The AI Gold Rush Is Ending. The Age of Intelligence Engineering Is Beginning.
July 18, 2026
If you remember, we have always maintained that trying to pick the exact top is a fool’s game. Markets possess an extraordinary talent for humiliating those who insist on playing hero, so instead of trying to forecast the precise moment enthusiasm collapses, we prefer to study the emotional forces driving it. Trends are born in psychology long before they become visible in price, and nowhere has that been more evident than in the current AI mania.
One of the biggest mistakes made in the West has been reducing artificial intelligence to large language models. LLMs are not AI. They are one branch of AI, and even then they are better described as highly sophisticated neural networks than the digital consciousness their marketing departments occasionally imply. Listen to the promotional language coming from the industry’s biggest names and one could be forgiven for believing humanity had stumbled upon an electronic life form. Scratch beneath the surface, however, and the miracle cure sometimes begins to resemble yesterday’s snake oil poured into a far more attractive bottle.
Because trends matter more than headlines, considerable time was spent over the past year working alongside people who actually build these systems rather than merely advertise them. Once the engineering reality is separated from the marketing mythology, an entirely different picture begins to emerge.
The Internet Is Knowledge and Noise Combined
The first observation is surprisingly simple. Most frontier models learned from the internet, yet the internet is one of the greatest collections of human knowledge ever assembled alongside one of the greatest collections of repetition, advertising, opinion, misinformation and outright nonsense. Depending on the subject, a remarkable percentage of the available information contributes little more than noise. Expecting wisdom to emerge automatically from that environment is rather like expecting a library filled with cookbooks to accidentally produce a Michelin-starred chef.
That is not how expertise develops.
How Real Expertise Actually Develops
A brilliant engineering student does not become exceptional by reading every journal ever published. The student studies under outstanding engineers who challenge assumptions, eliminate weak reasoning, compress decades of experience into practical judgement and teach something no textbook fully captures: how to distinguish signal from noise. Knowledge accumulates. Wisdom distils.
Artificial intelligence has largely been asked to perform the opposite exercise. It absorbs oceans of unfiltered information and is then expected to produce coherent judgement. The remarkable achievement is not that these systems occasionally hallucinate. It is that they perform as well as they do despite the quality of much of their training material. Because they are designed to sound helpful, they often deliver answers with extraordinary confidence while quietly saying very little. Confidence has never been reliable evidence of accuracy, although that distinction occasionally disappears somewhere between the engineering department and the marketing department.
When Genuine Expertise Enters the Conversation
Everything changes once genuine expertise enters the conversation.
An experienced engineer, physician, chemist or financier does not simply accept the model’s first answer. They interrogate it, redirect it, expose weak assumptions, refine the reasoning and force the system toward better conclusions. The expert becomes the missing layer of intelligence. Instead of wandering through an enormous landscape of information, the model is guided through it with purpose.
That observation explains why AI is far more likely to amplify expertise than replace it. A competent engineer becomes significantly more productive because experience allows mistakes to be recognised before they become decisions. A world-class engineer becomes even more formidable because judgement and computational speed reinforce one another. Someone lacking the underlying knowledge often mistakes polished nonsense for genuine insight because they possess no reliable mechanism for telling the difference.
Beyond the Obsession With Bigger Models
This is why the industry’s obsession with larger models may gradually give way to something far more interesting.
The future is unlikely to belong to one enormous system attempting to master every conceivable discipline. It is far more likely to belong to architectures capable of recognising the nature of a problem before routing it through specialised reasoning engines designed specifically for engineering, medicine, chemistry, mathematics, finance or physics. Every step reduces unnecessary computation, lowers power consumption, improves accuracy and produces better answers using fewer resources.
Ironically, the export restrictions imposed on China may have accelerated precisely this transition. Limited access to the most advanced hardware forced Chinese researchers to optimise software, refine architectures, improve inference, compress models and extract more intelligence from every available unit of compute. Comfort encourages expansion. Constraints encourage invention.
Perhaps the old saying deserves a slight revision. Necessity may be the mother of invention. Desperation is often the mother of inventions that change the rules.
