[Edaily Reporter Han Kwangbeom ] SamsungElectronics(005930)is evolving its Galaxy smartphones into an “Agentic AI Platform” that autonomously understands user context and acts on the user’s behalf. The company plans to enhance its proprietary on-device personalization data engine and introduce a next-generation user experience (UX) where the screen and user interface adapt to the situation, moving beyond a “phone that knows me” to a “phone that acts on my behalf.”
Lee Yoon-soo, Vice President of the AI Platform Team at SamsungElectronics’ MX Business Division, revealed the next-generation development direction for the mobile on-device AI platform during a presentation at the “Samsung AI Forum 2026” session held on the 30th at the Samsung headquarters in Seocho-dong, Seoul. “If AI phones up until now were ‘phones that know me,’ future phones will evolve into ‘phones that act on my behalf,’” he said.
In preparation for the upcoming era of agents, SamsungElectronics presented the following as the core pillars of its mobile on-device Agent AI platform: △ “Agentic UX,” where the user interface is optimized according to the situation; △ “Agent Memory,” which combines new data generated through interactions with the user with existing data platforms; and △ “Agent Orchestration,” a layer that integrates and coordinates the entire agent experience.
◇Building a Proprietary Personalization Data Engine… 100% On-Device Processing
He identified SamsungElectronics’ proprietary on-device data platform, the “Personal Data Engine (PDE),” as the most powerful weapon in this evolution of the Agentic platform. PDE is a system that analyzes data from apps, systems, sensors, and wearables within the phone using small-scale on-device models (sLM) and stores it as structured knowledge in the form of relational databases, knowledge graphs, and vector indexes.
Vice President Lee stated, “The experience of using foreign agent services like ChatGPT or Claude accounts for less than 10% of the total smartphone user experience,” adding "Samsung’s PDE collects user data generated from the remaining 90% or more of the overall phone experience—including watching movies, web browsing, and taking photos—in addition to agent interactions, and extracts meaningful context from it. This is our greatest differentiator as a manufacturer that others do not possess," he emphasized.
Regarding the chronic limitations of on-device AI—namely, limited computational and memory resources—he announced a plan to tackle these head-on through rigorous privacy protection and pipeline optimization.
He continued, “Since protecting users’ personal information is the top priority when processing PDE data related to personalization, we adhere strictly to 100% on-device processing.” He explained, “We continuously monitor the DEVICE’s status and run the analysis module only when resources are available. By pipelining a small model responsible for signal filtering with an on-device LLM responsible for deep analysis and generation, we achieve results equivalent to those of large models.”
◇The Key to Overcoming the Limits of Intelligence…Finding Solutions in the Design of the Human Brain
On this day, SamsungElectronics also shared its engineering design philosophy aimed at maximizing AI operational efficiency alongside the advancement of mobile AI. Hwang In-cheol, Vice President of SamsungElectronics’ MX Business Division, pointed out the enormous power and computational costs hidden behind the explosive growth of intelligence and proposed shifting the focus of AI design from “scaling” to “efficiency.”
He noted, “While training the latest frontier models consumes enormous resources—such as 100,000 NVIDIA GB200 racks (120 kW each)—the human brain efficiently processes multiple streams of information using just 400 kcal per day and 20 W of power, equivalent to a single light bulb,” and "For AI to become sustainable, we must engineer a system that embodies the brain’s principles of efficient information processing," he emphasized.
To this end, SamsungElectronics is applying five core efficiency pillars modeled after the brain’s mechanisms to its products: △ “Right Context,” which accurately extracts only the necessary context; △ “Right Computation,” which reduces predictable redundant computations; △ “Right Model,” which selects only the model and execution location appropriate for the task; △“Right Effort,” which adjusts the amount of thinking based on difficulty and the cost of failure; and △ “Right Learning,” which incorporates new information without forgetting what was previously learned.
Vice President Hwang stated, “Starting with the Galaxy S24—the first AI phone—the series is evolving into an ‘agent-based AI phone’ that will independently understand user needs and perform more tasks by the time of the S26 and Z Fold 8 series in 2026,” adding "We will deliver an optimal experience through a hybrid AI architecture that dynamically routes tasks—using on-device processing to ensure privacy protection and low-latency responses, while leveraging the cloud for complex generation and inference," he added.
