World-renowned AI researcher reveals breakthroughs accelerating AI agent practical applicationFujitsu's Challenge to Revolutionize AI Agent "Collaboration, Memory, and Quality"
Article | 2025-12-15
While 2025 is hailed as the "Year One of AI Agents" with the technology itself drawing significant attention, numerous barriers remain for actual implementation. An MIT report[1] points out that a major failure factor is AI's inability to correctly understand situations, leading to unstable workflows. Gartner[2] similarly warns that complex system design, data management, and security issues complicate adoption. Worse still, research from MongoDB [3] shows that when multiple agents don’t coordinate well, or when memory management and system structure aren’t robust, problems cascade quickly. In fact, their study found these issues derail anything from 40% to 80% of implementations.
Fujitsu has begun pioneering research and development [4] into AI agents that autonomously advance sophisticated tasks while collaborating with humans. Through this R&D, Fujitsu identifies three fundamental technological gaps that must be bridged to address these challenges: "Collaboration" to enable smooth multi-agent coordination, "Memory" to retain information without context loss, and "Quality" to optimize routing and ensure reliable output.
In this article, Dr. Kobashi, Fujitsu Research, introduces cutting-edge research tackling these gaps, featuring interviews with the authors of research papers.
Agent Data Protocol : Enabling “Collaboration” Among AI Agents
For multiple AI agents to work together effectively across diverse tasks, each agent needs strong, well-rounded training. Today, however, the shortage of high-quality supervised fine-tuning data - the kind of data required to teach agents how to collaborate smoothly - remains a major barrier. Without it, improving overall agent performance becomes harder.
The breakthrough addressing this challenge is the Agent Data Protocol[5]. The conceptual diagram below illustrates how it works. First, raw data is collected from a variety of agent datasets. It is then processed using a unified set of Actions and Observations defined by the Agent Data Protocol, and the resulting Trajectory is stored.
By standardizing these diverse datasets and converting them into a form that’s immediately ready for learning, the protocol dramatically reduces the preparation time required before reinforcement learning can begin. A dataset with more than 1.6 million training instances has already been released publicly, allowing anyone to start reinforcement learning with well-trained agents right away.
In this segment, Dr. Kobashi engages in a deep dive conversation with Professor Graham Neubig (Carnegie Mellon University) about the "Agent Data Protocol”.
