Anthropic CEO maps out humanity’s AI ‘adolescence’

In the movie Contact, when an astronomer is asked what single question she would pose to an alien civilization, her answer is simple: “How did you survive this technological adolescence without destroying yourself?”

Anthropic CEO Dario Amodei can’t seem to shake that question. Following his optimistic vision in Machines of Loving Grace, Amodei published a new essay on January 26, 2026, titled The Adolescence of Technology. His core message: humanity is about to be handed almost unimaginable power, and it remains deeply uncertain whether our institutions are mature enough to wield it.

Stepping away from doomerism

Amodei starts by pushing back hard against quasi-religious “doomerism.” He argues that during the height of AI risk panic across 2023 and 2024, sensationalist voices lacking concrete evidence dominated social discourse, sparking a severe backlash.

By 2025 and 2026, the political pendulum swung all the way toward blind opportunity, sidelining safety discussions entirely. But Amodei points out that technology doesn’t care about public mood swings. In reality, humanity is considerably closer to genuine danger in 2026 than it was three years ago.

To manage this, Amodei calls for surgical interventions. Instead of broad, economically destructive mandates that provoke political backlash, regulators and companies should implement lightweight, fact-based rules while gathering empirical evidence on imminent threats.

A country of geniuses inside a datacenter

What kind of AI is he actually worried about? Amodei reiterates his definition of “powerful AI,” which he believes could arrive in as little as 1 to 2 years.

This isn't just a smarter chatbot. It’s an entity with raw intelligence exceeding Nobel laureates across math, biology, and software engineering. It operates virtual human interfaces like keyboards, mice, and web tools, autonomously running tasks that take days or weeks.

By roughly 2027, compute clusters could support millions of these instances running concurrently at 10x to 100x human speed. Amodei summarizes this setup as a “country of geniuses in a datacenter.” Crucially, he reveals that at Anthropic, some of the strongest engineers have already handed over nearly all their coding to AI, creating a flywheel where the models are directly accelerating the next generation of AI development.

Why it matters

For engineers and builders, the takeaway is stark: frontier AI is no longer an assistant; it is becoming the primary workforce. If Anthropic’s top engineers are offloading almost all coding to internal models, anyone clinging to traditional software workflows risks falling rapidly behind. Learning to direct and verify autonomous cognitive labor is now the defining skill.

Politically, Amodei is attempting a difficult balancing act. He is trying to rescue AI safety from the realm of fringe sci-fi panic and reposition it as sensible, pragmatic governance. By advocating for narrow, burden-light regulation, he aims to avoid stifling competition while still putting guardrails in place before millions of super-fast digital minds come online.

We think this essay reveals a subtle unease at the frontier. When models begin writing the very software used to train their successors, the capability curve stops waiting for human consensus. Whether our systems can survive this technological rite of passage remains an open question—and Amodei is warning us that the clock is ticking faster than most policymakers realize.