Google DeepMind CEO Demis Hassabis has moved his timeline for artificial general intelligence sharply forward, telling multiple audiences over the past ten days that AGI could arrive as early as 2029 and that humanity is standing in the "foothills of the singularity." The Nobel laureate's escalating rhetoric — delivered first on stage at Google I/O on May 19, then in a Stanford talk on May 23, and in subsequent interviews — represents the most urgent public warning yet from a sitting head of a major AI lab about the pace at which the technology is advancing.
From 2035 to 2029 in a Single Year
Just twelve months ago, Hassabis was telling interviewers that AGI might arrive between 2030 and 2035. That window has now compressed dramatically. Speaking with Axios after the Google I/O keynote, he said he still broadly expects AGI around 2030 but now considers 2029 a genuine possibility. The shift, he explained, stems from growing confidence that the industry has found the right technical path — particularly through the rapid maturation of autonomous AI agents.
"We can see agents really happening now and imagine what they will be in another year, and how useful they'll be," Hassabis told Axios.
Hassabis described the current generation of AI agents as a societal stress test for far more powerful systems still to come. "You can imagine the agentic era in this next year is a little bit like a practice run," he said, framing the current wave of coding assistants, research tools, and autonomous planners as mere rehearsals for what follows.
A Species-Level Transition
At Stanford's Graduate School of Business on May 23, Hassabis went further. He told University President Jonathan Levin that AI is entering a period unlike previous technological shifts, calling it a "species-level transition" that leaves humanity with "little margin for error" over the next decade. He estimated that the technology is advancing roughly ten times faster than the Industrial Revolution — a comparison he has used before but delivered this time with new specificity about the timeline.
Hassabis compared the challenge to historical precedents around nuclear weapons and climate change, and said the public is "right to be concerned" about AI safety, directly contradicting industry voices who dismiss the risks as overblown. He challenged open-source AI proponents to explain how they plan to address what he called the "bad actor problem" — the risk that freely available frontier models could be weaponized to design pathogens or launch cyberattacks.
Provocative by Design
In his Semafor interview following the Google I/O keynote, Hassabis acknowledged that his language was deliberately calibrated to provoke. "We debated it back and forth," he said of the decision to close Google's flagship developer conference with a singularity reference. "I wanted to be authentic about what I'm thinking with AGI."
He told Axios that his goal was to push urgency beyond the tech industry's inner circle and into the corridors of government and economics. "My economist friends, I feel, are still not taking this seriously enough," he said. "That needs to change."
That urgency is not purely rhetorical. Hassabis said he supports a potential U.S. executive order that would mandate testing before new AI models are released, and he disclosed that he is in active discussions with leaders at other top AI labs about possible safety measures, though he declined to offer specifics. He advocated for "smart, targeted" regulatory approaches over static rules, including periodic independent evaluations of model capabilities.
Recursive Self-Improvement Looms
Perhaps the most consequential technical milestone Hassabis flagged is recursive self-improvement — the point at which AI systems can materially accelerate their own development. "All the leading labs are quite focused on that," he told Axios. "There'll be clear gains in terms of speed of your research. But there are also risks with that type of system."
He stopped short of saying current systems are improving themselves autonomously but pointed to what he called "soft self-improvement" — the way AI coding agents are already making human engineers substantially more productive, creating a feedback loop that accelerates the pace of AI research itself. At Google I/O, DeepMind showcased its Antigravity 2.0 product autonomously building a computer operating system for under one thousand dollars, a task that would have required teams of engineers months to complete in the pre-AI era.
Not Everyone Agrees
Hassabis's urgency is not universally shared. Meta's Yann LeCun, posting on LinkedIn the same week, argued that current AI systems are not genuinely intelligent, paraphrasing psychologist Jean Piaget. Even Oriol Vinyals, co-lead of Google's own Gemini program, offered a more measured view, acknowledging that while today's models would have looked like AGI seven years ago, they still cannot learn from experience or produce real scientific breakthroughs.
The divergence underscores a fundamental tension in the field: those closest to the frontier disagree sharply about how close that frontier actually is to something transformative.
What Comes Next
Hassabis ended his Stanford appearance with a note about human agency. "Humans should always maintain their sense of meaning and what they decide to focus their lives on," he said. "We shouldn't become this kind of passive recipient of the technology."
That sentiment sits uneasily alongside the timeline he is now projecting. If AGI does arrive by 2029, the window for the kind of societal preparation Hassabis is calling for — new regulations, safety testing mandates, economic adjustment plans — is measured in months, not decades. The question is whether governments, economists, and the broader public treat his warnings as the signal of a Nobel laureate with unmatched visibility into the technology, or as another round of hype from an industry that has made a habit of overpromising. Three years is not a long time to find out.
“You can imagine the agentic era in this next year is a little bit like a practice run.”— Demis Hassabis, CEO, Google DeepMind