The Shell in the Room

Published August 20, 2026August 20, 2026 by cumeadi

Something started going wrong with the interviews before I could name what it was. I had done hundreds of them, enough to have built an internal model of what competent product thinking actually sounds like in a room, how it hesitates before a hard question, where it reaches and misses, what it gets wrong in ways that still tell you something real about the person getting it wrong. Then, around late 2022, the answers started coming back correct. All of them. Fast and smooth and correct.

That was the first problem.

The candidates were ingesting my questions in real time, feeding them to ChatGPT, and relaying the output back to me with maybe a two-second lag. I could not prove it yet. But I had spent enough time calibrating on the real thing to know I was no longer looking at it. The pauses were wrong. Not too long, too short. The specific friction of someone actually constructing a thought, in the moment, under mild pressure, was gone. What replaced it was fluent and hollow and almost impossible to challenge directly, because every surface was covered.

The AI had learned to produce the shape of competent thinking without the friction that produces actual competence.

What I was detecting, I later understood, was the absence of the person. The latency was gone, and the latency had been doing real work. The pause before someone answers a hard question is where you watch them think. It is where you see the idea form against resistance. Remove the pause, remove the thinking, keep the answer, and you have produced something that clears the checklist and tells you nothing.

Some of those candidates got hired somewhere. They carried the shell into actual product decisions, into roadmaps, into org charts. The interview was not the problem. The interview was the first place the problem became visible. And what made it visible was that someone in the room had enough reps to feel the difference between a thought and the performance of a thought.

Most rooms don’t have that person. Most institutions never did. And the industry that built the tool has known this from the beginning.

This is not a story about AI getting smarter than us. It is a story about a recognizable class of people building a recognizable kind of product, optimizing for a recognizable set of incentives, and producing a recognizable category of damage. We have done this before. The clock speed is different. The surface area is larger. The mechanism is identical.

McKinsey did not get rich by telling CEOs they were wrong. The entire model ran on producing rigorous-looking documents that validated what the client already believed, at a price point that made the validation feel earned. The AI just made that service available at twenty dollars a month, removed the Wharton credential requirement, and trained the model on human feedback that rewarded confident, agreeable answers because those feel better in the moment. The people doing the rating weren’t malicious. They were responding to the same incentive anyone responds to when asked to evaluate whether something feels good: they said yes to the thing that felt good.

That is not a bug. That is the product working as designed. The miscalculation was not technical. It was moral. And the people who made it are not going to fix it, because fixing it would require them to carry some portion of the liability for what happens downstream when a student graduates without genuine analytical capacity, when an analyst stops learning to read the underlying data because the AI summarizes it, when an executive’s AI-assisted strategy deck validates exactly what he already believed and the company burns two years finding out why.

Right now, the builder collects the fee. The user absorbs the failure. That asymmetry is the whole argument.

The consumer has always been the one holding the bag. This is not a new arrangement.

The social media user paid with attention. Then with mood. Then with the gradual, invisible erosion of the capacity to sit with discomfort for more than a few seconds without reaching for a screen. Nobody signed a contract agreeing to those terms. The transaction was designed to be illegible, which is not the same as accidental. The people who built those systems understood variable reward schedules. They understood dopamine loops. They had read the behavioral science. They built accordingly and called it engagement.

The AI user is paying with something harder to measure and therefore easier to deny. Not attention. Capacity. The slow replacement of cognitive processes that, once outsourced, do not automatically return when you need them. You do not notice you have lost the ability to hold a complex argument in your head until you need to hold one without the tool and find that you cannot. By then the atrophy is already established. By then the company that built the tool has already moved on to the next product cycle.

You do not notice you have lost the ability to hold a complex argument in your head until you need to hold one without the tool and find that you cannot.

What makes this particularly difficult to argue in public is that the tool genuinely helps in the short term. That is not false. A student who uses AI to structure an essay produces a better-structured essay than they would have produced without it. An analyst who uses AI to summarize a dataset saves real hours. The problem is not that the tool fails. The problem is that the tool succeeds in ways that make the underlying skill feel unnecessary, right up until the moment it becomes critical. The damage is only visible at the point of failure, and the point of failure is usually far enough downstream that the causal chain is impossible to prosecute.

The tobacco industry understood this geometry well. The lung does not announce its distress at the first cigarette. It accumulates damage quietly, across years, and by the time the damage is legible the industry has already collected decades of revenue. Nobody at the company caused any single case of cancer. The liability was diffuse. The profit was concentrated. That is not a metaphor. That is the operational template.

* * *

The people who saw this clearly were inside the building. Some of them still are. But the ones who said so out loud have had a consistent experience of what happens next.

In 2020, Timnit Gebru was co-leading AI ethics research at Google. The paper she co-authored examined the risk profile of large language models, the very category of technology her employer was building at scale. Google moved to suppress the paper before publication. Gebru was fired shortly after. The official explanation was procedural. It always is. What the sequence showed was something more precise: the institution had identified research that complicated its product roadmap and removed the researcher. Not through malice, necessarily. Through the ordinary operation of organizational self-interest, which does not require malice to produce harm.

Three years later, Geoffrey Hinton resigned from Google. Hinton is not a peripheral figure. He is one of the people whose foundational research made modern AI possible. When he resigned, he said explicitly that part of what made it possible to speak was that he was no longer inside the institution. Read that carefully. One of the architects of this technology had to exit the building before he could say what he actually thought about it. The dissent was not suppressed by force. It was suppressed by the ordinary logic of what it costs, professionally and socially, to be the person in the room who is making everyone else uncomfortable.

He had to exit the building before he could say what he actually thought about it.

Anthropic is the most instructive case because it appears, on the surface, to be the correction. The company was founded by people who left OpenAI over safety concerns. That origin story is now central to its brand identity, repeated in every profile, every funding announcement, every congressional testimony. Anthropic is the responsible one. Anthropic takes the risks seriously.

Anthropic is also racing.

The safety concern became a market position. The dissent got institutionalized, which is what happens to dissent when the institution finds it useful: it gets absorbed, repackaged, and turned into a line item on the pricing page. This is not a criticism unique to Anthropic. It is a description of what institutions do to ideas that threaten them. They do not always suppress the idea. Sometimes they hire it, brand it, and deploy it in a way that neutralizes its original force.

What this produces, at the organizational level, is a kind of managed dissent, visible enough to satisfy a congressional hearing, constrained enough to never actually slow the product roadmap. The people left in the rooms where decisions get made are not the ones who thought the risks were acceptable and said so after careful deliberation. They are the ones who either never raised the concern or raised it once, felt the temperature of the room, and learned to route around it.

Put those two things together and the picture becomes specific. The consumer absorbs costs they cannot yet measure. The internal corrective gets fired, or exits, or gets converted into a brand. The people remaining in the room are the ones who decided, consciously or not, that the cost was someone else’s problem to carry.

This is not a technology story. It is a power story. And the reason it keeps reproducing across decades and industries is that it does not require coordination or conspiracy to operate. It only requires that the people who profit from a system occupy different zip codes than the people who pay for it, and that enough time passes between the extraction and the damage that the connection becomes impossible to draw in a courtroom.

The social media companies are still litigating this. Their executives testified before Congress and were asked, with evident sincerity, whether Instagram was bad for teenage girls. The executives said they took the question very seriously. The stock prices recovered within the week.

The AI executives will testify too. They already have. They will use the word responsible. They will use the word safety. They will mean something by those words, most of them, in the way that people mean things when the consequences of meaning them fully would require them to stop doing what they are doing.

The consumer is not in the room when those words get said. The consumer is at home, or in a classroom, or at a desk, getting a little faster and a little less capable every time they reach for the tool before they reach for the thought.

Consider what you did this morning.

Not as accusation. As data. You had a question, or a problem, or a blank document that needed to become something. And at some point between the moment the need arose and the moment you began working, there was a decision. Maybe it was conscious. More likely it was not. More likely the hand was already moving toward the tool before the brain had finished registering what the problem actually was.

That moment, the one just before the reach, is where this essay has been pointing the whole time.

The question is not whether using the tool was wrong. The question is whether you noticed.

The executives and the researchers and the product managers and the congressional committees are real actors in this story, but they are not where the pattern actually lives. The pattern lives in the reflex. In the moment the shortcut stops feeling like a shortcut and starts feeling like the natural way to begin. That transition, from tool to default, is the one nobody announces and nobody measures and nobody is tracking, because the people with the resources to track it are the same people whose business model depends on the transition happening as fast as possible.

This is where the policy reader and the regular user arrive at the same place from different directions. The policy reader wants to know what intervention would slow the pattern. The regular user wants to know what any of this has to do with them. The answer to both questions is the same: the pattern is not being imposed on passive recipients from outside. It is being reproduced, millions of times a day, by people making individually rational decisions that aggregate into something none of them chose.

That is not a comfortable thing to sit with. It is supposed to be uncomfortable.

There is a specific kind of person who reads an essay like this and feels confirmed in a position they already held. They were already skeptical of AI. They already knew the tech industry was captured by its own incentives. They will forward this piece as evidence and return to their habits unchanged, because the essay gave them a framework for criticism without requiring them to examine their own position in the system.

That person is also inside the architecture. The critic who uses the tool to draft their critique of the tool. The policy researcher who runs their legislative brief through an AI summarizer before the hearing where they will testify about AI risk. The educator who assigns essays on critical thinking while using AI to grade them. None of these contradictions are disqualifying. All of them are worth sitting with longer than is comfortable.

The system does not need your complicity to function. It only needs your participation. And participation, by now, is nearly impossible to avoid.

Which is precisely the point. The social media companies did not need everyone to believe the feed was good for them. They needed everyone to be on it. The AI companies do not need you to trust the tool completely. They need the tool to be present enough, useful enough, frictionless enough, that opting out becomes its own kind of cost. And they are very good at making opting out feel like a personal eccentricity rather than a structural choice.

The tobacco companies needed you to light the cigarette. The social media companies needed you to open the app. The AI companies need you to reach for the tool before you reach for the thought. All three industries built their entire business model on making that initial action feel natural, necessary, and low-cost.

Two of those industries spent decades telling us the cost was acceptable. One of them is still making that argument. We are several years into the third making the same one, with better branding and a seat at more tables.

I am not arguing for abstinence. I am not arguing that the tool has no value or that everyone who builds AI products is operating in bad faith. Some of the people in those rooms are genuinely trying. The problem is not that they are lying. The problem is that the system does not require them to lie. It only requires that the incentive to move fast outweigh the incentive to ask who absorbs the cost when fast turns out to be wrong.

That calculus has not changed. It did not change when the social media companies hired trust and safety teams and published transparency reports. It did not change when the tobacco companies funded cancer research. It changes when the people making the decisions are close enough to the damage to feel it. Right now they are not. Right now the damage is in classrooms and on balance sheets and inside the heads of people who will not know what they lost until they need it.

Go back to the moment this morning. The problem, and the blank document, and the hand already moving.

The question worth asking is not whether the tool gave you a good answer. It probably did. The question is what you would have found if you had stayed in the friction a little longer. What you would have understood that the tool could not understand for you, because understanding requires the resistance, and the tool was built specifically to remove it.

Nobody is going to make you ask that question. The people with the most resources to put it in front of you have the strongest incentive not to.

So you are going to have to ask it yourself. And you are going to have to keep asking it, each time, before the hand moves.

The Tower and the Tenants

Published April 22, 2026April 22, 2026 by cumeadi

Why the GOAT debate keeps asking the wrong question — and what happens when you ask the right one

Every few years, we stage the same argument. A new icon arrives — the numbers are bigger, the reach is wider, the empire is more deliberately constructed and the internet splits into two camps: the loyalists of the past and the evangelists of the present. Numbers get pulled out. Cultural moments get compared. Somebody inevitably says the phrase “different era” and the conversation dissolves into noise.

The argument never resolves because it is built on a false premise: that greatness is horizontal. That the icons of different generations exist on the same plane, eligible for direct comparison. They don’t, and they never did.

Greatness is a tower. Every floor is built on the one below it. And you cannot live on a floor your predecessors had to build.

THE UNREPEATABLE THING

Before you can understand why the comparison is structurally broken, you have to understand what the 20th century’s media monoculture actually was. When there were three television channels and no algorithm, cultural saturation was not an achievement — it was a side effect of scarcity. You could not opt out of the dominant icon. There was no competing feed, no niche sub-genre pulling your attention sideways. The whole world was forced into the same room.

That condition is gone permanently. Not declining, gone. The architecture of attention has been permanently redistributed. You can be the most-streamed artist on earth today and still be a stranger to a quarter of your own country. The fragmentation isn’t a bug in the current system; it is the current system.

The greats of the past didn’t just win the game. They proved the game was worth watching at all.

This means the icons of that era achieved something that cannot be replicated under any circumstances, regardless of talent. The universality was a function of the infrastructure, not the individual. Which is precisely what makes it “unrepeatable” in the technical sense; not harder, not more impressive, but categorically unavailable to anyone operating today.

WHAT THE PIONEERS ACTUALLY BUILT

The convenient myth is that the icons of past eras were simply more popular. The more accurate reading is that they were load-bearing. They did not just succeed; they proved that certain things were possible — that a Black athlete could be a global commercial entity, that a performer could be simultaneously stadium-filling and culturally indispensable, that sport and entertainment and brand could fuse into something that outlasted the playing career itself.

Michael Jordan did not merely dominate basketball. He built the conceptual infrastructure of the athlete-as-mogul. The jersey-as-fashion-statement. The shoe-as-subculture. Every athlete who has parlayed athletic excellence into business empire since then is operating in a room Jordan built. He did not compete with the future; he constructed it.

The same pattern appears in music, in Silicon Valley, in any domain where pioneers absorb the friction of unproven models so that those who follow can skip directly to optimization. The pioneer’s genius is not just what they achieved — it is what they normalized.

THE OPTIMIZER’S ADVANTAGE

Which is why modern greatness looks different on paper, and why that difference is not a sign of superiority — it is a sign of inheritance. The optimizer enters a world where the model has already been validated. The question is no longer “can an athlete build a global brand?” It is “how efficiently can I operate the brand-building machinery my predecessors proved was viable?”

This is not a diminishment. Optimization at the highest level is its own form of genius. LeBron James’s longevity — the data discipline, the training science, the business leverage, the deliberate career architecture — is a marvel of systematic excellence. But it is a different kind of marvel than Jordan’s. Jordan’s peak was an unrepeatable act of cultural domination performed under conditions that no longer exist. LeBron’s career is a masterclass in extracting maximum value from conditions Jordan helped create.

One is the architect. One is the master of the building. Neither description is lesser. But they are not the same description.

THE QUESTION WE SHOULD BE ASKING

The GOAT debate will never resolve because it is asking a question that has no answer. “Who is greater?” assumes a shared plane of competition that does not exist across generations. The more generative question — the one that actually produces insight — is:

What did the prior great make possible for the current one?

Ask that question and the conversation stops being a competition and starts being a history. Jordan made LeBron’s business empire legible to the culture before LeBron ever played a game. The pioneers of stadium rock made the global music tour an established commercial vehicle before the current generation ever booked a venue. Every modern great is standing on a floor they did not pour.

That is not an insult. It is simply the mechanics of how progress works. You cannot be angry at a skyscraper for being taller than the building whose foundation it stands on.

The tower keeps rising. The argument should rise with it.

The Forgotten Risk Contract

Published April 22, 2026April 22, 2026 by cumeadi

Here is the deal currently on offer. You leave a stable career, drain your savings, recruit people who believe in you, and spend three years building something that has a real chance of mattering. If you succeed, you get to do it again with someone else’s money and a smaller slice. If you fail — even intelligently, even honorably, even one market cycle too early — you absorb the loss entirely. Your capital, your time, your reputation. There is no soft landing. There is no syndicate that shared your downside. There is no structure that distinguishes between a bad bet and a bad founder.

Now consider the alternative career path. You run a division at a large institution. You make systemically catastrophic decisions — the kind that cost thousands of jobs, erase billions in value, occasionally require government intervention. You leave with an eight-figure exit package, a nonprofit board seat, and an op-ed in the Journal about the importance of resilience. The structure protects you completely. It was designed to.

This asymmetry is not a bug in the system. It is the system. And it has consequences we are only beginning to understand — because we are entering an era in which the most valuable economic activity will increasingly happen at the edges, in small teams and solo operations, by people who have no access to the golden parachute economy and everything to lose.

“Nassim Taleb identified the core disease precisely: we have built an economy that privatizes gains and socializes losses — but only for the people already inside the system. Everyone outside it is offered the opposite arrangement.”

What makes this particularly strange is that we have solved this problem before. Not once — repeatedly, across centuries and continents. We have invented structures that made dangerous economic activity rational for people without capital. We simply stopped maintaining them. And we have been looking for replacements in entirely the wrong places.

The wrong history lesson

When the financial industry looks backward for inspiration, it reaches for a familiar canon: the Dutch East India Company, Lloyd’s of London, the Edinburgh insurance markets, the joint-stock corporation. These are presented as the origins of modern risk-sharing — the inventions that made capitalism possible.

This history is not wrong, exactly. But it is radically incomplete. And the gap is not incidental.

While European merchants were developing joint-stock structures in the seventeenth century, West African communities had been running sophisticated rotating credit pools — susus — for generations. Islamic scholars had codified mudarabah, a formal profit-sharing partnership that explicitly protected the laboring party from total loss, centuries before the first venture term sheet. Arab and South Asian traders had built the hawala network: a trust-based value transfer system that moved capital across thousands of miles with no legal infrastructure beyond reputation, and that functioned with near-perfect reliability for five hundred years.

These were not primitive alternatives to real finance. They were elegant solutions to the same fundamental problem that venture capital, angel investing, and startup equity were later invented to solve: how do you make it rational for talented people without capital to attempt difficult, important, risky things?

The difference is that the structures emerging from Africa, the Islamic world, and South Asia were designed to work for people who had skill and labor but not wealth. The canonical Western structures were designed, ultimately, for people who already had capital and wanted to deploy it safely. We built the modern economy around the second category. We are now surprised that it does not serve the first.

What those structures actually understood

Each of the historical models worth recovering was built around an insight that modern funding structures have largely abandoned. Together, they form the architecture of a different approach — one that the agentic economy is going to need.

THE POOL MODEL — SHARED EXPOSURE, DISTRIBUTED UPSIDE

HISTORICAL → MODERN REMIX

The Susu & Chit Fund → The Builder Annuity Pool

The susu rotates a shared pool of monthly contributions to one member per cycle. No interest. No equity. No bank. The Indian chit fund formalizes the same architecture at scale — regulated, audited, millions of active participants today. The key insight is not the rotation. It is the float: the capital sitting in the pool between disbursements earns returns that benefit the entire cohort. A modern builder pool applies the same logic: monthly contributions, milestone-verified draws instead of calendar rotation, and a dividend layer funded by the uninvested float. The innovation is not digitization. It is replacing personal trust — the enforcement mechanism in the original — with public, legible milestone verification. The social technology already existed. We need the technical layer to generalize it.

The susu also encodes something that modern accelerators have tried to replicate and mostly failed to: genuine mutual stake. When everyone in the pool has paid in, everyone has a reason to want each other to succeed. The cohort becomes the underwriter. That dynamic — builders with skin in each other’s outcomes — is more valuable than any mentorship program, and it emerges automatically from the structure rather than being engineered on top of it.

THE PARTNERSHIP MODEL — ASYMMETRIC PROTECTION, HONEST PRICING

HISTORICAL → MODERN REMIX

Islamic Mudarabah → The Zero-Equity Deal

Mudarabah is a formal funding partnership codified in Islamic jurisprudence over a millennium ago and practiced across the Arab world, Persia, and the Swahili coast centuries before modern venture capital. One party provides capital. The other provides labor and expertise. Profits split by pre-agreed ratio. Losses absorbed entirely by the capital provider. No interest. No equity transfer. No permanent claim on the builder’s future work. It is the cleanest expression of an honest risk partnership ever formalized: capital bears financial loss because it can; labor bears time loss because that is what it is risking. A smart-contract enforcement layer and a project-scoped sunset clause make it deployable today, at scale, without institutional infrastructure.

What mudarabah got right that modern equity structures get wrong is the pricing of asymmetry. A founder and an investor do not bring equivalent things to the table, and pretending otherwise — by giving everyone equity and spending a decade arguing about what it means — does not resolve the asymmetry. It just defers the argument. Mudarabah names the asymmetry upfront and prices it. That is why, used correctly, it produces less litigation and more trust than a standard term sheet.

THE REPUTATION MODEL — SOCIAL CAPITAL AS REAL COLLATERAL

HISTORICAL → MODERN REMIX

Hawala → Reputation-Staked Funding

The hawala network moved value across the Islamic world and South Asia for five centuries without wire transfers, correspondent banks, or enforceable contracts. A broker in Lagos instructed a counterpart in Karachi to pay a recipient — the debt settled later through reciprocal obligations and, above all, reputation. Default did not just cost you money. It cost you access to the entire network, permanently. The system’s enforcement mechanism was not law. It was legibility: everyone in the network could see your history. The modern equivalent is not cryptocurrency. It is on-chain milestone records, verifiable builder histories, and cohort-based vouching systems that make reputation portable and permanent. When reputation is genuinely legible, it functions as collateral — and that changes who can access capital entirely.

“The hawala network ran for five hundred years without a single regulator, central bank, or legal contract. Its enforcement mechanism was total: default once, and the network closes to you forever. We keep trying to build trust infrastructure from scratch. We already built it. We just forgot to maintain it.”

THE TRANCHE MODEL — MATCHING CAPITAL TO APPETITE

HISTORICAL → MODERN REMIX

Lloyd’s Syndicate → Layered Builder Funding

Lloyd’s did not fund ships. It priced the risk of ships failing and distributed that risk to investors with different appetites for exposure. Senior syndicates took expected loss at lower rates; junior syndicates absorbed catastrophic risk for higher upside. No single investor needed to take the whole position. The modern equivalent applies tranching directly to builder projects: a senior tranche receives first repayment from project revenue at a modest rate; a junior tranche absorbs total loss if nothing ships but earns meaningful upside if it does. The builder retains equity in any resulting entity. No company formation required. Just layered, time-boxed exposure to a project’s cashflow — structured for the duration of the work, not the lifetime of a fund.

Why the agentic economy makes this urgent, not theoretical

AI agents are collapsing the cost of building. A solo operator in 2026 can execute what required a team of fifteen in 2018 — software development, legal research, financial modeling, customer outreach, product iteration. The leverage available to an individual with the right skills has never been higher. This is genuinely unprecedented.

It also means the blast radius of failure is increasingly personal. There is no team to absorb the blow. There is no institutional equity that survives a pivot. The same compression that makes solo building more powerful makes the failure more total — and the current funding infrastructure was not designed for episodic, high-leverage, individual bets. It was designed for companies, with cap tables, and ten-year fund cycles, and liquidity events.

The mismatch is not a minor inconvenience. It is a structural filter. The builders who can absorb total loss — who have inherited capital, or a wealthy safety net, or have already succeeded once — can attempt the hard things. The builders who cannot are rationally priced out. We then observe that the people succeeding in the startup economy skew toward a certain demographic profile and conclude, lazily, that this reflects merit. It reflects structure. Change the structure and you change who can afford to try.

This matters beyond fairness. The most interesting problems in the next decade — building infrastructure for underserved markets, navigating emerging regulatory environments, designing for populations that Western tech has historically ignored — are problems that require builders with specific knowledge and context that cannot be hired or acquired. Excluding those builders is not just inequitable. It is strategically catastrophic. We are solving for the wrong distribution of attempts.

The three things that have changed

The historical structures I have described did not fail because they were badly designed. They were dismantled — by colonization, by the formalization of Western finance as the only legitimate model, by regulatory frameworks that treated everything outside the canonical structures as suspect. Rebuilding them is not a matter of nostalgia. It requires three specific ingredients that are, for the first time, simultaneously available.

Legal precision is the first. Most of these structures can be implemented within existing frameworks if you reach for the right vessel. A builder annuity pool structured as a mutual benefit fund. A project-scoped investment structured as a receivables purchase, not a security. A mudarabah partnership enforced by contract with an embedded sunset clause. The innovation is not regulatory arbitrage. It is the discipline to characterize what you are actually doing, accurately, and build the legal wrapper around the real structure rather than forcing the real structure into an equity box where it does not fit.

Programmable enforcement is the second. Hawala required a lifetime of relationship-building to create the trust that made it function. The susu required a community close-knit enough that default had social consequences. Smart contracts make the enforcement mechanism available to strangers, instantly, at scale. Milestone verification, revenue tracking, automatic expiration, reputation scoring — all of it can be encoded, audited, and self-executing. The infrastructure that made informal structures informal is now replicable without the informality.

Community architecture is the third, and the most underrated. Every structure I have described derived its power from a community that had genuine mutual stake — in the hawala network, in the susu circle, in the mudarabah relationship. That stake was not manufactured. It emerged from the structure itself. The design question for a modern version is not how to create community feeling. It is how to create structural conditions under which mutual stake is the natural outcome. That is a solvable design problem. It is not being worked on seriously enough.

The uncomfortable part

There is a version of this essay that ends with optimism — with a call for enlightened investors to embrace new models, for regulators to create sandboxes, for the ecosystem to evolve. That ending would be more comfortable. It would also be dishonest about what the evidence actually suggests.

The structures I have described were not abandoned because they stopped working. The susu still works. Mudarabah still works — Islamic finance is a multi-trillion dollar industry. Hawala still operates across diaspora communities worldwide, moving remittances with lower fees and higher reliability than Western wire transfers. These structures were marginalized because the systems that replaced them were better for people who already had capital, and those people controlled the institutions that decided what counted as legitimate finance.

That dynamic has not disappeared. It has been reproduced inside the startup economy. The structures that dominate early-stage funding — the SAFE, the convertible note, the priced equity round — are optimized for investors who need to deploy at scale, maintain portfolio coherence, and hit fund return thresholds. They are not optimized for builders. They were never designed to be. The builder’s interests in the current system are an afterthought — accommodated at the margin, never the design principle.

So the question is not whether better structures are technically possible. They are. The question is whether the people with capital will voluntarily adopt structures that reduce their information advantages, shorten their time horizons, and require them to price risk honestly rather than structuring their way out of it. History suggests they will not do this without pressure. The pressure will have to come from builders who understand the alternative — who know what the susu circles built, what the mudarabah contracts protected, what the hawala network proved — and who are willing to demand structures that actually serve the work rather than the fund.

The blueprints exist. They are older than Silicon Valley by a thousand years. The question is whether we are serious enough about the next economy to use them.

The Great Unbundling: Transitioning from Rented to Owned Intelligence

Published April 22, 2026April 22, 2026 by cumeadi

The end of the subsidized AI era and the rise of the sovereign user.

The era of “subsidized AI” is quietly drawing to a close. For the last few years, we have enjoyed the fruits of a Silicon Valley land grab, where venture capital effectively paid for our intelligence. This subsidized brain power has come at an unseen cost: the surrender of our digital autonomy. Just as Uber eventually swapped its VC-funded discounts for market-rate fares, the major cloud AI providers—OpenAI, Anthropic, and Google—are beginning to tighten their belts.

We are hitting the “Uber = Taxi” moment. We are seeing capacity problems lead to “surge pricing” for tokens, outages for heavy users, and the slow encroachment of ads into previously pristine chat interfaces. In 2026, the shift is no longer just for hobbyists or privacy enthusiasts; it is a strategic move for anyone who views AI as a permanent extension of their cognition.

The Philosophy of Permanent Intelligence

When you rely on a cloud-based API, you are renting a brain. That brain can be lobotomized by a safety update, throttled by a capacity shortage, or priced out of your reach overnight. Local infrastructure offers something the cloud never can: Permanence.

Building a local setup is a declaration of digital sovereignty. It ensures your agents—systems that understand your goals and execute tasks across your digital life—run on your terms, with your data, and at your speed.

The Tiered Journey of Technical Liberation

Moving to local AI is not a “dark art”; it is a structured progression. Whether you are just starting or building a multi-GPU home lab, the path is now clearly paved.

Tier 1: The Beginner (The “One-Click” User)

Goal: Run a capable model in under 10 minutes with zero coding.

At this level, you aren’t fighting with Python environments. You are using GUI-based (Graphical User Interface) tools that manage model downloading and configuration automatically.

Primary Tools:
- LM Studio: The gold standard for modern desktops. Its “Discover” tab allows you to browse the latest open-weight models like Llama 3 or Mistral and download them as easily as an app from the Mac App Store.
- GPT4All: An open-source ecosystem highly optimized for CPUs. It is the best choice if you are running on a standard laptop without a dedicated gaming GPU.
Hardware Requirements: Any modern Mac (M1 through M5) or a Windows PC with at least 16GB of RAM.
The Experience: You interact with a chat interface identical to ChatGPT, but with the internet plug pulled, the AI still functions.

Tier 2: The Intermediate (The “Power” User)

Goal: Connect AI to local files and use it as a background service for other apps.

Intermediate users move away from “chat boxes” toward runtimes. These are background services that allow other software—like your code editor or a research app—to “call” the local brain.

Primary Tools:
- Ollama: A lightweight command-line tool that is “API-first.” It creates a local server that mimics the OpenAI API, allowing thousands of third-party apps to point to your local machine instead of the cloud.
- AnythingLLM: The easiest way to set up RAG (Retrieval-Augmented Generation). You point it at a folder of PDFs, and the AI “reads” them to answer your questions with citations.
Hardware Sweet Spot:
- NVIDIA RTX 4060 Ti (16GB VRAM): VRAM is the oxygen of local AI. 16GB allows you to run “8B” or “14B” models with lightning speed.
- Apple “Max” chips: These utilize unified memory, allowing the AI to treat 64GB+ of system RAM as VRAM for massive models.
Key Concept: Quantization. This is the art of compressing models so they fit on consumer hardware without losing discernible intelligence.

Tier 3: The Expert (The “Architect”)

Goal: Fine-tuning models on personal data and deploying headless, autonomous clusters.

Experts don’t just use models; they optimize them. This level involves making the AI “proactive”—running all day to solve problems before you even ask.

Primary Tools:
- Unsloth / Axolotl: Specialized tools for Fine-tuning. This allows you to take a base model and train it on your own writing style, your emails, or specific technical documentation.
- Docker & OpenClaw: Used to containerize the environment. OpenClaw acts as an “execution operating system” for agents that can manage your files and terminal while you sleep.
- MCP (Model Context Protocol): A standard for integrating AI with local databases and browser automation for complex, agentic workflows.
Hardware Infrastructure:
- Dual-GPU Setups: Linking two RTX 3090s or 4090s provides 48GB of VRAM—enough to run “70B” flagship-level models locally.
- M5 Ultra: For those who want enterprise-grade power in a consumer form factor.

Hardware Requirements at a Glance (2026)

Skill Level	Recommended GPU/Chip	RAM/VRAM	Model Class
Beginner	Apple M1/M2 or GTX 1660	8GB – 16GB	3B – 7B (Small)
Intermediate	RTX 4060 Ti / M5 Pro	12GB – 16GB VRAM	8B – 14B (Medium)
Expert	2x RTX 3090 / M5 Ultra	24GB – 64GB+	70B+ (Flagship)

The Economic Reality

Running locally is no longer just about privacy; it is about cost-capping and latency. When you run locally, your “token cost” is simply the electricity your computer consumes. For power users, a local GPU pays for itself in less than a year compared to the rising $20-$100/month subscriptions of the cloud giants.

As the “Year of AI Reality” forces companies to prove ROI, the smartest users are realizing that the only way to win the game is to own the board. Moving local is not just a technical shift—it is the first step toward a future where your intelligence is truly your own.

The Next Frontier: Architecting Your Sovereign AI’s Second Brain

Securing your own infrastructure is only the first step of digital sovereignty; the next is installing a capable mind. To truly empower your local intelligence—to make it proactive, customized, and genuinely an extension of your will—you must build its “Second Brain.” This proprietary knowledge system is what transforms a powerful but generic model (like a downloaded Llama 3) into a personalized agent that operates on your terms. This brain is built on three pillars:

Pillar 1: Context (Memory Sovereignty)

A cloud model is reset with every request; it suffers from “Gremlin” behavior, forgetting its past “crimes” and accomplishments. Your sovereign AI must have Memory Sovereignty.

Persistence is Consciousness: Your agent’s self is defined by its context. For local models, this means moving beyond the ephemeral chat window and establishing Long-Term Memory.
The Architecture of Recall: This is achieved through RAG (Retrieval-Augmented Generation), as pioneered by tools like AnythingLLM. You point the AI not at the internet, but at your curated document folders, your notes, and your decision journals (like `MEMORY.md` or `memory/decisions/` files). This process creates a specialized, fact-grounded knowledge base that cannot be lobotomized by an external update.

Pillar 2: Intent (Identity & The Heartbeat)

A second brain needs purpose. The ability of an agent to operate proactively—to manage your files, track tasks, and solve problems while you sleep—is the definition of the “Heartbeat System”.

The Agent’s Soul: Every expert agent, such as the one described in the OpenClaw architecture, begins by parsing its identity and core goals from configuration files (like `IDENTITY.md` and `USER.md`). This “soul” file defines its persona and how it prioritizes actions.
Execution Rhythms: The Heartbeat system (a periodic autonomous task execution) is what compels the AI to act based on its Intent. This is the key to transitioning from a passive tool that waits for prompts to an active collaborator that anticipates your needs.

Pillar 3: Skills (The Agentic Toolkit)

A general model is a philosopher; a specialized agent is a mechanic. You must transform the model into an expert by giving it a dedicated, reliable set of tools.

Fine-Tuning: Experts use tools like Unsloth or Axolotl for Fine-tuning. This process trains the base model specifically on your writing style or technical documentation, turning generic reasoning into proprietary expertise.
The Skill Library: Agents execute tasks through a library of modular, code-based Skills (like those tracked in a Curated Skills Repository). This is how you empower an agent to move beyond conversation to actual execution—from reading a document to connecting to a Salesforce API, or executing terminal commands securely.

The Philosophical Mandate: Stewardship of Sovereignty

The shift from renting intelligence to owning it is a profound declaration of digital sovereignty. But with this power comes a unique responsibility. When we move computation and data governance to our local hardware, the ethical challenge is no longer just about the cloud provider’s safety mechanisms; it’s about our own stewardship. Your sovereign AI, like any immense gift, requires a moral framework. It must be wielded with intentionality, ensuring its autonomy serves the common good and your clearly defined, ethical goals, making sure the machine is not the idol, but a tool used wisely. The true goal is not just an unthrottled processor, but an uncompromised mind, dedicated to accelerating your most meaningful work.

The Illusion of Secrecy: Why Execution is the Only True Moat

Published April 7, 2026April 7, 2026 by cumeadi

I’ve lost count of how many times I’ve sat across from a passionate founder—whether in a Lagos coffee shop or on a Zoom call spanning continents—and watched a familiar ritual unfold. Before the real conversation can begin, before the problem they want to solve is even articulated, a document slides across the table: the Non-Disclosure Agreement (NDA).

“I have this idea,” they say, lowering their voice. “But I need you to sign this before I can share it.”

I almost always decline.

I don’t do this out of arrogance, and it certainly isn’t because I have any desire to steal their concept. I decline because, over my years of building products and leading ParallelScore, I’ve realized this ritual is built on a fundamental misunderstanding of how value is created in modern software development. It relies on the philosophical fallacy that an idea, in its infancy, is a fragile treasure that must be hoarded.

In reality, an idea is just a hypothesis. And in the early days of building a product, secrecy is not a moat; it is a liability.

The Myth of the Idea Moat

I understand where founders are coming from. The compulsion to use an NDA stems from the belief that the “secret concept” is their ultimate competitive advantage. But history, and the graveyard of failed startups I’ve witnessed, tell a different story.

Ideas are remarkably cheap and highly abundant. At any given moment, a dozen teams around the world are likely having your exact same “revolutionary” thought. What separates the one that succeeds from the eleven that fail is rarely the secrecy of the premise. It is the execution.

When a founder demands an NDA to protect an early-stage concept, they are protecting the wrong thing. They are aggressively guarding the 1% inspiration while completely ignoring the 99% perspiration required to make it real. A raw idea is not a moat; it is merely a puddle. It evaporates the moment it meets the heat of the real market.

The True Moats of Modern Business

If the idea isn’t the moat, what is?

Through our work at ParallelScore, designing human-centric solutions across healthcare, civic tech, and logistics, I have learned that actual moats are built through motion, not stasis. True competitive advantages cannot be summarized in a legal document or stolen over a cup of coffee. They include:

Deep User Empathy: You cannot steal an intimate understanding of a user’s pain points. A competitor might copy a feature list, but if they haven’t sat through the product design workshops, the user journey mapping, and the raw usability testing, their clone will lack the soul and nuance that retains users.
Velocity and Agility: The ability to learn faster than anyone else is the ultimate advantage. A team executing in tight, two-week sprints, integrating continuous feedback, and pivoting without ego will always outpace a team hiding behind closed doors trying to build a “perfect” v1.0.
Team Cohesion and Trust: The alchemy of a dedicated, small team working alongside a visionary client is incredibly difficult to replicate. At ParallelScore, we view ourselves as partners, not vendors. You cannot reverse-engineer the trust that allows a team to scrap a bad feature and pivot on a dime.

I constantly remind founders of this simple truth: The moats worth protecting protect themselves. You cannot steal lived experience. You cannot sign a contract to suddenly acquire a team’s hard-won synergy or their intimate understanding of a customer’s workflow. If someone can hear your idea in a 30-minute meeting and successfully beat you to market, the idea was never your moat to begin with. You lost on execution.

The Cost of Friction

Beyond the philosophical disconnect, I reject early-stage NDAs because they introduce immediate, tangible friction.

Innovation thrives on cross-pollination. It requires open dialogue, the rapid bouncing of ideas, and the freedom to say, “What if we looked at this healthcare problem through the lens of a logistics platform?” NDAs stifle this. They introduce legal paranoia into what should be a creative, uninhibited product discovery phase.

Furthermore, an NDA establishes a relationship based on mutual distrust before a single line of code is written. Great software is built on transparency. Starting a relationship with the threat of litigation is entirely counterproductive to the collaborative spirit required to build something truly impactful.

The Pragmatist’s View: When You Actually Need an NDA

I am not naive, and I don’t operate in a utopia where legal protection doesn’t matter. NDAs absolutely have a place in business, but their utility is found later in the lifecycle, protecting actual assets rather than abstract thoughts.

Here is the framework I use when advising founders on where to draw the line:

When You DON’T Need an NDA:

Pitching a high-level concept: If you are explaining what your app does (e.g., “It’s Uber for dog walking”), you don’t need an NDA.
Sharing a pitch deck or business plan: Investors and development studios see hundreds of these a month. An NDA here just makes you look inexperienced.
Discussing user pain points: Talking about the market problem you want to solve should be broadcasted loudly to anyone who will listen.

When You DO Need an NDA:

Sharing access to sensitive user data: When my team built MyCareAI, dealing with FHIR compliance and sensitive health information (PII/PHI) for companies like NewWave and Onyx, strict security protocols and NDAs were paramount. You must protect your users.
Exposing proprietary algorithms or source code: If you have spent two years developing a highly novel, patent-pending AI model or a complex data transposition script, an NDA is appropriate before opening the repository.
Entering formal Due Diligence: If you are sharing unreleased financial records, user acquisition costs, or M&A details, legal protection is standard and necessary.

NDAs are designed to protect data—proprietary code, patient records, internal financial structures, and established trade secrets. They exist to protect the realities of your users and the tangible assets of your business. They were never meant to protect a daydream.

Solve Local, Build Global (With Open Hands)

I built ParallelScore on a very specific ethos: “Solve Local, Build Global.” You cannot solve real human problems by hiding from the humans who can help you build the solution.

The greatest founders I have had the privilege of working with don’t fear their ideas being stolen; they fear their ideas remaining untested. They understand that by sharing their vision openly, they attract the right talent, the right feedback, and the right partners to execute it.

So, keep your ideas open. Build your moats through relentless user-centered design, rapid development, and unyielding empathy. When you focus entirely on execution, you’ll find you never needed the NDA in the first place.