Eric Schmidt WARNS: "You Have No Idea What's Coming In 2027"

The current moment in AI

• Eric Schmidt believes we are only 10 or 15% into the impacts of AI, with hardware like robots taking longer than software.
• Recursive self-improvement, where systems improve themselves, is not happening yet, but reasoning systems already act as perfect partners for humans, for good and bad.
• Even if development stopped today, which Schmidt says is not stoppable by any government, individual or corporation, these reasoning agents would still have advanced humanity.

The San Francisco consensus

• Schmidt describes a "San Francisco consensus": this is the year of agents, with scaling of agents and reasoning growing enormously.
• Everyone is out of hardware and electricity; he calls it the biggest boom he has seen across three or four cycles in his career.
• Once recursive self-improvement arrives, intelligence could learn faster than humans because we are biologically limited.
• His example: a tech company with a thousand AI researchers could run perhaps a million AI research agents, limited only by electricity, producing a superintelligence moment. The belief in San Francisco is this occurs within 2 to 3 years.
• The channel narrator pushes back, arguing recursive self-improvement is not exponential because technology shows diminishing improvements as it matures.

Claude Opus 4.5 and programming

• Everyone in the Bay Area doing software says the split went from 80/20 to 20/80 with Claude Opus 4.5, because the underlying LLM produces better quality tokens and thinks longer.
• Schmidt, a former programmer, says there is nothing he could do that it cannot do, citing a C compiler written in Rust.
• Top programmers will become more valuable, not less, because these systems need human control.

Education and risks

• Schmidt proposes universities stop everything and design a prompt engineering course for all freshmen starting in September, covering every discipline including liberal arts.
• Key risks he names: eventual jobs impact (already visible in software and customer service), maintaining moral values while racing China, teenagers being harmed by LLMs, and unpredictable effects from orchestrating agents from non-compatible vendors.

Google's legacy and chips

• The Transformer, TPU and protein folding all happened at Google under his tenure; protein folding does in an hour what took a PhD student 4 years, roughly 300 million times more efficient.
• TPU version two's algorithm changes made it the perfect inference engine, positioning Google well; Nvidia managed to build real supercomputers that Intel never could.

China and robotics

• China is a competitor, not an enemy, with money, intelligence, a strong work ethic and dominance of key industries.
• Schmidt argues China will win low-cost robotics because the electric vehicle industry produces the same motors and systems; letting China dominate EVs was an error.
• The narrator adds China built the Tesla Shanghai Gigafactory in 168 days and has dozens of humanoid robot makers producing thousands of robots annually, while Boston Dynamics has worked on Atlas for almost 15 years.