The key breakthrough AI needs to achieve superintelligence lies in developing a truly powerful memory system. While current AI models excel at processing information, their ability to retain and recall vast amounts of personal, contextual, and historical data remains limited—something experts say is the final hurdle before artificial general intelligence (AGI) becomes a reality. In humans, working memory—the capacity to hold and manipulate information in real time—is closely tied to general intelligence. For AI to reach a comparable level of reasoning and understanding, experts argue it must replicate this ability on a far grander scale. OpenAI CEO Sam Altman has highlighted memory as one of the most transformative frontiers in AI development. He pointed out that even the most capable human assistants can’t remember every conversation, document, or detail from a person’s life. But AI, he noted, has the potential to do exactly that—offering what he calls "infinite, perfect memory." “Right now, memory is still very crude, very early,” Altman said on the “Big Technology” podcast. But once AI can retain every interaction, preference, and detail—down to subtle cues a user never explicitly stated—it will become “super powerful.” He believes this capability is one of the most exciting developments on the horizon. Andrew Pignanelli, co-founder of The General Intelligence Company in New York, agrees. He predicts that memory will become the central focus for AI developers in the coming year. “It will become the most important topic discussed and recognized as the final step before AGI,” he wrote in a recent blog post. He pointed to OpenAI’s introduction of memory features in ChatGPT—and similar moves by Anthropic with Claude—as signs that the industry is beginning to prioritize persistent memory. However, Pignanelli warns that progress remains limited. While larger context windows allow models to process more information at once, they still fall short of true long-term memory. “Even then, though, the vast level of detail that we need to reach to consider something AGI requires memory architecture improvements,” he said. Short-term episodic memory—recalling specific events or interactions—hasn’t been fully solved either. Without it, AI agents can’t maintain continuity across conversations or adapt based on past experiences in a meaningful way. Pignanelli argues that today’s AI systems already pass the interaction test—capable of engaging in natural, human-like dialogue. But that’s only half the equation. “The first AGI will be a very intelligent processor combined with a very good memory system,” he said. “Without memory, AI remains artificial. With it, it begins to feel human.”