Establishing AI Agent Technology Capable of Reliable Business Idea Generation
The AIST Group and Stockmark Inc. are launching the "Stockmark - AIST Co-creation Project for Intellectual Creation Technology using Large Language Models" to develop an "autonomous idea generation AI agent." This agent aims to create business value by producing innovative ideas, moving beyond traditional AI applications. The collaboration leverages Stockmark's AI technology and AIST's research expertise to drive innovation in Japan by establishing AI agent technology that can derive highly feasible business plans.
📋 Article Processing Timeline
- 📰 Published: April 15, 2026 at 22:00
- 🔍 Collected: April 15, 2026 at 13:31
- 🤖 AI Analyzed: April 15, 2026 at 19:36 (6h 5m after Collected)
The AIST Group (National Institute of Advanced Industrial Science and Technology (President & CEO: Kazuhiko Ishimura, hereinafter 'AIST') and AIST Solutions Inc. (President & CEO: Kiyoharu Aisaka) have achieved substantial research results through their previous joint research with Stockmark Inc. (CEO: Toru Hayashi, hereinafter 'Stockmark'), and have accumulated a track record of leading the research field through impactful presentations at prominent academic conferences and international events.
Now, both parties have decided to expand the scale of their collaboration and transition to the 'Co-creation Project,' a new framework newly launched by the AIST Group from FY2025.
We are pleased to announce that we will launch the 'Stockmark - AIST Co-creation Project for Intellectual Creation Technology using Large Language Models' from April 2026 to further strengthen our collaboration.
Evolution from 'Joint Research' to 'Co-creation Project'
In this project, we will focus on research and development aimed at creating business value through an 'autonomous idea generation AI agent' capable of generating new ideas that would typically only be conceived by humans possessing vast and advanced knowledge.
The applications of generative AI are rapidly expanding beyond knowledge exploration and text generation to higher human creative activities such as 'idea generation' and 'intellectual creation.'
We aim to develop new LLMs and create pioneering business case studies by strategically leveraging Stockmark's generative AI technology with AIST's research resources and expertise.
Research and Development of 'Autonomous Idea Generation AI Agent'
The joint research between the AIST Group and Stockmark has moved beyond the initial theme of 'document summarization' and is currently focused on developing an 'AI agent' capable of generating useful ideas for formulating rigorous business plans in real business scenarios.
▶︎Related Information: Research on Autonomous Idea Generation AI Agent for Developing High-Feasibility Business Plans (March 7, 2025)
A key characteristic of our research achievements to date is the utilization of cutting-edge generative AI technologies such as 'multi-AI agent technology,' which derives ideas through discussions among multiple AI agents, based on preconditions like the features of proprietary technologies (seeds) and the background of proposals, rather than free thinking that is not bound by fixed rules or frameworks.
In business decision-making, merely listing novel ideas is insufficient; rigorous evaluation of whether they are 'worth executing' in light of the company's strategy and market environment is essential.
Only when both the 'idea generation agent' responsible for original creation and the 'evaluation agent' that verifies logical consistency and market validity are in place can an AI agent withstand the stringent demands of complex business environments.
Through the new 'Co-creation Project,' the AIST Group and Stockmark aim to develop an 'autonomous idea generation AI agent' that can derive business plans with high feasibility by autonomously repeating the 'idea generation' and 'evaluation' processes, thereby minimizing human refinement.
Track Record of Previous Efforts by the AIST Group and Stockmark
The AIST Group and Stockmark have steadily built achievements towards establishing LLMs capable of deriving business plans.
Beginning with research into document summarization technology, we have established 'QA automatic generation technology,' developed the proprietary LLM 'Stockmark-13b,' and evolved into an 'autonomous idea generation AI agent' leveraging forced brainstorming and multi-AI agent technologies.
**2023: Establishment of 'QA Automatic Generation Technology' Enabling In-depth Information Digging**
In the early stages of joint research, we focused on technology that automatically generates highly accurate questions and answers from business documents.
This achievement was presented at the 'Information Processing Society of Japan Natural Language Processing SIG' and the international conference 'PACLIC 2023,' laying the foundation for AI to not only read text but also deeply understand its content and extract implications.
**2023: Development of 'Stockmark-13b,' Japan's Leading Business-Specific LLM**
In the same year, we jointly developed 'Stockmark-13b,' a proprietary large language model (LLM) specialized for the Japanese business domain, and publicly presented the results at the '30th Annual Meeting of the Association for Natural Language Processing.'
This was a groundbreaking achievement that enabled accurate grasping of the 'Japanese business context,' which had been difficult for general-purpose AI.
**2024: Realization of 'Continuous Learning' to Maintain High-Frequency Knowledge**
Furthermore, to reflect the constantly changing economic landscape in LLMs, we researched knowledge update methods using continuous learning with news data, and the results were accepted by 'PACLIC 2024.'
The characteristic of 'always possessing the latest knowledge' is precisely a crucial core technology in current AI agent development.
**2025: Establishment and Advancement of 'Multi-AI Agent Technology' Through Multi-faceted Discussion**
In 2025, we established 'multi-AI agent technology,' which derives ideas through discussions among multiple AI agents, and the results were accepted by the international conference 'SIGDIAL 2025.'
The framework, where agents with diverse professional personas such as physicists and psychologists engage in mutual critique and revision, eliminates single-perspective bias and enables the formulation of highly rigorous business plans.
Through these processes, the AIST Group and Stockmark have moved beyond the scope of 'information summarization' to reach a more advanced phase of social implementation: 'AI agents' that think autonomously and adapt to the latest business environments.
With the launch of this 'Co-creation Project,' we will consolidate these insights and further accelerate business transformation through AI agents.
Future Outlook
Through the Co-creation Project, we will accelerate research on multi-agent configurations that autonomously repeat idea generation and evaluation, establishing 'AI agent technology' capable of deriving highly feasible business plans tailored to companies' proprietary technologies and market environments.
The cutting-edge knowledge gained in the Co-creation Project will be promptly implemented into Stockmark's product 'SAT,' powerfully supporting strategic planning and new business creation for client companies, starting from their technical seeds.
Leveraging the AIST Group's strong community, we will present academic achievements at major domestic and international conferences.
Furthermore, through workshops and Shared Task initiatives, we will form an open community for AI agent research and lead discussions.
By accelerating the multi-faceted social implementation of 'autonomous idea generation AI agents' across a wide range of industrial fields, the AIST Group and Stockmark will enhance Japan's overall innovation creation speed.
Now, both parties have decided to expand the scale of their collaboration and transition to the 'Co-creation Project,' a new framework newly launched by the AIST Group from FY2025.
We are pleased to announce that we will launch the 'Stockmark - AIST Co-creation Project for Intellectual Creation Technology using Large Language Models' from April 2026 to further strengthen our collaboration.
Evolution from 'Joint Research' to 'Co-creation Project'
In this project, we will focus on research and development aimed at creating business value through an 'autonomous idea generation AI agent' capable of generating new ideas that would typically only be conceived by humans possessing vast and advanced knowledge.
The applications of generative AI are rapidly expanding beyond knowledge exploration and text generation to higher human creative activities such as 'idea generation' and 'intellectual creation.'
We aim to develop new LLMs and create pioneering business case studies by strategically leveraging Stockmark's generative AI technology with AIST's research resources and expertise.
Research and Development of 'Autonomous Idea Generation AI Agent'
The joint research between the AIST Group and Stockmark has moved beyond the initial theme of 'document summarization' and is currently focused on developing an 'AI agent' capable of generating useful ideas for formulating rigorous business plans in real business scenarios.
▶︎Related Information: Research on Autonomous Idea Generation AI Agent for Developing High-Feasibility Business Plans (March 7, 2025)
A key characteristic of our research achievements to date is the utilization of cutting-edge generative AI technologies such as 'multi-AI agent technology,' which derives ideas through discussions among multiple AI agents, based on preconditions like the features of proprietary technologies (seeds) and the background of proposals, rather than free thinking that is not bound by fixed rules or frameworks.
In business decision-making, merely listing novel ideas is insufficient; rigorous evaluation of whether they are 'worth executing' in light of the company's strategy and market environment is essential.
Only when both the 'idea generation agent' responsible for original creation and the 'evaluation agent' that verifies logical consistency and market validity are in place can an AI agent withstand the stringent demands of complex business environments.
Through the new 'Co-creation Project,' the AIST Group and Stockmark aim to develop an 'autonomous idea generation AI agent' that can derive business plans with high feasibility by autonomously repeating the 'idea generation' and 'evaluation' processes, thereby minimizing human refinement.
Track Record of Previous Efforts by the AIST Group and Stockmark
The AIST Group and Stockmark have steadily built achievements towards establishing LLMs capable of deriving business plans.
Beginning with research into document summarization technology, we have established 'QA automatic generation technology,' developed the proprietary LLM 'Stockmark-13b,' and evolved into an 'autonomous idea generation AI agent' leveraging forced brainstorming and multi-AI agent technologies.
**2023: Establishment of 'QA Automatic Generation Technology' Enabling In-depth Information Digging**
In the early stages of joint research, we focused on technology that automatically generates highly accurate questions and answers from business documents.
This achievement was presented at the 'Information Processing Society of Japan Natural Language Processing SIG' and the international conference 'PACLIC 2023,' laying the foundation for AI to not only read text but also deeply understand its content and extract implications.
**2023: Development of 'Stockmark-13b,' Japan's Leading Business-Specific LLM**
In the same year, we jointly developed 'Stockmark-13b,' a proprietary large language model (LLM) specialized for the Japanese business domain, and publicly presented the results at the '30th Annual Meeting of the Association for Natural Language Processing.'
This was a groundbreaking achievement that enabled accurate grasping of the 'Japanese business context,' which had been difficult for general-purpose AI.
**2024: Realization of 'Continuous Learning' to Maintain High-Frequency Knowledge**
Furthermore, to reflect the constantly changing economic landscape in LLMs, we researched knowledge update methods using continuous learning with news data, and the results were accepted by 'PACLIC 2024.'
The characteristic of 'always possessing the latest knowledge' is precisely a crucial core technology in current AI agent development.
**2025: Establishment and Advancement of 'Multi-AI Agent Technology' Through Multi-faceted Discussion**
In 2025, we established 'multi-AI agent technology,' which derives ideas through discussions among multiple AI agents, and the results were accepted by the international conference 'SIGDIAL 2025.'
The framework, where agents with diverse professional personas such as physicists and psychologists engage in mutual critique and revision, eliminates single-perspective bias and enables the formulation of highly rigorous business plans.
Through these processes, the AIST Group and Stockmark have moved beyond the scope of 'information summarization' to reach a more advanced phase of social implementation: 'AI agents' that think autonomously and adapt to the latest business environments.
With the launch of this 'Co-creation Project,' we will consolidate these insights and further accelerate business transformation through AI agents.
Future Outlook
Through the Co-creation Project, we will accelerate research on multi-agent configurations that autonomously repeat idea generation and evaluation, establishing 'AI agent technology' capable of deriving highly feasible business plans tailored to companies' proprietary technologies and market environments.
The cutting-edge knowledge gained in the Co-creation Project will be promptly implemented into Stockmark's product 'SAT,' powerfully supporting strategic planning and new business creation for client companies, starting from their technical seeds.
Leveraging the AIST Group's strong community, we will present academic achievements at major domestic and international conferences.
Furthermore, through workshops and Shared Task initiatives, we will form an open community for AI agent research and lead discussions.
By accelerating the multi-faceted social implementation of 'autonomous idea generation AI agents' across a wide range of industrial fields, the AIST Group and Stockmark will enhance Japan's overall innovation creation speed.