"Memory and context editing features make artificial intelligence (AI) agents smarter and enable continuous learning."
Ben Mann, co-founder of Anthropic, said this at the SK AI Summit 2025 held on the 4th at COEX in Samseong-dong, Seoul. Mann said, "Memory helps AI models store and use information outside the chat window, and the context editing feature tidies up unnecessary information during (AI model) execution to improve efficiency."
Anthropic is a U.S. artificial intelligence (AI) Start - Up founded in 2021 by former OpenAI researchers and is headquartered in San Francisco. Through its Claude large language model series, it competes with OpenAI's ChatGPT and Google's Gemini, and it is also known as a company that received a $4 billion (about 5.5 trillion won) investment from Amazon.
Since 2023, Anthropic has accelerated AI progress by releasing eight major models on a rapid cadence. Mann said, "Each AI model was not a simple upgrade but a meaningful technological leap," adding, "The Claude platform is designed so builders (developers) can keep innovating."
Introducing the latest model Claude Sonnet 4.5 and a new development environment, he said it is "a model with world-class coding and reasoning capabilities." He added, "Sonnet 4.5 outperforms competitor models not only in coding but also in long-term reasoning and complex agent-building capabilities, and it can be used for real-time data analysis and multistep task automation."
Mann also introduced Claude's browser extension and beta features for VS Code. Developers can use Claude directly on the web or in an IDE (integrated development environment) to automate code edits, repository management, and bug fixes. He said, "Not only engineers but also designers and non-majors will be able to build real-time prototypes, dramatically speeding up product development."
Mann cited three core capabilities that AI agents should pursue: "contextual intelligence," "long-term execution," and "genuine collaboration." He said, "They must evolve from mere tools that carry out commands into entities that understand the user's goals and work context and can adjust and supplement themselves." To that end, Anthropic supports real-time integration of AI with existing systems through the Model Context Protocol (MCP), and it is improving work efficiency while cutting expense by 90% by introducing prompt caching.
Mann said, "AI will not replace humans but will be a colleague that amplifies human capabilities," adding, "Going forward, AI will reduce the time spent handling spreadsheets, and people will focus on strategy and creative judgment."