliably traverse the long tail of requirements and execution, a future that’s already nearly upon us: METR (opens in new tab) finds the length of tasks frontier agents can complete end-to-end with 50% reliability has doubled g approximately every 7 months.

Think about what this means structurally. Today, a consultant might resell a large SaaS product and capture a thin margin on GMV that flows through someone else's platform, with minimal equity exposure to the underlying value being created. But they’re the ones who deeply understand their client's specific needs and how to satisfy them. That expertise and relationship is now where value accrues. Tomorrow, that same consultant — armed with AI-generated software at a fraction of the old build cost — becomes the platform. They own the workflow. They own the data. They own the relationship economics.

There's a natural objection here: why wouldn't the enterprise customer just build everything in-house? If software is cheap, why do you need the intermediary at all? But this overestimates how many organizations actually want to be DIY. Most enterprises — like most people — don't want to build and maintain every tool they use. They want someone trusted to handle it for them. This is the deep underlying signal behind the rise of forward-deployed engineers and the persistence of consulting as a category even as AI eats away at consulting's traditional deliverables. The role of the consultant — the trusted advisor who shows up, understands your problem, and manages the solution — doesn't go away. It becomes more prized and more specialized because trust and expertise are the scarce inputs in a world where the software itself is abundant.

Anthropic senses this. That’s why they’re spinning up a dedicated consultancy (opens in new tab) that can cultivate and own the relationships needed to convince businesses to use their platform. And it’s a pattern that extends beyond consulting. Paraform (opens in new tab) is executing on this strategy. Instead of just recruiting on behalf of a company like Google, where only a subset of the network of job applicants is a fit, they recruit on behalf of dozens of fast-moving companies. A recruiter with a deep network can now activate that entire network — making placements at Anduril, at a Series B startup, at a public company — rather than leaving most of their relationships dormant. It’s a play to maximize network and trust in an experience previously defined by bottlenecks.

Networks are splintering away from centralized platforms and reconstituting around trust. The great aggregators assumed that software itself was the moat. It wasn't. The network was. And now the network can have its own software.

II. The Splintering of the Job Itself

Knowledge workers currently represent over half of the workforce (opens in new tab). Most of that work is, at its core, information processing. Researching, analyzing, synthesizing, drafting, deciding. These are exactly the tasks AI will do better, faster, and cheaper than people.

If the optimistic AGI thesis plays out — and I believe it will — what remains of this work in a generation? Two categories of people: the true experts who sit at the frontier of human knowledge and push it forward, and the network-holders — the connectors, the trusted intermediaries, the people who own relationships rather than just contracts. These are the roles that AI augments rather than replaces, because their value is fundamentally social and reputational, not computational.

Here is the question I keep coming back to: what parts of a job are we actually trying to protect?

Take medicine. AI will soon be superior to most physicians at diagnosis. Pattern recognition across imaging, labs, and patient history is a computational problem, and computers are very good at those. If we are honest about optimizing for outcomes, AI should make the call.

And yet.

There is something irreducibly human about receiving a diagnosis. The moment a doctor sits across from a patient and says here is what is happening in your body, and here is what we are going to do about it is not an information transfer. It’s the moment your life changes. It is a relationship. It requires trust, presence, and accountability. The patient needs to feel that a person, not an algorithm, is responsible for their care.

So the job of the doctor doesn't disappear. It splinters. The diagnostic function migrates to AI. The relational function — delivering, explaining, advocating, guiding — remains human. Regulatory frameworks will likely formalize this split because society will demand it.

The same logic applies across knowledge work. A teacher may not walk a student through a textbook because AI tutors will do that better, with infinite patience and perfect personalization. But a teacher as a mentor, as a trusted adult in a young person's life, as someone who sees a student's potential before the student does — that role doesn't just survive, it becomes more important.

An attorney may not draft the contract. But someone still needs to look a client in the eye before they sign their company over.

The great irony of the AI era may be this: by stripping away the computational and analytical functions of professional work, we are left with the parts that were always the most human, and the most valuable. The jobs that survive won't be diminished versions of what came before. They will be something closer to their essence.

Mercor (opens in new tab) is an early proof point of what this looks like in the wild: instead of hiring a lawyer or a doctor, you increasingly buy slices of their judgment on demand, and the “job” starts to unbundle into discrete, priced tasks. Done well, that shifts human work up the stack, as experts use these systems to offload routine analysis and concentrate on accountability, taste, and edge-case judgment.

III. Campuses > Cities

The built environment has always been a lagging mirror of the economic forces shaping how we live and work.

The industrial revolution didn't just move production into factories. It reorganized human geography around them. Cities swelled. Neighborhoods clustered by trade. Then the automobile and the post-war boom gave us suburbs: the original hub-and-spoke city, where workers radiated outward from a downtown core each morning and retreated each evening. The physical world encoded the economic logic of its era in concrete and steel.

We are at another one of those inflection points.

If AI splinters knowledge work into AI-powered output and human connection, cities organized around producing the bundled knowledge work of the past century will also change. Again, the question becomes what remains?

COVID gave us a clue. The pandemic provided the first moment of dislocation between knowledge work and the cities that fostered it. Suddenly, a third of the workforce could go anywhere. And, predictably, they did. But it's what came next that gives us a clearer sign of where things are headed. Most of those people moved back (opens in new tab). New York City's real estate market rebounded (opens in new tab), and then some. It's a phenomenon that's bigger than return-to-office policies: retirees flocking to major cities (opens in new tab), an explosion in spending on concerts – moments of pure co-presence – that could have fueled an entire country's econom (opens in new tab)y.

People came back to cities because it's the closest to what they're looking for. But their post-pandemic spending and behaviors reveal preferences for the village-like qualities of cities, not the commute-to-a-tower arrangement that cities were optimized for. As AI further decouples output from connection, I predict those preferences will increasingly define the built environment.

When machines do the work, we get to reimagine what a life of abundance actually looks like and why it is we gather. We have the opportunity to design for leisure and fulfillment rather than the demands of a commute. Instead of hub-and-spoke cities organized around office towers and daily commutes, I predict the rise of campus-like environments: dense, walkable, mixed-use communities that bundle together dining, fitness, entertainment, healthcare, micromobility, and — most importantly — the social infrastructure of friends and family. Not gated communities, but genuine villages. (Somewhere, Jane Jacobs is nodding and smiling.) Environments designed not around the demands of a job, but around the texture of a life.

We’re already seeing our biggest cities begin to adapt. “Third spaces” (opens in new tab) like bathhouses, membership clubs, and coworking spaces that lend themselves to a more distributed, communal experience of work are booming. Tinder has lost 600,000 users (opens in new tab) to a wave of new startups that get daters offline. Cities from Paris to New York are aggressively rezoning for mixed-use development, pedestrian zones, and outdoor gathering spaces. The 15-minute city concept — where everything you need is within a short walk or bike ride — is gaining traction not as urban planning theory but as actual policy. The city was a transit mechanism for labor. The campus is designed for living.

This shift will be the slowest to materialize, and may take fifty years to fully play out. But the direction is clear, and it is already beginning.

What is precious doesn’t scale

History has a reliable pattern: when production costs collapse, power reconcentrates around whoever seizes the new chokepoints. The railroad barons didn't democratize transportation — they monopolized it. The internet flattened information-sharing, and then created Google and Meta. Cloud gave us AWS. So why should AI be any different? Why won't we just end up with a "Meta of trust networks" or a "Blackstone of micro-campuses"?

A platform that claims to hold your trusted relationships holds none of them; the moment a network becomes a product, it stops being a network. The chokepoints of prior eras were physical or infrastructural — pipes, rails, data centers — and physical infrastructure rewards consolidation. What we're describing here is an economy increasingly organized around things that actively degrade when centralized.

In the era of monoliths, scale was a competitive advantage. Now what scales becomes a commodity. Our next economy will organize itself around the scarce goods that don’t scale: trust, presence, accountability, and human connection. The Great Splintering isn't a story about things falling apart. It's a story about things rebuilt around what actually matters. The biggest winners will earn their communities, not hold them captive.