Polimill 始于 Surfvote，一个让市民就政治和社会议题交换观点的平台。随着公司与地方政府合作，它发现了一个结构性挑战：公共部门团队被日常运营消耗殆尽，几乎没有时间将市民的声音反映到政策中。
为了扩大公民参与，Polimill 首先需要让政府工作本身更高效。公司开始为公共部门工作流程构建生成式 AI 平台，并于 2024 年 10 月发布了 QommonsAI。如今，QommonsAI 在议会答复、公共服务、社会福利和法律检索等领域提供专业化 AI。日本全国约 1,050 个地方政府和约 55 万名公务员正在使用它。QommonsAI 基于 OpenAI 技术构建，正从员工生产力工具演变为 Polimill 所设想的日本市政工作公共操作系统。
在劳动力短缺日益严重的背景下，利用 AI 提高政府工作效率至关重要。但引入各自独立的工具可能会在地方政府之间造成服务差距。这就是我们希望 QommonsAI 成为平等支持每个地方政府的共同基础，并成长为支撑日本政府的公共操作系统。
——Masahiro Wakabayashi，Polimill CAIO
构建跨地方政府知识库
将生成式 AI 引入政府面临的最大障碍之一是数据碎片化。每个地方政府都有自己的工作流程和文档格式，历史记录分散各处。例如，准备议会答复可能需要人工查阅多年的会议记录，以确保答复与都道府县政策保持一致。如果底层数据没有组织好，即使是最强大的 AI 模型也无法产生实用的洞见。
Polimill 收集并标准化了日本各地的议会会议记录，然后使用 AI 添加元数据，构建了一个跨地方政府、跨时间范围的高精度检索基础。
公司将该结构从议会扩展到福利、法律及其他行政领域。通过将专业数据和 AI 整合到一个平台，并通过统一界面实现可检索性，QommonsAI 将分散的行政信息转化为日常工作中可用的知识。这一结构是 Polimill 将 QommonsAI 打造为政府共享平台愿景的基础。
平衡政府级安全与日常易用性
OpenAI 的 GPT 模型是 QommonsAI 的核心。在公共部门，信息管理和审计就绪性至关重要。QommonsAI 包含运营控制功能，让管理员可以查看功能使用历史，并根据组织政策限制可用模型。
与此同时，Polimill 的 CAIO Masahiro Wakabayashi 指出，GPT 模型广泛的能力和熟悉度是采用它们的主要原因。当公务员在系统上线后需要使用新工具时，ChatGPT 的广泛认知度很重要。即使他们不熟悉技术术语或其他模型名称，知道系统基于广泛使用的 ChatGPT 技术也能降低初始采用门槛。
据 Polimill 称，在 QommonsAI 的通用对话功能中，GPT 模型是日常工作中最常被选用的选项。从读取文件到支持日常对话，GPT 在单一模型中可靠响应各种工作请求的能力，帮助 QommonsAI 在基层扎根。
Codex 与手把手支持使开发速度提升 3-5 倍
OpenAI 技术也改变了 Polimill 的开发流程。公司在整个工作流中采用了 Codex，从需求定义到与现有 GitHub 代码的一致性检查、实现和测试。工程师可以专注于审查 AI 生成的计划并做出高层决策，而 AI 则更自主地承担实现工作。
凭借 AI 编码速度和持续的后台工作，开发速度提升至此前的 3-5 倍。Wakabayashi 表示，构建原型、向地方政府团队展示、收集反馈的验证循环已大幅加速。
OpenAI 的手把手支持进一步加速了这项工作。合作包括分享全球先进案例的最佳实践、提供针对 Polimill 定制的信息，以及帮助设计开发流程本身，包括哪些步骤应由 AI 处理、哪些环节需要人工审查并承担责任。
随着模型和能力的发展，AI 辅助开发实践也在不断变化。单个公司很难跟踪全球先进案例并将其融入自己的工程组织。Wakabayashi 认为，持续向 OpenAI 学习并将这些洞见落实到 Polimill 内部的能力，与 API 本身的性能同样宝贵。
成果一览
以 OpenAI 技术为核心，Polimill 持续将 QommonsAI 扩展为面向市政工作的平台。主要采用和使用成果包括：
- 日本全国约 1,050 个地方政府和约 55 万名公务员正在使用 QommonsAI。
- Polimill 构建了统一日本各地分散的议会会议记录和行政信息的基础设施，并实现跨组织检索。
- Codex 和 OpenAI 的手把手支持加速了验证和实施循环，使开发速度提升 3-5 倍。
- GPT 模型的高级推理能力帮助经验较少的员工利用 AI 和积累的行政信息起草政策提案，其质量接近资深官员的提案水平。
超越效率：捕捉资深官员的隐性知识
QommonsAI 的价值不仅限于更快的检索。在 Polimill 的验证中，经验较少的员工使用 AI 和积累的行政信息起草政策提案，这些提案获得的评价接近经验丰富官员的提案。
与此同时，经验丰富官员的提案仍然获得最高评价。Polimill 分析认为，差异来自隐性知识：手册中没有记载的实践判断，例如使政策提案落地所需的程序或居民可能存在的顾虑。
Polimill 计划记录资深官员如何指示 AI 进行研究以及如何修改输出，将此前未被文档化的判断转化为组织知识。目标不是用 AI 取代熟练员工，而是放大他们的能力并将他们的诀窍传递给下一代。QommonsAI 可以成为 Polimill 为政府设想的这种知识传递的基础。
我认为那些未被知识或数据捕捉的部分对结果质量做出了贡献。换句话说，当拥有这类经验的人使用 AI 时，他们可以创造出更好的工作成果。
——Masahiro Wakabayashi，Polimill CAIO
迈向公共部门工作的超级代理
2026年秋季，Polimill计划全面推出Qommons ONE，这是一个允许外部公司为市政当局提供应用程序的商店。其核心将是一个超级代理，它整合了多个专门的AI系统和私营部门应用。用户将能够陈述一个目标，系统将调用必要的AI或应用程序来产生实用的交付成果，从研究到演示文稿。
为了实现这一愿景，Polimill认为GPT模型的广泛输入和输出能力以及OpenAI积累的代理开发专业知识是重要的基础。OpenAI还通过技术验证和个别示例支持开发。随着模型和开发方法的不断演进，若林认为与OpenAI紧密合作，将最新见解引入Polimill具有重大价值。
通过将这些见解迅速融入开发，Polimill可以持续演进Qommons ONE。其雄心超越了行政效率。AI创造的时间可以让公务员更贴近地倾听公民的声音并支持他们。这个公共操作系统是Polimill正在努力实现的政府未来。
迈向政府、企业和公民通过AI连接的新一代公共基础设施，Polimill的工作才刚刚开始。
Polimill began with Surfvote, a platform where citizens can exchange views about politics and society. As the company worked with local governments, it saw a structural challenge: public-sector teams were consumed by daily operations and had little time to reflect citizens' voices in policy.
To expand civic participation, Polimill first needed to make government work itself more efficient. The company started building a generative AI platform for public-sector workflows and released QommonsAI in October 2024. Today, QommonsAI provides specialized AI across areas such as assembly response, public services, social welfare, and legal search. About 1,050 municipalities and about 550,000 public employees across Japan now use it. Built with OpenAI technology, QommonsAI is evolving beyond a staff productivity tool into what Polimill sees as a public OS for municipal work in Japan.
Amid a worsening labor shortage, using AI to make government work more efficient is essential. But introducing separate tools can create service gaps between municipalities. That is why we want QommonsAI to become a common foundation that supports every municipality equally—and grow into the public OS that supports Japan's government.
Masahiro Wakabayashi, CAIO, Polimill
Building a cross-municipality knowledge base
One of the biggest obstacles to bringing generative AI into government was fragmented data. Each municipality has its own workflows and document formats, and past records are scattered. For example, preparing assembly responses can require manually reviewing years of minutes to ensure an answer is consistent with prefectural policy. Even the most capable AI model cannot produce practical insight if the underlying data is not organized.
Polimill collected and standardized assembly minutes from across Japan, then used AI to add metadata and build a high-precision search foundation that works across municipalities and time periods.
The company expanded this structure beyond assemblies into welfare, laws, and other administrative domains. By bringing specialized data and AI into one platform and making it searchable through a common interface, QommonsAI turns distributed administrative information into knowledge that can be used in daily work. That structure is the foundation for Polimill's vision of QommonsAI as a shared platform for government.
Balancing government-grade security with everyday usability
OpenAI's GPT models are core to QommonsAI. In the public sector, information management and audit readiness are essential. QommonsAI includes operational controls that let administrators review feature usage history and limit which models are available according to organizational policy.
At the same time, Masahiro Wakabayashi, Polimill's CAIO, points to GPT models' broad capability and familiarity as the main reasons for adopting them. The widespread awareness of ChatGPT matters when public employees need to use a new tool after rollout. Even if they are not familiar with technical terms or other model names, knowing the system is based on widely used ChatGPT technology lowers the initial barrier to adoption.
According to Polimill, GPT models are the most frequently selected option for everyday work in QommonsAI's general conversation feature. From reading files to supporting daily dialogue, GPT's ability to respond reliably to a wide range of work requests in a single model helps QommonsAI take root in the field.
Codex and hands-on support made development 3-5x faster
OpenAI technology has also changed Polimill's development process. The company adopted Codex across the workflow, from requirements definition to checking consistency with existing GitHub code, implementation, and testing. Engineers can focus on reviewing AI-generated plans and making high-level decisions, while AI carries out more of the implementation work autonomously.
With AI coding speed and continued background work, development speed rose to 3-5x previous levels. Wakabayashi says the validation cycle of building prototypes, showing them to municipal teams, and gathering feedback has accelerated dramatically.
OpenAI's hands-on support has further accelerated that work. The collaboration includes sharing best practices from advanced global examples, providing information tailored to Polimill, and helping design the development process itself, including which steps AI should handle and where humans should review and take responsibility.
As models and capabilities evolve, AI-assisted development practices also keep changing. It is difficult for individual companies to track advanced global examples and incorporate them into their own engineering organizations. Wakabayashi sees the ability to keep learning from OpenAI and implement those insights inside Polimill as valuable as the API's performance itself.
Results at a glance
With OpenAI technology at its core, Polimill has continued expanding QommonsAI toward a platform for municipal work. Key adoption and usage results include:
- About 1,050 municipalities and about 550,000 public employees across Japan are using QommonsAI.
- Polimill built infrastructure that unifies assembly minutes and administrative information distributed across Japan and makes them searchable across organizations.
- Codex and OpenAI's hands-on support accelerated validation and implementation cycles, increasing development speed by 3-5x.
- GPT models' advanced reasoning helped less-experienced employees use AI and accumulated administrative information to draft policy proposals that approached the quality of proposals from veteran officials.
Beyond efficiency, capturing the tacit knowledge of veteran officials
QommonsAI's value extends beyond faster information search. In Polimill's validation, less-experienced employees used AI and accumulated administrative information to draft policy proposals, and those proposals received evaluations close to proposals from experienced officials.
At the same time, the experienced officials' proposals still received the highest evaluations. Polimill analyzes that the difference came from tacit knowledge: practical judgment that is not written in manuals, such as the procedures required to make a policy proposal real or the concerns residents may have.
Polimill plans to record how veteran officials instruct AI to research and how they revise outputs, turning judgment that has not previously been documented into organizational knowledge. Rather than replacing skilled employees with AI, the goal is to amplify their abilities and pass their know-how to the next generation. QommonsAI can become a foundation for the kind of knowledge transfer Polimill envisions for government.
I think the parts that are not captured in knowledge or data contributed to the quality of the outcome. Put another way, when people with that kind of experience use AI, they can create even better work.
Masahiro Wakabayashi, CAIO, Polimill
Toward a super agent for public-sector work
In fall 2026, Polimill plans a full rollout of Qommons ONE, a store where outside companies can provide applications for municipalities. At its center will be a super agent that brings together multiple specialized AI systems and private-sector apps. Users will be able to state a goal, and the system will call the necessary AI or apps to produce practical deliverables, from research to presentations.
To realize that vision, Polimill sees GPT models' broad input and output capabilities and OpenAI's accumulated agent development know-how as important foundations. OpenAI is also supporting development through technical validation and individual samples. As models and development methods continue to evolve, Wakabayashi sees major value in working closely with OpenAI to bring the latest insight into Polimill.
By quickly bringing those insights into development, Polimill can keep evolving Qommons ONE. Its ambition goes beyond administrative efficiency. The time created by AI can let public servants listen more closely to citizens' voices and support them. That public OS is the future of government Polimill is working toward.
Toward next-generation public infrastructure where governments, companies, and citizens connect through AI, Polimill's work is just beginning.
本文内容采集自官方网站，排版和翻译可能与原页面存在差异。
阅读官方全文