Polimill has developed QommonsAI using OpenAI technology and released the platform in October 2024. The generative-AI system is now used by about 1,050 municipalities in Japan and by roughly 550,000 public employees. Polimill positions QommonsAI as more than a productivity tool for staff — the company intends it to evolve into a shared, next-generation public operating system for municipal work.
Origins and problem statement
Polimill started with Surfvote, a platform for citizen discussion on political and social issues. Through work with local governments, the company identified a structural problem: public-sector teams are often consumed by day-to-day operations and have limited time to reflect citizens' voices in policy. To expand civic participation, Polimill first had to make government work itself more efficient.
Data fragmentation and the technical solution
A major obstacle to adopting generative AI in government was fragmented data: each municipality has its own workflows and document formats, and historical records are scattered. For instance, preparing assembly responses can require manually reviewing years of minutes to ensure consistency with prefectural policy. Without organized source data, even the best AI model cannot produce practical insight.
To address this, Polimill collected and standardized assembly minutes from across Japan, then used AI to add metadata and build a high-precision search layer that works across municipalities and time periods. The company expanded this approach to welfare, legal, and other administrative domains, bringing specialized data and AI into a single searchable platform. That structure turns distributed administrative information into usable knowledge for daily municipal tasks.
Core technology and governance controls
OpenAI’s GPT models are central to QommonsAI. Because information management and audit readiness are critical in the public sector, QommonsAI includes operational controls allowing administrators to review feature usage history and restrict which models are available according to organizational policy.
Masahiro Wakabayashi, Polimill’s CAIO, cites not only GPT models’ broad capabilities but also the familiarity of ChatGPT as reasons for adoption. Widespread awareness matters when public employees must start using a new tool: knowing the system is based on commonly known ChatGPT technology lowers the initial barrier to adoption. In QommonsAI’s general conversation feature, GPT models are the most frequently selected option, able to handle a wide range of workplace requests from reading files to supporting everyday dialogue.
Development changes: Codex and OpenAI collaboration
Polimill adopted Codex across the development workflow — from requirements definition and consistency checks with existing GitHub code to implementation and testing. Engineers now focus on reviewing AI-generated plans and making high-level decisions, while AI carries out a larger share of implementation work autonomously.
With AI-assisted coding and ongoing background work, Polimill reports development speed increased to about 3–5× previous levels. The validation cycle — building prototypes, showing them to municipal teams, and gathering feedback — has accelerated considerably. OpenAI’s hands-on support, including sharing advanced global practices and providing company-specific guidance, further sped up development. The cooperation also helped determine which steps should be handled by AI and where human review and responsibility are required.
Practical outcomes and knowledge transfer
Polimill’s validation shows QommonsAI’s value goes beyond faster search. Less-experienced employees used AI and accumulated administrative information to draft policy proposals that scored close to proposals from experienced officials. However, senior officials’ proposals still scored highest; Polimill attributes the gap to tacit knowledge — practical judgment not captured in manuals, such as procedures required to make a proposal actionable or anticipating residents’ concerns.
Polimill plans to record how veteran officials instruct AI and how they edit AI outputs, converting previously undocumented judgment into organizational knowledge. The aim is not to replace skilled employees with AI but to amplify their abilities and transmit know-how to the next generation.
Qommons ONE marketplace and the super agent
Polimill plans a full rollout of Qommons ONE in fall 2026: a marketplace where outside companies can provide applications for municipalities. At its center will be a super agent that orchestrates multiple specialized AI systems and private-sector apps. Users will state a goal, and the system will call the necessary AI or apps to produce practical deliverables, from research to presentations.
Polimill views GPT models’ wide input/output capabilities and OpenAI’s accumulated agent-development expertise as key foundations for that vision. OpenAI is supporting development with technical validation and sample implementations. As models and development methods evolve, Polimill values the ability to keep learning from OpenAI and to integrate those insights quickly into its engineering work.
Broader ambition: time to listen to citizens
Beyond administrative efficiency, the company argues the time saved by AI can allow public servants to spend more time listening to and supporting citizens. Polimill’s work aims toward a next-generation public infrastructure where governments, companies, and citizens connect through AI — a vision that is only beginning to be realized even as QommonsAI is already in use in over a thousand municipalities.
