Something unusual is happening in technology.
The people building some of the world's most important AI systems are no longer describing AI as another software upgrade. They're talking about abundant intelligence, self-improving products, disappearing jobs, AI assistants for everyone, autonomous agents, and AGI arriving within years rather than generations.
What's fascinating is that they don't completely agree on what happens next.
Sam Altman sees intelligence becoming abundant and dramatically cheaper. Spenser Skates believes that intelligence will increasingly live inside products themselves, creating a generation of self-improving products. Dario Amodei warns about serious disruption to entry-level white-collar employment. Jensen Huang believes AI will amplify humans and create entirely new economic opportunities. Satya Nadella says decades of technological change are being compressed into just a few years. Demis Hassabis believes we have reached a pivotal moment in human history.
At first, these sound like different predictions. I think they're different chapters of the same story.
Put them together, and they offer a surprisingly clear picture of what may be coming—and how we should prepare for it.
SAM ALTMAN: INTELLIGENCE IS BECOMING ABUNDANT
OpenAI has increasingly framed AI through the lens of the economics of abundance. The basic idea is simple: as useful intelligence becomes more capable and dramatically cheaper, activities that once required expensive human expertise suddenly become economically possible at enormous scale.
That idea is bigger than it first appears.
For most of human history, useful intelligence has been expensive because it lived inside people. If you wanted expertise, society first had to educate someone for years and then compete for that person's limited time. A great lawyer can serve only so many clients, a programmer can write only so much code, and a scientist can investigate only so many hypotheses.
AI changes those economics because machine intelligence can be copied, scaled, and made continuously available.
Imagine intelligence gradually becoming something closer to electricity. We don't employ electricity or schedule meetings with it. We simply access as much as we need, whenever we need it.
A five-person startup could have analytical and technical capabilities that once required a multinational. A small business could have something resembling its own researcher, programmer, designer, and strategist. Scientists could investigate questions that previously weren't worth the thousands of human hours required to explore them.
We built an economy around intelligence being scarce. AI may turn intelligence into infrastructure.
And when something moves from scarcity toward abundance, economic value usually moves somewhere else.
SPENSER SKATES: PRODUCTS WILL START IMPROVING THEMSELVES
Then Amplitude co-founder and CEO Spenser Skates takes the argument one step further.
What happens when all that abundant intelligence doesn't simply sit inside a chatbot waiting for us to ask questions? What happens when we put it inside the product itself?
Skates has been talking about a future of self-improving products: products capable of continuously understanding how they're being used, discovering problems and opportunities, experimenting with improvements, learning from the results, and feeding those lessons back into what gets built next.
That phrase deserves much more attention.
For decades, software development has operated through a relatively slow human loop. A company builds something, customers use it, data accumulates, analysts study what happened, product managers decide what should change, designers create alternatives, engineers build them, and eventually another version reaches customers.
Then the process starts again.
Every stage depends on humans initiating the next one. AI begins to compress that entire loop.
First, the product gets built faster. AI coding tools dramatically reduce the time required to create software and modify existing products.
Then the product observes. AI agents can continuously watch how customers actually behave instead of waiting for someone to manually investigate dashboards.
Then it understands. Those agents can identify friction, unusual behavior, conversion problems, and opportunities that humans may never have thought to investigate.
Then it experiments. New experiences can be created and tested at a speed traditional product teams couldn't realistically achieve.
Then it learns. Behavioral data reveals whether a change actually improved conversion, engagement, retention, or another important outcome.
And finally, it improves. What the system learns feeds directly back into what should be built next.
Today, humans still control most of those transitions. But imagine what happens as the loop becomes increasingly autonomous.
A product notices that a particular group of customers repeatedly abandons onboarding. An AI agent investigates their behavior, identifies the likely source of friction, and proposes several alternatives. Experiments test those alternatives, behavioral data reveals which one works, and that learning feeds directly into the next improvement.
The product is no longer simply software that people use. It becomes software that learns how to become better software.
That changes the competitive equation.
AI coding is rapidly making it easier for everyone to build. If every company can ship faster, simply producing more features becomes less of an advantage. The advantage moves from how fast you build to how fast you learn.
There is an important distinction here. AI can generate software, but creating something and knowing whether humans actually want it are completely different problems.
The first requires intelligence. The second requires understanding human behavior.
That may be why the next generation of great products won't simply have AI inside them. They'll have a continuous feedback loop connecting intelligence with actual human behavior.
The best product may not win.
The product that learns fastest might.
DARIO AMODEI: WHAT HAPPENS TO THE FIRST RUNG?
Anthropic CEO Dario Amodei sees a much more uncomfortable consequence of increasingly capable intelligence.
He has warned that AI could disrupt a large share of entry-level white-collar employment within just a few years. Whatever the exact number eventually turns out to be, the structural problem beneath his warning deserves serious attention.
What happens to the first rung of the career ladder?
Companies employ junior analysts partly because somebody needs to research competitors, manipulate spreadsheets, and prepare presentations. Law firms employ young lawyers to review enormous amounts of material. Software companies give junior developers relatively straightforward coding tasks.
But those jobs perform two functions. They produce work for the company while simultaneously producing experienced humans.
The analyst eventually develops judgment and becomes a director. The young lawyer reviewing documents eventually handles complex negotiations. The programmer fixing boring bugs gradually learns how real systems behave when things go wrong.
If AI becomes exceptionally good at beginner-level work, companies have an obvious economic incentive to employ fewer beginners.
That creates a paradox we haven't solved: AI could remove the work people traditionally performed while learning how to do the work AI cannot yet perform.
We may automate the bottom of the ladder without figuring out how the next generation reaches the top.
JENSEN HUANG: EVERYBODY GETS SUPERPOWERS
NVIDIA CEO Jensen Huang sees another side of the equation.
His prediction is wonderfully simple: “Everybody will have an AI assistant.”
Huang's broader argument is that AI won't simply substitute for human labor. It will amplify what individuals can accomplish, potentially creating entirely new demand as the cost of intelligence and production falls.
History gives that argument some credibility.
Spreadsheets didn't eliminate accountants. Computers didn't eliminate office workers. The internet destroyed some professions while creating industries that previously couldn't have existed.
AI may therefore eliminate enormous amounts of work without eliminating the same number of jobs.
If software becomes ten times cheaper to create, perhaps society doesn't need one-tenth as many developers. Perhaps humanity decides to build fifty times more software. If scientific investigation becomes dramatically cheaper, perhaps we don't need fewer scientists; perhaps scientists attempt millions of experiments that previously weren't economically feasible.
This means Huang and Amodei could both be right. Some professions may contract dramatically while entirely new categories of economic activity explode.
Technological revolutions rarely choose between destruction and creation. They usually deliver both, but not necessarily to the same people at the same time.
SATYA NADELLA: THE REAL PROBLEM MAY BE SPEED
Microsoft CEO Satya Nadella has described the current platform shift with a sentence I find particularly important:
“Thirty years of change is being compressed into three years!”
The word compressed may matter more than almost anything else in this debate.
Humanity has survived enormous technological transformations before. Electricity transformed factories and homes. Cars redesigned cities. Computers changed offices. The internet reorganized communication, commerce, and media.
But societies had time to adapt. People retired while younger generations learned different skills. Universities changed curricula, governments created regulations, companies formed and failed, and entire professions gradually evolved.
AI may create a different problem because technological adaptation could happen faster than human adaptation.
Imagine someone beginning a four-year university degree today for a profession whose economics could look fundamentally different by graduation. Or consider a 45-year-old professional who spent twenty years becoming exceptional at a task that suddenly becomes available almost instantly and at almost no cost.
The biggest danger may not be technological change itself. It may be the widening gap between the speed at which machines improve and the speed at which humans, companies, and institutions can reinvent themselves.
DEMIS HASSABIS: THIS MAY BECOME BIGGER THAN SOFTWARE
Then we arrive at Google DeepMind CEO Demis Hassabis.
In August 2026, Hassabis described the current period as “a pivotal moment in human history” and said that, after working toward AGI throughout his career, he believes it is now “close at hand.”
This takes the conversation far beyond workplace productivity.
Once AI becomes capable of materially helping humans advance mathematics, biology, chemistry, materials science, engineering, and medicine, AI stops being simply another technology industry. It becomes a technology that accelerates other technologies.
AI can improve robotics, help discover drugs, design new materials, write software, assist scientific research, and help engineers design better computers, which in turn can run better AI.
This creates a feedback loop: better AI helps humans build better technology, better technology helps humans build better AI, and the cycle accelerates.
That's why comparing AI with smartphones may dramatically underestimate what is happening. The smartphone changed how humans accessed information.
Advanced AI could change how quickly humanity discovers new things.
THEY MAY ALL BE RIGHT
At first, these leaders appear to be telling us different stories. Put their arguments in sequence, however, and they begin to look surprisingly compatible.
Altman says intelligence becomes abundant. Capabilities that were once expensive become available to almost everyone.
Skates says products themselves begin learning and improving. Abundant intelligence gets embedded into the product-development loop, connecting building, observing human behavior, experimenting, learning, and improving.
Amodei says abundant intelligence disrupts knowledge work. Particularly vulnerable may be the junior tasks on which traditional careers were built.
Huang says AI dramatically amplifies individual humans. Lower costs create new demand, new companies, and entirely new categories of work.
Nadella says all of this happens extraordinarily quickly. Society gets far less time to adjust than during previous technological revolutions.
Hassabis says AI eventually accelerates discovery itself. Intelligence begins helping humanity invent the next generation of technology.
These aren't necessarily six competing predictions. They may be six layers of the same transformation.
FOLLOW THE SCARCITY
If these people are even partly right, there is a surprisingly practical lesson for anyone thinking about their career, company, or investments:
Don't build your value around something AI is making abundant. Build it around what becomes scarce because AI is abundant.
Attention becomes scarce. When content becomes effectively unlimited, capturing genuine human attention becomes more difficult and therefore more valuable.
Judgment becomes scarce. When analysis can be produced instantly, the difficult question is no longer whether you can obtain an analysis. It's whether you know which analysis deserves to be trusted.
Direction becomes scarce. When AI can generate hundreds of ideas and strategies, another idea isn't necessarily useful. Knowing which problem deserves to be solved becomes the advantage.
Original thought becomes scarce. When competent writing, images, software, and presentations can be produced by almost anyone, having something genuinely different to say becomes more valuable.
Trust becomes scarce. Machines will increasingly communicate convincingly while synthetic content becomes easier to manufacture. Human credibility therefore becomes more valuable, not less.
Taste becomes scarce. AI can create endless possibilities. Someone still has to decide which one is actually good.
And self-improving products create one more scarcity that may matter enormously: knowing what the system should optimize for.
A machine can become extraordinarily good at achieving an objective. A product can become extraordinarily good at improving a metric. But was it the right objective? Was it the right metric?
That is no longer primarily a technology question.
It's a human judgment question.
DON'T TRY TO OUTRUN THE MACHINE
For decades, managers told humans which tasks to perform. Tomorrow, leaders may increasingly define objectives for combinations of humans, agents, and self-improving systems.
Leadership then becomes less about supervising execution and more about defining direction, constraints, incentives, and meaning.
I wouldn't spend the next decade trying to become slightly better at something machines are improving at exponentially. I'd learn to direct them while deliberately strengthening the abilities that become more valuable precisely because intelligence itself is becoming abundant: judgment, curiosity, courage, relationships, leadership, responsibility, and taste.
The winners of the AI era may not be the people who know the most about AI. They may be the people who understand what to do with intelligence once intelligence itself is no longer scarce.
For thousands of years, intelligence gave humans extraordinary power. Now we're building machines that may make intelligence available everywhere, all the time, while the products around us begin observing, learning, and improving.
Perhaps the defining question of the AI era isn't whether machines will eventually think like us.
It's something much more uncomfortable:
What should humans become when being intelligent is no longer enough to make us special?
