AGI, short for artificial general intelligence, means an AI system that can do any thinking task a human can, across any subject, without being specially trained for each one. Today's AI tools are already impressive but they are specialists: a tool built to write contracts cannot also manage your logistics calendar. AGI would remove that boundary entirely.
Demis Hassabis runs Google DeepMind, which is Alphabet's main AI research division and one of the two or three most powerful AI labs on the planet. He also won the 2024 Nobel Prize in Chemistry for using AI to solve a problem in biology that had stumped scientists for fifty years. When he speaks about timelines, it carries weight. At Google's annual developer conference last week, he said we are standing in the "foothills of the singularity." In a follow-up interview with Axios, he put AGI at 2030, with 2029 now on the table as a genuine possibility. Twelve months ago, his stated range was 2030 to 2035.
He is not an outlier. Google co-founder Sergey Brin, who has no reason to hype anything, said he expects AGI just before 2030. Elon Musk has claimed it could already be here. Sam Altman at OpenAI and Dario Amodei at Anthropic are both pointing at the late 2020s. The clustering of these predictions is more significant than any single name. Independent forecasters on Metaculus, a public forecasting platform with nearly 2,000 participants, currently put a 25% chance on AGI arriving by 2029 and a 50% chance by 2033.
Why is Hassabis moving his estimate earlier? He pointed specifically to AI agents: software that does not just answer a question but takes a sequence of actions over time to complete a task. A scheduling agent, a procurement research agent, a customer response agent. These are already in commercial use in 2026. Hassabis called the current wave of agents "a practice run" for far more capable systems. The logic is that as agents prove themselves on real work, the remaining gap to full human-level capability looks shorter than it did two years ago.
Here is where most business operators get confused. AGI arriving in 2029 does not mean your business is unaffected until then. The agents being deployed today, the ones Hassabis is calling a practice run, are already capable of handling significant chunks of white-collar work. Data processing, customer service, financial analysis, procurement research, legal document review: these are all areas where AI tools in 2026 are producing real output at a fraction of the previous cost. The gap between current tools and AGI is real, but it is shrinking, and the tools along the way are already commercially relevant.
Hassabis has a specific test he calls the Einstein test: give an AI model knowledge up to 1901 and ask whether it can independently arrive at what Einstein figured out in 1905. He says today's systems clearly cannot pass it. But he also says he sees no reason they will not be able to in the future. That is a careful, non-hyped way of saying the direction is clear, even if the exact date is not.
One thing worth noting: there is real disagreement outside the labs. A survey of researchers from the American Association for Artificial Intelligence found that 76% of respondents believe scaling up current AI approaches is unlikely to reach AGI on its own. The people building the systems are more optimistic than many of the people studying them from the outside. That gap matters.
For a business operator, the useful framing is this: plan as though capable AI tools will keep getting cheaper and more useful every year for the foreseeable future, because that part is already proven. AGI as a formal milestone is a theoretical line. The practical pressure on how your company is staffed, how your services are priced, and how your competitors operate is not waiting for that line to be crossed.