I'm sitting in three meetings this week with Austrian founders. All three have projects on the table that were the right call eighteen months ago. None of the three have noticed that they're wrong today.
This isn't personal. It's structural. The ground has shifted under their feet, and it's shifting faster than a typical management team processes information. I want to show in this essay why that is, what the people closer to the source are actually saying about it, and what it means concretely for an Austrian founder.
The curve nobody sees
Start with the person who says it most clearly. Dario Amodei, CEO of Anthropic, published an essay in January 2026 called The Adolescence of Technology. It runs about twenty thousand words, and its central argument is uncomfortably simple: public perception of AI swings between "it's hitting a wall" and "everything will change in six months," but behind that noise, he writes, there has been "a smooth, unyielding increase in AI's cognitive capabilities."
"Smooth" is not a reassuring word here. It means no dip, no pause, no saturation. The curve that became visible to the public in 2022 with ChatGPT is still climbing with the same slope. It just runs much higher now.
Amodei also describes the mechanism. "AI is now writing much of the code at Anthropic," he writes, "which is substantially accelerating our own pace of progress. This feedback loop is gathering steam month by month." This is the part most management teams haven't internalized: AI is now building the tools that build the next AI. The curve is building itself. And it's accelerating, not slowing down.
On what that means for work, Amodei is not diplomatic. He predicts that "AI could displace half of all entry-level white-collar jobs in the next 1 to 5 years, even as it accelerates economic growth and scientific progress." A system with the cognitive capacity of an entire country full of expert researchers, which he calls powerful AI, is "as little as 1 to 2 years away" as a realistic possibility. This is not vague futurism. This is the CEO of a company building these systems, staking his reputation on it.
The man who stopped writing code
If you think Amodei is a self-interested marketer, listen to Andrej Karpathy instead. Karpathy was a founding member of OpenAI, then Director of AI at Tesla, and now runs the education startup Eureka Labs. He's one of the best AI engineers alive, and he has no reason to hype anything.
In March 2026, Karpathy did an interview on the No Priors podcast where he mentioned, almost in passing, that since December 2025 he has effectively stopped writing code himself. He "expresses his intent," as he puts it, to AI agents, sixteen hours a day. He no longer programs. He directs.
In a side project called AutoResearch, he handed an agent a model he had hand-tuned for years and let it run overnight. By morning, the agent had found optimizations he had missed with all his expertise. His conclusion is calm: "The skill that's becoming valuable is judgment. What to delegate. How to specify. How to review it fast."
Translated for the Austrian mid-market, that means: the lawyer still drafting his own briefs today is about one step behind where Karpathy was a year ago. He will, like Karpathy now, eventually realize that a reasonably competent agent had already drafted it before he opened his laptop.
2026, 2027, the next decade
Sam Altman, CEO of OpenAI, has written a calendar for the coming years in his essay The Gentle Singularity, and he does it without the usual hedging. "2025 has seen the arrival of agents that can do real cognitive work. Writing computer code will never be the same," he writes, and that claim maps exactly onto what Karpathy describes from daily practice. What 2026 brings, Altman states in the present tense: "systems that can figure out novel insights." These are no longer assistants. They are tools that independently generate new scientific or strategic insights. 2027, Altman writes, may bring robots doing tasks in the real world. Not as a stage demo. As part of supply chains, workshops, nursing homes, restaurants.
Marc Andreessen, the investor behind a significant share of American tech over the last two decades, put the historical scale of this into context on The a16z Show's 2026 Outlook in January 2026. This is, he says, "the biggest technological revolution of my lifetime. Bigger than the internet. Comparable to the microprocessor, the steam engine, electricity." The economic forecast that follows is relevant to every business: the cost of intelligence, meaning the price of having a cognitive task performed, will "drop like a stone." In six to twelve months, Andreessen expects, small models will have the capabilities of today's frontier models. Free or nearly free. Running locally on a laptop. No API fee.
That's the point where the story gets concrete for a founder in Salzburg, Linz, Innsbruck, or Graz.
What this means for Austria
Taking these four voices together, and there's no serious reason not to, many of the decisions being made in Austrian boardrooms right now structurally look like this:
A company hires a new marketing department while an agent pipeline could do the same work in a fraction of the time at a fraction of the cost. Another invests six figures in a CRM setup that requires three full-time people to maintain, when a custom system with AI automation could be built in two to three weeks. A third is building a content team while competitors abroad already produce with text-to-video, text-to-image, and multi-agent workflows. A fourth is buying per-seat software licenses when, by Andreessen's math, the same function will run for free inside any small model six months from now.
None of these projects is badly executed. From the perspective of 2022, all of them are rational. From the perspective of 2026, they're questionable. From the perspective of 2028, they are strategically obsolete. Money, time, and attention are flowing into horse-breeding while the world is already taking driving lessons.
The news trap
Why does almost no one notice? Because the news itself creates the feeling of being informed. Reading one article per week on ChatGPT feels like keeping up. And that routine is exactly the mechanism that hides the real shift.
Between two Thursdays, the frontier labs collectively ship fifteen, twenty, sometimes thirty new capabilities, model versions, and tools. Most of it is noise. Two or three change the industry. And those two or three rarely reach the headlines that show up in Austrian business newsletters. They live in Reddit threads, GitHub commits, research papers, conference talks, Twitter threads from people like Karpathy that are read by three thousand followers and three months later have changed the working habits of an entire profession.
If you aren't watching the curve daily, you underestimate its slope. And if you underestimate the slope, you plan for the world you just came from, not the one your market is headed into.
Three consequences
Let me make this concrete. If the four voices in this essay are right, and there's no serious reason to assume they aren't, three consequences follow for anyone running a company today.
The first is that your current roadmap is probably substantially obsolete. Not entirely. But "partially obsolete" means you're burning budget on directions that won't exist in eighteen months. That money isn't just gone; it's time you'd need to build the company that will still make sense in 2028.
The second is that your competition is no longer the firm next door. Your competition is a young founder with three agents and a laptop, building a product in six months that you'd need a whole department for. They rarely sit in Salzburg. They sit in San Francisco, Lisbon, Dubai, Bangalore. And they come for your market as soon as they translate the language and the local nuances. That translation itself is now a domain where AI is better than most humans doesn't make things easier.
The third is that the only honest strategic move is not "introduce AI." The honest move is the question of what your company would look like if you built it from scratch today, with everything available in 2026. The answer to that question is almost always uncomfortable. And it is almost always eighty percent the right company.
Closing
I know this essay sounds like many others written in the last three years. But the four people I'm quoting no longer sound like they did three years ago. They sound more serious, more specific, more precise. They give specific years instead of vague decades. They cite percentages instead of promises. They describe their own working day, and that working day no longer resembles one you or I had five years ago.
When the CEO of Anthropic writes that half of all entry-level office jobs could be gone in five years or less, that is no longer a forecast. That is a planning assumption.
The world is driving a car. You can keep optimizing your horse. The horse will get more beautiful, faster, healthier. It will be a very nice horse.
But it will still be a horse.
