The shock is coming whether we're ready or not.
Dario Amodei, fundador da Anthropic, avisa que a inteligência artificial representa uma rutura sem precedentes no mercado de trabalho — não apenas mais uma transição tecnológica, mas uma transformação que ameaça a própria utilidade cognitiva humana. Ao contrário das revoluções agrícola e industrial, que deslocaram trabalhadores mas abriram novas portas, a IA avança a uma velocidade e amplitude que podem não deixar portas alternativas. O crescimento económico que ela promete pode coexistir com o empobrecimento da maioria — e é essa contradição que Amodei considera o verdadeiro teste da nossa época.
- A IA já escreve programas inteiros onde há dois anos mal conseguia uma linha de código — a velocidade desta transformação não tem paralelo na história tecnológica.
- Metade de todos os empregos de escritório pode desaparecer em cinco anos, segundo previsões que Amodei recusa classificar como alarmismo.
- Ao contrário das revoluções anteriores, a IA não substitui tarefas específicas — substitui o trabalho humano em geral, ameaçando quem não tem para onde reconverter as suas capacidades cognitivas.
- Startups construídas de raiz com IA vão competir e potencialmente eliminar grandes empresas, acelerando a adoção muito além do que os mercados de trabalho conseguem absorver.
- O poder económico concentra-se no Vale do Silício, aprofundando desigualdades geográficas e criando o risco de uma subclasse permanente de desempregados estruturais.
Dario Amodei, fundador da Anthropic — a empresa por detrás do sistema de IA Claude —, publicou um ensaio no seu blogue pessoal que contraria o otimismo habitual da indústria tecnológica. Sim, a IA vai provavelmente acelerar o crescimento económico, com taxas anuais de PIB entre 10 e 20 por cento a parecerem plausíveis. Mas crescimento e prosperidade partilhada não são a mesma coisa.
O argumento central de Amodei é que esta revolução tecnológica é fundamentalmente diferente das anteriores. Quando a mecanização agrícola deslocou trabalhadores rurais, estes tornaram-se operários. Quando a indústria entrou em declínio, surgiram os empregos do conhecimento. Cada transição foi dolorosa, mas cada uma abriu novas possibilidades. A IA ameaça fechar essas portas: não substitui uma profissão específica, substitui o trabalho cognitivo humano em geral. Quem tiver menor capacidade de adaptação intelectual não encontrará um setor alternativo — porque esse setor também preferirá máquinas.
A velocidade agrava tudo. Há dois anos, os sistemas de IA mal escreviam uma linha de código. Hoje já produzem programas completos. Em breve, poderão executar todas as funções de um engenheiro de software. Os mercados de trabalho são lentos a reequilibrar-se; a tecnologia não espera. Amodei prevê que metade dos empregos de escritório possa desaparecer em cinco anos, acelerada por startups que, ao contrário das grandes empresas, são construídas desde o início para operar com IA.
O segundo grande receio é a concentração de poder. Os benefícios da IA não serão distribuídos de forma equitativa. O Vale do Silício funcionará como uma economia à parte, a uma velocidade diferente do resto do mundo. A riqueza gerada concentrar-se-á em quem controla a tecnologia, enquanto milhões de trabalhadores noutras regiões enfrentarão o desemprego sem perspetivas de recuperação. O choque que Amodei antecipa não é uma transição gradual — é algo mais rápido, mais amplo e muito mais difícil de sobreviver.
Dario Amodei, the founder of Anthropic, has spent considerable time thinking about what artificial intelligence will do to the people who work for a living. In a recent essay shared on his personal blog, he laid out a vision of the coming years that is far darker than the usual tech-industry optimism about productivity gains and economic growth. He believes we are entering what he calls a turbulent and inevitable rite of passage—one that will test who we are as a species.
Amodei and his sister Daniela started Anthropic in 2021 after leaving OpenAI, and the company has built Claude, a conversational AI system that competes directly with ChatGPT. But unlike many of his peers in the industry, Amodei does not seem entirely convinced by the promise of technological progress. Yes, he acknowledges, artificial intelligence will almost certainly accelerate economic growth. He estimates that sustained annual GDP growth of 10 to 20 percent is plausible. New systems for scientific research, manufacturing, supply chains, and financial efficiency will all move faster. The economy will expand.
But expansion is not the same as shared prosperity. Amodei's central worry is this: what happens to the majority of humans in a world where machines can do most of what humans do? He points out that previous technological revolutions—the mechanization of agriculture, the rise of industrial manufacturing—did displace workers, but those workers eventually found new kinds of work. Farmers became factory workers. Factory workers became knowledge workers. Each transition was painful, but each one opened new doors. Amodei fears that artificial intelligence will be different. It will be faster, broader in scope, and far more difficult to adapt to.
Consider the speed alone. Two years ago, AI systems could barely write a single line of code. Now they can write entire programs for some people, including engineers at Anthropic itself. Soon they will be able to perform every task a software engineer performs, from start to finish. This acceleration is unlike anything in previous technological history. Humans and labor markets are slow to react and rebalance. The transition, Amodei writes, will be "exceptionally painful" in the short term.
But speed is only part of the problem. The deeper issue is scope. When machines replaced farm labor, they did not replace the cognitive abilities that made humans valuable in factories and offices. There were new jobs waiting. Artificial intelligence, by contrast, is approaching something closer to general human cognition. It is not replacing specific skills or specific professions. It is replacing human work itself. This means that unlike previous disruptions, which affected people in particular occupations, AI threatens to affect people based on their intrinsic cognitive abilities. Those with lower intellectual capacity will have nowhere to go. They cannot simply retrain for a new sector if the new sector also prefers machines. Amodei fears the creation of what he calls a "subclass" of permanently unemployed or severely underemployed people.
He also worries about the speed of adoption. Many economists argue that the diffusion of AI through the economy will be slow enough to allow for gradual adjustment. Amodei is skeptical. He points to a 2025 prediction that AI could replace half of all office jobs within five years. He does not dismiss this as alarmism. Startups, he notes, will serve as accelerators of adoption. If established companies move slowly, new companies built from the ground up to use AI will simply outcompete them. In some cases, startups might replace large enterprises entirely, doing the same work with a fraction of the workforce.
This brings him to his second major concern: the concentration of economic power. The benefits of AI will not be evenly distributed. Silicon Valley, already the center of technological innovation, will become something more like its own economy, operating at a different speed than the rest of the world and leaving it behind. Geographic inequality will deepen. The wealth generated by artificial intelligence will concentrate in the hands of those who control the technology. Meanwhile, millions of workers in other regions and sectors will face displacement without the prospect of new opportunities.
Amodei is not arguing that economic growth will not happen. He is arguing that growth and widespread prosperity are not the same thing. A society can become richer while most of its people become poorer. That is the shock he sees coming—not a gradual transition like the ones before, but something faster, broader, and far more difficult to survive.
Notable Quotes
We are entering a turbulent and inevitable rite of passage that will test who we are as a species.— Dario Amodei, Anthropic founder
The transition will be exceptionally painful in the short term because humans and labor markets are slow to react and rebalance.— Dario Amodei
The Hearth Conversation Another angle on the story
You're saying AI is different from the steam engine or the assembly line. What makes you so sure?
The steam engine replaced muscle. The assembly line replaced repetitive motion. Both left room for human thinking. AI is starting to replace thinking itself. It's not specialized—it's general. When a machine can do what your brain does, there's nowhere left to migrate.
But people adapted before. Why not now?
Speed and breadth. Agricultural workers had generations to move into factories. Factory workers had decades to move into offices. AI is doing this in years. And it's not just one sector—it's potentially all of them at once.
You mention a "subclass." That's a stark word.
It is. If AI can do cognitive work better than humans, what happens to someone with lower intellectual capacity? They can't retrain into something harder. There's no ladder up.
Silicon Valley becoming its own economy—what does that actually mean?
It means the wealth and opportunity concentrate there while everywhere else falls further behind. Startups replace big companies with fewer workers. The gap between the haves and have-nots becomes geographic, not just individual.
Is there a way to prevent this?
That's the question Amodei is asking, not answering. He's warning that we need to think about it now, before the disruption hits. The shock is coming whether we're ready or not.