Two of the Smartest Research Shops on Wall Street Disagree on AI. They Are Both Missing the Point.

In 1997 everyone was certain the internet would reshape the economy in five years. They were right about everything except what mattered most: the timing, the mechanism, and what you were supposed to do about it.

We are doing it again with AI. And the people who get this right will not be the ones who picked the correct forecast.

I founded my first technology company in that era. The internet was going to eliminate every middleman, destroy retail, make geography irrelevant, and reshape the global economy inside of five years. The pundits were both completely right and spectacularly wrong at the same time. I am watching the same movie again with AI. And just like then, the most interesting question is not who wins the macro debate. It is what you do about it while everyone else is still arguing.

Two smart people. Two very different conclusions.

CitriniResearch issued a research note in February outlining a fictional scenario set in 2028. The S&P is down 38% from its 2026 highs. Unemployment has hit 10.2%. White collar layoffs have triggered a consumer spending collapse, a private credit crisis built on SaaS debt that assumed ARR would stay recurring, and early stress in the $13 trillion residential mortgage market. The mechanism is not a classic financial bubble. It is a feedback loop: AI improves, payrolls shrink, spending softens, companies buy more AI to protect margins, AI improves again. A self-reinforcing cycle with no brakes.

That same week, Citadel Securities published a rebuttal. Their argument is grounded in actual data. AI adoption at work, measured by the St. Louis Fed's Real Time Population Survey, is not accelerating non-linearly. It looks more like a standard S-curve. Job postings for software engineers are rising, up 11% year over year. New business formation is expanding rapidly. Displacing white collar work at scale requires orders of magnitude more compute than currently deployed, and if automation expands rapidly, compute costs rise, creating a natural economic barrier to further recursion. Recursive capability, they argue, does not imply recursive adoption.

Both papers are serious. Both are worth reading. And both miss the point for anyone trying to build or lead something right now.

The macro debate is the wrong question.

What I know from building and taking a company from zero to $200 million in revenue through a public offering is that the macro environment is the water you swim in. You cannot choose it. You cannot wait for the researchers to agree before you act. The companies that win technology transitions are not the ones with the best forecast. They are the ones that build the right thing faster than the transition makes the old thing obsolete.

The Citrini scenario is not prediction, it is a risk map. The Citadel paper is not reassurance, it is a pacing metric. Read them together and you get something useful: the disruption is real, the timing is uncertain, and the window for builders is open right now, before the feedback loop closes or the S-curve plateaus.

What Citrini gets right is the mechanism. The companies most vulnerable to AI are becoming its most aggressive adopters. ServiceNow cutting 15% of its workforce while selling workflow automation. Every dollar saved on headcount flowing into the capability that enables the next round of cuts. That is not dystopia. That is rational corporate behavior under pressure, and it is already happening. The question is not whether this continues. It is whether you are building the platform that benefits from it or the platform that gets rationalized away.

What Citadel gets right is the velocity of organizational change. Change is slow. Change in regulated environments is slower. There is an enormous gap between what AI can demonstrate in a proof of concept and what an enterprise will actually trust with its mission-critical workflows. That gap is measured in years, not quarters, and it is where most of the real business value gets built.

What this means for builders right now.

Three points I would make to anyone building or leading during this period.

First, the data advantage is the moat. The Citrini scenario describes AI agents routing around friction, collapsing intermediation, and destroying businesses built on information asymmetry. That is true. But proprietary data that cannot be replicated, real transaction history, behavioral patterns at scale, performance data across thousands of deals, that is not friction. That is the foundation of a defensible platform. The businesses that survive this transition are the ones sitting on data that gets more valuable as AI improves, not less.

Second, the workflow complexity premium is real and temporary. Citadel is right that regulated, high stakes environments have coordination friction, liability constraints, and trust barriers that slow AI adoption. That premium exists today. It will not exist in five years. The window to build the category defining platform inside that premium, before the friction disappears, is what gives this moment urgency.

Third, the leadership question is not about AI literacy. It is about judgment under uncertainty. Every executive I know is consuming the same research, attending the same conferences, and asking the same questions. The difference is not who understands the technology. It is who can make clear decisions with incomplete information, build and execute with a team while the strategy is still evolving, and maintain a long view when the short term data is volatile. That is the differentiation between founders and operators, and between operators and administrators.

Just about every article about AI stops here. They treat the macro debate as the whole story and leave the most important question untouched. I want to go somewhere most AI commentators do not bother to go: the neuroscience of why original human thought may represent a permanent ceiling that AI cannot reach, not eventually, but in principle.

Why AI will never reach original human thought. The science is clear.

This is the section most AI commentators gloss over. Not because the research does not exist, but because it requires coming to terms with an uncomfortable truth: there may be a structural gap between what AI achieves and what original human thought represents. Not in five years. I don't think ever.

The first pillar is embodiment. When you have a gut feeling that something is wrong with a deal, that is not noise. That is your body processing risk signals that have not yet reached your conscious reasoning. Antonio Damasio's somatic marker hypothesis demonstrated that emotion and bodily state are not separate from reasoning. They are constitutive of it (Damasio, 1994). We think with our bodies. Patients whose brain regions that integrate emotion into decision making were damaged did not become more rational when their emotional processing was disrupted. They became incapable of making decisions at all. The implication for AI is direct: a system with no hormonal architecture, no visceral sense of risk, no embodied history of consequence, is not reasoning. It is pattern matching at an unprecedented scale. These are not the same thing.

The second pillar is what happens when you stop trying. Your brain has a network that activates specifically when you are not focused on a task. When you are in the shower, driving without thinking, or staring out a window, that network is running. Scientists refer to this as the Default Mode Network. It turns out to be central to original creative thought. Beaty and colleagues found that the most creative people show stronger connectivity between this rest-state network and the executive network that evaluates and focuses (Beaty et al., 2016). This is the neuroscience of the 3am insight, the idea that arrives when you step away from the problem. Large language models do not have a Default Mode Network. They do not rest. They do not daydream. They only process when prompted. The architecture that generates human creative breakthroughs does not exist in any current AI system, nor is there an apparent computational analog.

The third pillar is the one nobody has figured out yet. We do not have a scientific explanation for why experiencing something feels like anything at all. Why does seeing red feel like something? Why does grief feel like something? David Chalmers called this the hard problem of consciousness in 1995, and the field has made little progress toward resolving it since (Chalmers, 1995). Original ideas frequently emerge from subjective states: grief, wonder, confusion, the particular texture of a lived experience that cannot be retrieved from a training corpus because it was never written down. AI processes tokens. It has no inner experience. Whether that gap is bridgeable in principle is genuinely unknown, but nothing in the current trajectory of AI development addresses it at all.

I want to be precise about what this does and does not mean. AI will get dramatically better at everything humans do deliberately: retrieval, synthesis, pattern recognition, rapid iteration on existing frameworks, stress testing ideas against vast bodies of knowledge. These are enormous capabilities and anyone not using them is leaving a real advantage on the table. But the original thought that decides what to build, which problem is worth solving, what the insight is that nobody else has seen yet, that still has to come from a human who has lived something, felt something, and has the embodied architecture to generate a connection that falls outside the distribution of everything that has been written before.

AI is the best recall and synthesis engine ever built. That makes it a powerful amplifier for human original thought. The original thought still has to come from you.

The honest answer.

I do not know if the Citrini scenario unfolds in 2028. Neither does Citrini. Neither does Citadel. What I know is that every significant technology transition in my career rewarded the people who stopped debating the macro and started building the right thing at the right moment.

The neuroscience tells us that the limited resource in an AI-abundant world is not data, compute, or even domain expertise. It is the human capacity for original thought: embodied, emotionally integrated, and architecturally unique.

Your 401(k) will figure itself out. The more interesting question is what you are building while everyone else watches it gyrate.

References