Anthropic insiders expect Nobel-winning AI by 2030, but the public isn't so sure

If you ask frontier AI researchers what work looks like five years from now, you might think they are describing a sci-fi novel. If you ask an average worker on the street, they will probably tell you AI is just a glorified spellchecker.

Anthropic just published a fascinating survey comparing a nationally representative sample of US adults with its own staff. The results show that while both groups mostly agree on what AI can do today, they are living in completely different realities when it comes to where this is all headed.

The inside view vs. the outside world

The gap kicks off with raw scientific capability. When asked when AI will make a Nobel-level scientific discovery, 91% of Anthropic staff expect it to happen by 2030, compared to just 39% of the US public.

The timeline for complex research is even more stark. While roughly a quarter of both groups believe AI can design, run, and publish scientific studies right now, 9 in 10 Anthropic staffers expect AI to manage that by next year. Among the general public, only 38% agree.

Perhaps most telling is the ceiling people place on the tech. Depending on the task, between 19% and 40% of US adults believe AI will never be able to replace human professionals. Inside Anthropic, that figure caps out at just 2%.

Living in the automated future

When the questions turn to personal livelihoods, the divide becomes a chasm.

Anthropic employees expect AI to be capable of handling 75% of their own current work hours by 2030. More than half of them expect that share to hit at least 80%. In contrast, the average US worker expects AI to do just 28% of their job, and only 1 in 20 thinks 80% or more of their work could be automated.

Even non-Anthropic workers in computer science and engineering who use AI daily only forecast 40% of their work being automated, landing much closer to the average cashier than to an Anthropic engineer.

Why such a gap? Because Anthropic staff are already living inside an accelerated workflow. Roughly 59% of staff say tasks suited for AI take a quarter of the time or less, compared to just 26% of AI-using US adults. Inside Anthropic, engineers tripled the lines of code they added in the first month after adopting coding agents, while deal velocity and recruiting throughput saw similar spikes.

Outside the lab, adoption is running into messy real-world hurdles. When asked what holds adoption back, everyday workers point to customers preferring to deal with humans, data privacy risks, unreliable outputs, and legal liability.

Why it matters

We think this survey highlights a massive, dangerous blind spot between the people building tomorrow's infrastructure and the people meant to use it.

For everyday knowledge workers, writing off AI as a neat party trick is becoming an increasingly reckless bet. If the very people engineering Claude genuinely believe three-quarters of their high-skill engineering and research roles will be handled by autonomous systems in four years, the window to learn agentic workflows is narrowing fast.

For the industry, Anthropic's disclosure is a clever flex that doubles as a regulatory warning. But it also exposes a distinct Silicon Valley bias. Anthropic staffers work in pristine digital environments with unlimited compute and deep technical literacy. In the real economy, legacy databases, institutional inertia, and liability concerns slow things down dramatically. The technology might move at Anthropic's pace, but society almost certainly will not.