NP Nithin Palal

Chapter 14: The Commodity Trap

Part IV: Conscious Creation

· 18 min read · Discuss · Plain text

AI-Proof: What Machines Will Never Be — Volume II: Application, Part IV: Conscious Creation, Chapter 14: The Commodity Trap

The numbers were perfect. That was the problem.

Chidi Okonkwo checked the dashboard from a rented desk in a co-working space in Yaba, Lagos, and felt the specific exhilaration of a line going up and to the right. Four hundred paying users in three weeks. Two thousand by month three. A seed round from a Lagos-based venture fund by month four. The product — ClarityAI — took a large language model’s API, wrapped a clean interface around it, added prompts optimised for African small business owners, and charged a subscription for the ability to draft contracts, write marketing copy, generate financial projections, and answer basic legal questions. It worked. It worked immediately. It worked the way things work when you are twenty-nine and have studied computer science at the University of Lagos and spent two years at a fintech on Victoria Island and left to build something of your own and you are fast, resourceful, and technically gifted — the kind of founder who ships over a weekend and has customers by Monday.

Chidi loved the speed. Not as a strategy. As a feeling. The feeling of the numbers rising was the feeling of being right, and being right at velocity was the purest form of professional joy he had known. He refreshed the dashboard the way his mother checked the weather — not because the information would change his plans but because the act of checking was itself a comfort. Revenue: climbing. Churn: low. Acquisition cost: falling. Everything pointed the same way.

He did not notice what the numbers were not telling him.

They were not telling him that his product’s entire value — the interface, the prompts, the subscription page — could be described in a design document that a competent team would read in an afternoon. They were not telling him that the model provider was building a simplified interface of its own. They were not telling him that a competitor in Nairobi was already building a near-identical product with better localisation across three African languages. They were not telling him that the thing the numbers measured — growth — and the thing the business needed — depth — were not the same thing.

The numbers told him he was winning. He was evaporating.

ClarityAI died the way most AI wrappers die: not in a single catastrophe but in a convergence. The model provider released its own free interface. The Nairobi competitor launched — built in under a fortnight. An API update broke several of ClarityAI’s custom prompts, and Chidi spent two weeks rebuilding functionality his users had already learned to get elsewhere. His churn rate climbed. His acquisition costs doubled, then tripled. His two employees — a designer and a customer-support coordinator, both friends from university — sat across from him in the same co-working space and waited for him to say what everyone already knew.

Seven months. Launch to collapse. April to November.

On a Tuesday evening in November, Chidi went home to his apartment in Surulere and lay on the bed and stared at the ceiling fan and let the question arrive that his velocity had been outrunning since April. Not the strategic question — he would get to that. The simpler question. The one that felt like a hand on his chest.

What did I build?

An interface. Some prompts. A subscription page. A layer of convenience wrapped around someone else’s intelligence. When the intelligence improved, the layer shrank. When the intelligence provider offered its own interface, the users left. When a competitor used the same intelligence to build a similar layer, the differentiation vanished. ClarityAI was not a company. It was a feature of a platform that had not yet built that feature itself.

The lesson of every platform cycle — a lesson Chidi knew intellectually and had failed to feel in his body until that Tuesday evening — is that platforms build their own features. Eventually. Without malice. Without even, necessarily, noticing the companies they absorb in the process.

14.1 Why Most AI Startups Will Die

The claim requires qualification, because without it the claim is just doom and doom is cheap.

Nobody knows how many AI startups launched between 2023 and 2025. The estimates range from tens of thousands to over a hundred thousand, depending on how you count and what you count. What is not in dispute is the scale: the wave of AI company formation in this period was among the largest in the history of technology, rivalling and possibly exceeding the dot-com surge of 1999–2000 and the mobile app explosion of 2009–2012. The structural forces acting on that wave — platform absorption, competitive replication, model commoditisation — suggest that the majority of these companies will not survive as independent entities.

But “majority” is not “all.” Some companies that began as wrappers survived — by accumulating proprietary data, by embedding into enterprise workflows, by building distribution moats the platform could not replicate by adding a feature. The wrapper was the entry point, not the business. The business was what grew around it. This chapter is about the difference between those two things — and about why the difference is invisible to a founder moving at speed.

The diagnostic is straightforward: a thin wrapper is a product whose core value is access to a capability the underlying model already provides. If a competitor can replicate your product in a fortnight using the same API, you have a wrapper, not a business. That test is sufficient for recognition. What it does not explain is compulsion. Why do intelligent, capable, well-warned founders keep building wrappers?

Because the wrapper is what the market rewards first.

Investors want traction. Users want solutions. The competitive landscape punishes hesitation. And a wrapper, in its first months, is indistinguishable from a real business. It has revenue. It has growth. It has users who genuinely value the product. The metrics are right. The problem is that “right” is temporary, and the window during which the metrics hold is shrinking.

Cloud computing took roughly seven years to mature from product to infrastructure — from the launch of Amazon Web Services in 2006 to the point, around 2013, when building on cloud was no longer a differentiator but a default. The large-language-model landscape is compressing the same transition into a fraction of that time. The window during which a wrapper can capture value before the platform absorbs the function has narrowed from years to months. Founders who would have had time, in an earlier cycle, to deepen their product before competitive pressure arrived now find that the deepening and the pressure happen simultaneously.

Chidi had the time of a season. He used it to build faster. Speed was the only tool he trusted, and speed is the commodity trap’s perfect bait: it produces the appearance of progress while consuming the time that depth requires.

14.2 The Phone Call That Changed the Product

In São Paulo, a woman named Isabella Cavalcanti built the same trap and escaped it by accident.

Her product was a legal-document generator for Brazilian microempresas — another wrapper, another API, another set of carefully engineered prompts. The users were small-business owners in the informal economy who could now generate standard contracts and tax filings that would have cost them hundreds of reais in legal fees. Forty reais a month. Twelve hundred subscribers by month three. The growth chart looked like Chidi’s. The danger was the same.

On a Thursday in September, Isabella called a user named Cláudia who ran a cleaning service in Guarulhos. This was not a survey. Isabella had made a decision, three weeks earlier, to call ten users a week and ask one question: what would you do if this product disappeared tomorrow?

Most users said they would find something similar. This was the answer Isabella expected and dreaded, because “something similar” meant “another wrapper,” which meant her product was interchangeable, which meant it was dying on a timeline she had not yet calculated.

Cláudia said something different.

“I would call my cousin’s husband,” she said. “He is a lawyer. He is expensive. But he knows the rules.”

Isabella asked what she meant by “the rules.”

“The rules for Guarulhos,” Cláudia said. “The rules for a cleaning service. Not the general rules. The specific ones. The ones that changed last month. Your product knows them. The free version does not.”

Isabella sat in her apartment in Pinheiros, phone pressed to her ear, and felt something shift. She had built the product around AI-generated contracts. Cláudia was paying for regulatory specificity — the fact that the prompts had been engineered to reflect the tax regime for MEIs, the contractual language required by São Paulo prefeituras, the quarterly regulatory updates that no general-purpose model tracked. The AI was the delivery mechanism. The knowledge was the product.

Cláudia could not have said it that way. She said: your product knows the rules. But what she meant was: your product knows my world. And knowing someone’s world is not a feature. It is a relationship.

Isabella did what Chidi did not do: she stopped improving her prompts and started building the thing her users actually valued. She hired a paralegal to monitor Brazilian tax and labour law changes. She built a database of jurisdiction-specific requirements that her tool updated weekly, feeding the changes into the model’s context so that every generated document reflected the current legal reality, not the training data’s snapshot. Within six months, the product could do something no general-purpose model could replicate: generate a contract that was correct for a specific business type, in a specific municipality, under this week’s regulatory framework.

The wrapper had become something else. Not because the technology changed. Because the founder’s attention changed. Isabella had been looking at the dashboard. Cláudia’s phone call made her look at the world behind the dashboard — the specific, local, constantly shifting reality that her users navigated every day. The dashboard measured growth. The phone calls revealed what the growth was made of.

This is the difference the commodity trap exploits. A wrapper accumulates users. A business accumulates something that exists nowhere else — data, knowledge, relationships, the kind of understanding that requires a human being who has listened carefully enough to know what a specific population actually needs. Users are valuable but rented. They arrive for the convenience and leave for greater convenience. The thing that stays is what you built underneath the convenience. Isabella built a regulatory intelligence layer. When the model improved, it made her layer more useful, not less — the better the AI’s output, the more her jurisdiction-specific corrections ensured that output was correct, current, and real. The same force that killed wrappers strengthened her product.

14.3 Six Questions

The anti-wrapper test is not a checklist. It is a discipline of self-interrogation, and the discipline matters more than the answers, because the answers change as the product evolves. A company that passes the test today can fail it in six months if attention drifts.

The first question is the one Chidi answered on a whiteboard in Yaba in January, after closing ClarityAI in December, after his investors had been understanding in the way investors are when the loss is small enough to write off, after the designer had found work at a fintech on Victoria Island and the support coordinator was at an NGO in Lekki, after the co-working space was quiet in the way only a failed startup’s desk is quiet.

If the underlying model disappeared tomorrow, what would remain?

Not the interface. Interfaces are trivial. What data have you accumulated that exists nowhere else? What integrations have you built that would cost months to replicate? What relationships survive a change in technology? Chidi’s answer: nothing. Isabella’s answer, by then: a regulatory database covering three hundred Brazilian municipalities, updated weekly, with no equivalent anywhere in the market.

Can a competitor replicate your product in a fortnight using the same API?

This is the speed test. Founders fail it because they confuse the difficulty of building their specific product with the difficulty of building something equivalent. Chidi spent weeks refining his prompts. A competitor matched the result in twelve days with different prompts. The effort was real. The moat was not.

Does the product improve the more people use it?

A product with network effects compounds over time. A product without them has the same value at ten thousand users as at ten, adjusted only for revenue. This distinction is existential, and it is the one most founders lie to themselves about.

If the model provider built your feature tomorrow, what percentage of your users would stay?

Users stay for three reasons: the product does something the platform cannot, it is embedded deeply enough that switching is prohibitively costly, or the relationship — brand, community, trust — is something a product announcement cannot replicate. If the honest answer is that most users would leave, the user base is a loan, not an asset.

What is the product’s judgement layer?

Every valuable product contains a point where human judgement has been applied — a decision about what to prioritise, what to suppress, how to frame the output, for whom. If that layer is limited to prompt engineering, the judgement is thin and replicable. If it includes proprietary data, editorial decisions, and user-informed preferences refined through months of real use, the judgement runs deep. Deep judgement is a moat. Thin judgement is a configuration file.

Would the product be worth building if the model were free?

If the entire value proposition is making AI accessible, a free model eliminates the need. If the value would remain — because of data, community, integration, domain expertise, curation — then the AI serves the product rather than constituting it.

Chidi answered all six in January. Each answer: nothing. He did not erase the whiteboard. He sat with it for a long time, the way a doctor sits with a scan that confirms what the symptoms had already told him.

Then he went to visit his uncle.

14.4 Feature vs. Infrastructure

The distinction is old. Stripe did not solve payments. Stripe made it possible for any developer to build a product that accepts payments — and in doing so, became the layer on which an ecosystem grew. Shopify did not sell products online. Shopify made it possible for anyone to build a shop. The pattern: the companies that survive platform transitions are the ones that become platforms themselves. A feature solves a problem. Infrastructure creates the conditions in which problems are solved.

Most AI startups are features. They solve a specific problem using a specific model in a specific way. The solution is real. But the product does not create the conditions for an ecosystem to form around it. It sits on someone else’s infrastructure and adds a layer of value that the infrastructure provider can absorb at any time.

The infrastructure play is harder and slower. It requires data pipelines that improve with use, integration points that deepen over time, feedback loops that make the product more valuable in ways the raw model is not. It requires depth. And depth is what the market selects against, because depth produces less visible progress in the early months. The founder building data pipelines has less to show an investor than the founder shipping features. The market selects for wrappers because wrappers are legible. They have metrics. They have the shape of success.

Infrastructure has the shape of patience.

The conscious founder plays both games simultaneously — shipping features fast enough to maintain relevance while building infrastructure that makes the product impossible to replicate. The ability to hold both, to ship and question in the same breath, is not a strategy. It is an act of consciousness: the awareness that sees the short term and the long term without collapsing into either.

Not every founder needs to build a global platform. The infrastructure play has a smaller version available to every AI product: becoming infrastructure for a specific community, a specific workflow, a specific domain. Isabella’s product is not Stripe. It is infrastructure for Brazilian microempresas navigating tax compliance. That is enough. That is a business — because a road, once built, is used even when the engines change.

14.5 Building Your House on Someone Else’s Land

Model dependency is the operational expression of the commodity trap. It manifests in three forms.

Price dependency: your margins are the model provider’s pricing decision. An API cost increase can destroy your unit economics overnight. A rate-limit adjustment can make your product unusable at scale. You are a tenant whose rent can be raised without negotiation.

Capability dependency: your product’s boundaries are the model’s boundaries. When the model improves, every product using it improves equally — which means your improvement is not a differentiator. When the model degrades between versions in areas your product relies on, your product degrades with it, and you cannot fix the underlying problem.

Strategic dependency: the most dangerous, because the least visible. The model provider has a roadmap not shared with you. When the provider decides to enter your market, you learn about it when the rest of the world does — at a keynote, as a line item, while your business model is mentioned only in the Q&A.

It is worth stating, because an informed reader will note it regardless, that not all model providers behave identically. Some have signalled that they do not intend to compete with their developer ecosystem. Others have systematically absorbed successful use cases into their own product offerings. The variation is real. But signals are not contracts, and a company whose survival depends on a competitor’s restraint is a company built on a promise.

The escape is not model avoidance. AI is infrastructure. Avoiding it is choosing irrelevance. The escape is model agnosticism — treating the model as a replaceable component the way a web application treats a database engine. You do not build your company around a specific database. You build around the data, the logic, and the experience. You swap the engine when a better one arrives.

But architecture alone does not solve the deeper problem. If your value is primarily intelligence-dependent — if the product is fundamentally “make the AI easier to use” — then model agnosticism protects you from operational risk without protecting you from strategic risk. The real escape is the same escape as from the commodity trap itself: build value that resides not in the intelligence but in the consciousness — the human judgement, domain knowledge, and trust — that the intelligence serves but cannot replace.

Chidi built his house on someone else’s land. He knew it, the way you know a headache is coming: a pressure you can feel but choose not to name, because naming it would require you to stop doing the thing that feels like progress. The knowledge that the land is not yours is the kind of knowledge that is easy to suppress when the numbers are rising. It requires consciousness to hold.

14.6 What Emeka Knew

Chidi’s uncle Emeka had run a trading operation in Balogun Market for twenty-two years, connecting textile suppliers across West Africa with retailers in Lagos, Accra, Abidjan, and Douala. The business had no app, no dashboard, no investor deck. It had Emeka — sitting in a plastics chair in front of his shop on a Tuesday afternoon in February, receiving phone calls, drinking Fanta, and making decisions that involved the livelihoods of dozens of families in ways that no algorithm could map and no model could score.

Chidi went to visit him. Not for advice. For proximity to something real.

He stayed a week. He watched.

On the second day, a new supplier arrived — a young man from Aba named Tobechukwu, who wanted to sell wax-print fabric through Emeka’s network. He brought samples. The prints were clean, the colours fast, the pricing competitive. He had references: a cousin in the market, two retailers in Surulere who had already placed orders. On paper — if paper existed in Balogun Market, which it mostly did not — the supplier checked every box.

Emeka examined the fabric the way Rajan examines cotton — with his hands, holding it to the light, folding and refolding. But where Rajan reads the fabric, Emeka read the person. He asked about Tobechukwu’s father. About how long the family had been in the textile trade. About who had taught him to assess dye quality. He asked whether Tobechukwu owned his workshop or rented it. He asked, casually, about the rains — whether the last season had been good for transport to Aba.

These were not small-talk questions. Each one was a probe. Emeka was mapping a network of reliability indicators that did not exist in any database: family reputation, which predicted whether a supplier would honour commitments even when it cost them. Ownership versus rental, which predicted whether they would be in business next year. Knowledge of dye chemistry, which predicted whether they actually understood their own product or were reselling someone else’s. The question about the rains was the subtlest — if Tobechukwu knew the transport impact of the rainy season without hesitating, he was making the journey to Lagos regularly, which meant he was maintaining the relationship, not just seeking a transaction.

Tobechukwu hesitated on the rains.

Emeka did not refuse him. He placed a small order — far smaller than the samples warranted. A test. If the delivery came on time and matched the samples, the next order would be larger. If it came late or short, there would not be a next order. This was the credit system of Balogun Market: trust extended incrementally, withdrawn absolutely, accumulated over years of consistent behaviour. No contract. No app. No AI. Just a man in a plastics chair who had spent two decades learning to read the distance between what people said and what they would do.

Chidi watched this and something turned over in him.

He had built ClarityAI to help people like his uncle — African small business owners navigating complex decisions. But he had heard the problem wrong. He had heard: these people need better tools. What Emeka’s Tuesday afternoon was telling him was: these people have extraordinary tools. They are invisible because they are relational, experiential, and embodied in human beings who have been doing this for decades. The problem is not that the tools do not exist. The problem is that they cannot compound. When Emeka retires, his knowledge retires with him. The network of trust he has built over twenty-two years — the assessments, the reputations, the nuanced understanding of who can be relied upon and why — is not written down. It is not structured. It cannot be transferred, searched, or built upon by the next generation.

That was infrastructure worth building.

Not a wrapper around intelligence. A platform around consciousness — the accumulated human judgement of traders who had been making relational assessments for decades, now captured, structured, and made compoundable. AI was the delivery mechanism. Emeka’s knowledge — and the knowledge of every trader like him across West Africa — was the product.

Chidi did not build it immediately. He spent two more months in the market, talking to traders, mapping the relationships and information flows that constituted the market’s invisible architecture. He was doing what this book calls conscious building: beginning not with the technology but with the human system the technology would serve, understanding that system deeply enough to know where AI could amplify human capability without replacing human judgement.

The product he built — the second company, the one that survived — is the subject of what follows. What matters here is not the product but the pause that preceded it. Chidi did not become someone else. His speed was still there. His resourcefulness was intact. What changed was where he pointed his attention.

The commodity trap does not catch slow founders. It catches fast ones — the ones so good at building that they build before they understand what they are building. It catches the ones for whom the act of creation is so rewarding that they do not ask whether the thing they created deserves to exist independently or is merely a temporary expression of a technology that the market will absorb.

Emeka, in his plastics chair, had never thought about any of this. He had no framework for moats or wrappers or model dependency. He had something better: twenty-two years of attention to the difference between what people say and what they do. That attention was his product. It was, in the language of this book, consciousness applied so consistently and for so long that it had become a form of capital.

Chidi is building again. The whiteboard in Yaba still has the six questions on it. He has not erased them. They are not a test he passed. They are the questions he asks every week, the way Emeka asks about the rains — not because the answer matters on its own but because the willingness to keep asking is the whole point.

↑ Back to top

Discuss

Loading the discussion…

More in Artificial Intelligence
  •  ·  26 min read

    Epilogue

    Late in the Gītā, the one who knows most says: ponder it thoroughly, then act as you like. A machine can now set out the reasons on every side. Saying “I answer for this” then is not pride. It is a reply, for your part, to those owed one.

  •  ·  37 min read

    Chapter 40: What This Series Cannot Tell You

    The print edition’s last chapter reduced the book to one sentence, and rested human worth on what machines cannot do. I withdraw both. A book can date its claims and correct them in public. It cannot sit in the room where you decide.

  •  ·  6 min read

    Part IX: The Unfinished Chapter

    The print edition’s last Part said that consciousness never concludes. This one is a single chapter that turns the series on itself: what a series can stand behind, what it owes those who read the print edition, and what it leaves open.