For Japanese companies' AI agent strategies, a core challenge in moving from PoC to production use is designing systems that retain data as an asset while preserving freedom of choice among AI options.
This series organizes the challenges Japanese companies face when using and deploying AI agents from the perspectives of the 4 Cs: Context, Choice, Control and Cost. This article focuses on Choice, or options, and examines how to retain data and context as corporate assets while maintaining the ability to freely choose which AI to use.
AI democratization is discussed in terms of what it means to be able to choose, and although enabling people to use AI without specialized knowledge is often viewed as important, research by GovAI and the UK's CMA points out that democratization encompasses different meanings, including use, development, benefits and governance, and cannot simply be treated as a single concept.
Looking at the configuration of an AI agent as a whole, the central element is an application called a harness. It carries out tasks by receiving requests and interacting iteratively with an LLM, while managing context and memory, the use of tools, and the loading of skills. The LLM offers particularly broad scope for choice, while there is also room to choose how the harness and data are managed.
For production use beyond the PoC stage, this design philosophy focuses on how to retain data and context as assets while ensuring that the model used by the agent and the harness itself can continue to be chosen.