January 13, 2026 Anthropic’s president says the artificial general intelligence debate may already be outdated because, in some domains, the milestone has quietly been crossed. In an interview with CNBC, Daniela Amodei argued that AI systems have already reached – and in some cases surpassed — human-level performance in specific tasks, even if they fall short of the broader, sci-fi notion of AGI.
In her words: “AGI is such a funny term because many years ago it was kind of a useful concept to say when will artificial intelligence be as capable as a human. And what’s interesting is by some definitions of that, we’ve already surpassed that.”
Coding, Amodei said, is the clearest example. She pointed to Anthropic’s Claude models, which she said can now write software code at a level comparable to many engineers inside the company itself – a striking claim given Anthropic’s reputation for employing elite technical talent. At the same time, she stressed that Claude remains highly specialized, unable to perform many tasks humans can do. That mismatch, she suggested, exposes a flaw in how AGI is commonly framed.
“The construct itself is now wrong or maybe not wrong, but just outdated,” she said.
Productivity gains are already reshaping work
Anthropic’s internal data backs up the scale of the perceived shift. Engineers at the company now use Claude for roughly 60 per cent of their work, delivering a reported 50 per cent productivity boost. That represents a two-to-three-fold increase in AI-assisted output compared with just a year ago.
The gains are not limited to anecdotal usage. In December 2025, Anthropic announced that Claude Opus 4.5 became the first model to surpass 80 per cent accuracy on SWE-Bench Verified, a demanding benchmark designed to measure real-world software engineering performance. The result placed Claude at the forefront of coding-focused AI systems.
Despite the rapid progress, Amodei cautioned against assuming limitless acceleration. Breakthroughs, she noted, are not guaranteed. “The exponential continues until it doesn’t,” she said, adding that even Anthropic’s own researchers are often surprised by how long the current pace has held.
She also drew a sharp line between technical capability and real-world deployment. Even as AI systems improve, adoption can lag behind due to procurement rules, organizational inertia, and the challenge of redesigning workflows around new tools.
Still, Amodei believes the direction of travel is clear. If AI has already reached human-level performance in narrow but economically critical domains like coding, the question is no longer whether such systems will arrive but how societies, companies and governments adapt to their presence.
