Google DeepMind co-founder Demis Hassabis estimates that artificial general intelligence could arrive as early as 2028 or 2029, condensing previous forecasts and intensifying the debate over societal readiness.
Shifting Timelines and the Drive Toward 2028
The timeline for artificial general intelligence has compressed significantly. In December 2025, Demis Hassabis estimated to Axios that AGI remained five to ten years away. By May 2026, he suggested 2030, possibly before. Now, the horizon has narrowed further to 2028 or 2029.
During an interview published by The Wall Street Journal Magazine, the British researcher and Nobel laureate summarized the accelerated outlook with a direct phrase: C’est proche
(It is close). This altered projection is not arbitrary. Hassabis deploys a more urgent tone deliberately to prompt governments, economists, and the general public into treating the prospect with immediate seriousness.
Autonomous agents are driving this renewed confidence. While a traditional chatbot waits for a single prompt, these systems can chain multiple actions together, utilize tools, and take certain decisions during a task. Hassabis described these current frameworks as a dress rehearsal for artificial general intelligence.
Technical Barriers Facing DeepMind
Despite rapid advancements, current systems remain constrained by fundamental bottlenecks. During an April exchange with Y Combinator, Hassabis highlighted three core limitations that prevent today’s models from achieving human-like flexibility and reliability.
Continuous learning remains absent. Human beings absorb an experience and reuse that understanding in other situations. Models do not naturally update their knowledge with every new piece of information.
Memory persistence also poses a structural challenge. Even with expanded context windows, assistants can lose the thread. Reasoning represents the final major hurdle. An AI can solve a complex math problem or produce solid code, then make a mistake on elementary logic, or give a bad answer with aplomb.
What Autonomous Systems Change
The definitive leap from narrow machine learning to AGI centers on operational flexibility. Current utility involves using isolated tools—such as ChatGPT for travel, Gemini for message summarization, or Claude for drafting text. A true general intelligence would execute complex, multi-step workflows across novel domains with much less supervision.
In practical application, an advanced agent could compare prices, coordinate agendas, and book a stay almost alone. Enterprise deployment would shift administrative, research, and analysis tasks to autonomous systems. Scientific discovery offers a proven precedent. Hassabis shared the 2024 Nobel Prize in Chemistry for AlphaFold, which revolutionized protein structure prediction.
Yet these capabilities introduce risks regarding employment, privacy, security, and errors of decision by more autonomous systems. Google DeepMind has published a framework to measure progress toward AGI, but a universal benchmark verifying its arrival does not yet exist.
