The definition of AGI has always been a moving target. Matei Zaharia, who just received one of computing's top honors from the ACM (Association for Computing Machinery), thinks the target has already been crossed - and that the field keeps treating it as a future problem.
Zaharia co-founded Databricks in 2013 and created Apache Spark, the open-source data processing framework that became one of the most widely deployed data infrastructure tools in the world. The ACM recognition validates a career built on real-world systems, not just papers. That background gives his AI commentary weight that pure theorists lack.
The Goalpost Argument
AGI - artificial general intelligence, the theoretical threshold at which AI can perform any intellectual task a human can - has historically been defined to stay just out of reach. Chess mastery wasn't AGI. Writing code wasn't AGI. Zaharia's argument is that the collection of capabilities in current systems - reasoning through novel problems, generating working software, synthesizing research across fields, holding complex multi-turn conversations - would have qualified as general intelligence by any definition from a decade ago. The definition has been moving, not the milestone.
That's not just semantic. If we're already operating in the zone that was historically labeled "AGI territory," then safety frameworks, oversight mechanisms, and policy conversations that have been deferred to some future moment need to be applied to the systems that exist right now.
AI for Research, Not Just Retrieval
Zaharia is now focused on applying AI to scientific research - not tools that retrieve or summarize existing knowledge, but systems that actively participate in hypothesis generation and experimental analysis. Most AI products on the market today help humans work faster. What he's working toward is AI that contributes to generating genuinely new knowledge.
That raises harder questions about verification. When an AI writing tool produces a draft, you can tell quickly whether it's useful. When an AI system contributes to a research finding, determining whether the conclusion is actually correct requires something far more rigorous. Building that verification layer is one of the harder unsolved problems in making research AI genuinely trustworthy.
For daily AI tool users, Zaharia's framing has a straightforward implication: the capabilities that people have been assuming were still a few years off are the same ones available in tools they can sign up for today. The question of how to use them responsibly wasn't ever meant to be a future question.