OpenAI's Codex and GPT models have become the backbone of a public-sector AI platform now used across approximately 1,050 Japanese municipalities, serving roughly 550,000 public employees. The deployment, built by Tokyo-based startup Polimill, represents one of the largest known integrations of AI coding agents into government operations, yet it has largely remained out of the spotlight until OpenAI published a customer story on August 31, 2026.

What Is QommonsAI and How Does It Work?

Polimill built QommonsAI, a generative AI platform designed to handle routine public-sector workflows, and launched it in October 2024. The platform now supports tasks ranging from drafting assembly responses to searching legal and welfare regulations and handling general public-service work. Unlike a pilot program or proof of concept, QommonsAI is a live, operating system embedded in the daily administrative work of Japanese local governments.

The platform relies on two distinct OpenAI products working in tandem. On the user-facing side, GPT models power the day-to-day interaction layer, allowing public employees to interact with the system much like they would with ChatGPT. On the development side, Polimill adopted Codex, OpenAI's coding-agent product, across its entire engineering workflow, from defining requirements to implementation and testing.

"With Japan facing a worsening labor shortage, using AI to make government work more efficient is essential," said Masahiro Wakabayashi, Chief AI Officer at Polimill, describing his stated goal for QommonsAI to become "a common foundation that supports every municipality equally" and eventually "the public OS that supports Japan's government."

Masahiro Wakabayashi, Chief AI Officer at Polimill

Why Is This Deployment Significant for AI Adoption?

Japan's public sector is known for strict regulatory requirements and fragmented bureaucracy. Each municipality typically maintains its own workflows and document formats, with historical records scattered across systems that don't communicate with each other. Getting AI tools adopted at scale in that environment is substantially harder than selling to a single enterprise customer. That Polimill has achieved this scale suggests that AI coding agents can work in highly regulated, complex institutional settings.

The deployment also fits into a broader pattern of OpenAI's investment in Japan. The company opened its first Asian office in Tokyo in April 2024, appointing former AWS Japan president Tadao Nagasaki to lead the market. OpenAI has since described local-government AI adoption "from Saitama to Fukuoka" as part of its broader economic vision for the country.

How Codex Accelerated Polimill's Development Speed

Polimill's use of Codex in its engineering workflow demonstrates how AI coding agents can reshape software development at scale. According to OpenAI's account, engineers now spend more time reviewing AI-generated plans and making high-level decisions, while Codex handles more of the implementation work. Combined with what OpenAI describes as "hands-on support" from its team, Polimill reports that its overall development speed increased by a factor of 3 to 5.

This acceleration matters because it allowed a small startup to build and maintain a platform serving over half a million public employees across an entire country's municipal system. Without AI-assisted development, that scale would have required a much larger engineering team.

Steps to Understand How AI Coding Agents Are Entering Government

• Direct Integration: Rather than a top-down government procurement, QommonsAI entered Japanese municipalities through a startup partner, allowing for faster adoption and iteration without requiring national-level approval processes.
• Workflow Automation: The platform automates routine administrative tasks like drafting responses and searching regulations, freeing public employees to focus on higher-level policy work and citizen engagement.
• Governance and Oversight: QommonsAI includes administrative controls that let organizations review usage history and restrict which models are available, addressing concerns about AI transparency in government settings.
• Labor Shortage Response: Japan's demographic challenges make AI-assisted government work a practical necessity rather than a luxury, creating strong institutional incentives for adoption.

What Remains Unclear About the Deployment?

While the scale of QommonsAI's adoption is striking, important details remain undisclosed. Neither OpenAI's customer story nor Polimill's materials name a specific GPT model version, leaving unclear whether QommonsAI runs on GPT-4o, GPT-5-series models, or another variant. Additionally, no named list of participating municipalities has been published, and the figures come from OpenAI and Polimill themselves without independent verification by a government registry or procurement filing.

Financial details about Polimill, including funding raised, revenue, and valuation, are largely undisclosed in public sources. Polimill plans a full rollout of "Qommons ONE," a broader agent and app platform, in fall 2026, which could expand the system's capabilities further.

How Does This Fit Into Broader AI Infrastructure Trends?

The QommonsAI deployment occurs alongside broader infrastructure developments in AI routing and orchestration. NVIDIA recently released Switchyard, a Rust proxy and library that routes and translates LLM (large language model) traffic across OpenAI and Anthropic APIs. This kind of infrastructure allows organizations to use multiple AI models and providers without rewriting their applications, a capability that becomes increasingly important as government systems need to integrate AI from different vendors while maintaining compatibility with existing workflows.

Similarly, Zoho announced major enhancements to Catalyst, its Platform-as-a-Service offering, adding support for both Anthropic's Claude Code and OpenAI's Codex, along with new integrations for agentic AI coding assistants. These developments suggest that the market is moving toward platforms that can orchestrate multiple AI coding agents, allowing organizations to choose their preferred tools while maintaining a unified deployment and governance layer.

The QommonsAI story demonstrates that AI coding agents are no longer confined to startup engineering teams or tech companies. They are now operating at scale inside government institutions, handling real administrative work for millions of people. As more jurisdictions face labor shortages and budget constraints, this pattern is likely to accelerate, making the infrastructure and governance questions around AI agents increasingly urgent for public-sector leaders worldwide.