The world just changed in a way we have never seen in all of human history, and almost nobody is paying attention. People are walking around, living day by day, not understanding the implications of what has already happened, let alone what is coming next.
There is a wall going up. And if you care about having a sizable, lasting impact on this world, you are in a race against time.
░▒▓█▓▒░
You will hear people argue that it is a long road between large language models and artificial general intelligence. You will hear people argue that pattern recognition and machine learning will never produce a system truly capable of surpassing human intelligence, one capable of genuine discovery and creation. These claims rest on a set of assumptions that deserve closer scrutiny.
Let's start by examining what we actually mean by human discovery.
Human discovery is sequential trial and error over a prolonged period of time. It is the act of picking up from where someone else left off and layering your own thinking on top of theirs. In the truest sense, no idea is entirely original. Ideas are the product of taking everything you know and everything you have experienced, both consciously and subconsciously, and applying creativity to restructure the pieces. Given enough time, you restructure them enough times to land on a configuration that people perceive as valuable.
When people argue that large language models are fundamentally incapable of new discovery, they are making a philosophical claim and presenting it as settled fact. The deeper question is whether discovery requires conscious understanding or whether it can emerge from sufficiently powerful recombination at scale. That question is far from resolved. But here is what we do know: the practical output of these systems is already difficult to distinguish from the kind of work that, a decade ago, would have required deep domain expertise and months of effort.
The equation, in its simplest form, is this: trial and error, executed enough times by systems with expert-level knowledge, equals increased discovery potential. Whether we call that "real" intelligence is a philosophical debate. What it produces is not.
░▒▓█▓▒░
Take the first electronic digital computer as an example. In 1941, Konrad Zuse, a German civil engineer and pioneer of computer science, built the Z3. It could perform one calculation every five seconds. Eighty-five years later, the world's first verified exascale supercomputer performs 1.2 quintillion calculations per second. Meanwhile, the device in your pocket is billions of times more powerful than the Z3 ever was.
You have every right to call me delusional, but I believe we will see that same order of technological leap compressed into the next decade. Just as the world of 1941 could not have fathomed where we are today, we cannot fathom what the next several decades will look like. That is not speculation. That is the pattern.
░▒▓█▓▒░
For as long as enterprise has existed, people have competed to gain an edge, driven by the ambition to deliver more value than their competitors. To build something better. To win, or to watch someone else win in their place.
This reality becomes sharpest when you examine the role urgency has played throughout history.
World War II was a catalyst that propelled humanity's capabilities well beyond what anyone considered possible. The Manhattan Project was handed a blank check and 125,000 people, including many of the world's top scientific minds. No committee. No long-term outlooks. No ROI analysis. Just a will to solve it now, or lose everything.
What came out of that era was not just the atomic bomb. It was radar, sonar, jet propulsion, rocketry, new classes of drugs, and what would eventually become the most powerful force of all: computers. The Manhattan Project produced a generation of scientists and builders who learned to solve impossible problems on impossible timelines. They rejected the very idea of impossible, proving it to be a fickle construct.
Then the Cold War sustained that pressure. DARPA, the internet, GPS, and semiconductors were all born from an urgency to get there first.
The lesson here is not that war drives growth. War simply creates urgency at scale. It is the urgency that drives growth. That distinction is everything.
Fast forward to the late 1990s and the dotcom boom. Entrepreneurs everywhere carried an urgency to build the next Yahoo. Venture capital had billions flowing in, fueled by its own urgency to fund the winners. Even after the bubble burst, venture capital did not slow down for long, and neither did the competitive drive to deliver more value than everyone else.
From 1941 to today, we have built upon one another's ideas, using unprecedented resources, layering unoriginal thought on top of unoriginal thought, to create things that were previously unimaginable. That pattern is important, because it is exactly what large language models are about to do at a speed and scale we have never seen.
░▒▓█▓▒░
Now bring that lens to AI.
Seven years ago, in June of 2018, OpenAI released GPT-1. Almost nobody noticed. It could return information in broken English before becoming incoherent after a few sentences. Today, frontier models like Claude Opus 4.6 are capable of extended reasoning, coding, vision, tool use, agentic workflows, web search, and file creation. In a single prompt, these systems can pass professional exams, write production code, analyze research papers, and build functional applications.
Consider the comparisons.
It takes three years of law school and months of studying to pass the bar exam. LLMs today pass it on the first attempt, with zero preparation, in a fraction of the time. It may take a software engineer six to twelve months to learn a new programming language. LLMs already operate fluently in over a hundred. A 20-page research summary might take a human team days to weeks. It takes an LLM seconds to minutes.
But there are three comparisons that matter even more than those.
We sleep for roughly 26 years of our lives. LLMs can operate around the clock, every day, without rest. It takes a lifetime to master one field. LLMs function with expert-level knowledge across a multitude of fields simultaneously. And when it comes to the rate of improvement, a human professional typically peaks after decades of practice. LLMs have roughly doubled in measurable capability every twelve months since 2020, as tracked across major benchmarks like MMLU, HumanEval, and professional licensing exams.
Let me be direct about what that trajectory means.
It is not long before humans can no longer deliver value that AI cannot deliver better, faster, and cheaper. The exception is peer-to-peer relational work, roles that require the irreplaceable element of being human in relationship with another human. Beyond that, it is not a hunch. It is the trajectory. The train has left the station, and it is moving at such velocity that there is no longer enough friction available to slow it down.
░▒▓█▓▒░
The wall is going up, and this is the moment to put your footprint in the inevitable.
The urgency to create value has never been this high. I do not have a crystal ball, but my base case is that we have roughly ten years left to build before there is nothing left for us to build that AI will not be building on our behalf.
Beyond that horizon, things get speculative. What happens next becomes a product of your belief system, your eschatology, your outlook on life, and your experience. It is defensible to expect some form of universal income. It is equally defensible to believe that humanity will thrive during this period, finding new kinds of meaning and purpose. Others prefer the doom-and-gloom dystopia. None of us know which one it will be.
But here is what we do know right now.
For those who are obsessed with an idea and are on a mission to deliver that idea to the world, the window is open and it is closing. There is an urgency to push beyond what has been considered possible, because it has become as clear as day that impossible is a fickle construct.