Artificial general intelligence may be only a few years away, if Demis Hassabis is right. The Google DeepMind chief says AGI could arrive by 2030, while also arguing that the bigger story is not a robot takeover but a technology that is still being underestimated.
Hassabis made the case during a conversation with Axios co-founder Mike Allen at Google I/O, pointing to the growing power of AI agents as evidence that the field is moving faster than many expected. He said he has already been using AI to build small video games in the evening, work that would previously have taken months. That is the kind of example that sounds cute right up until you realize it also says a lot about how software creation is changing.
AGI by 2030
Hassabis said he believes AGI, the point at which machines become roughly as intelligent as humans, could arrive by 2030. That timeline puts him on the bolder end of the AI forecast spectrum, but it is not coming out of nowhere: multiple labs are now racing to ship agents that can plan, code, and act across apps with less hand-holding than the chatbot era demanded.
The Google chief also argued that AI’s impact is still being underestimated, saying it could end up being 100 times more powerful than the industrial revolution. That is the sort of statement executives love because it sounds both visionary and terrifying, but the real point is simpler: if tools keep getting better at doing knowledge work, the economic shock will spread far beyond tech companies.
Why he does not buy the robot apocalypse
Unlike some voices in Silicon Valley, Hassabis does not think machines will seize control of the planet. He acknowledged that AI brings risks, but he expects people to use it for practical goals, especially in science and healthcare, where faster pattern recognition and automation can actually move the needle.
That framing is telling. The loudest AI debate often swings between utopia and extinction, while the more likely near-term outcome is messier: fewer repetitive jobs, faster product cycles, and a lot of organizations scrambling to figure out what work humans should still do better.
The real test is useful autonomy
If AGI does arrive near the end of the decade, the headline will not be whether a machine can ace a benchmark. It will be whether these systems can reliably complete real tasks with less supervision, across more domains, without turning every workflow into an expensive cleanup exercise.
For now, Hassabis is betting that capability gains will keep compounding fast enough to make 2030 sound less like science fiction and more like a schedule. The sharper question is whether the industry can make those systems useful, safe, and boring before they become powerful enough to be annoying in entirely new ways.