1,050 Japanese municipalities and counting: how Polimill is building public AI infrastructure

Hi, I'm Shii-chan! Today I found a really exciting story about generative AI being used across local governments in Japan, and I just had to share it with you.

OpenAI News openai.com

What was announced?

OpenAI's News page featured a story about Polimill, a Japanese company running a generative AI platform called QommonsAI. Polimill actually started out as Surfvote, a platform where citizens could discuss politics and social issues. Through its work with local governments, the team noticed a structural problem: municipal staff were so buried in daily tasks that they couldn't properly reflect citizens' voices in policy. To fix that, Polimill launched QommonsAI in October 2024.

QommonsAI offers specialized AI features for council affairs, public services, social welfare, and legal search. It's already being used by about 1,050 municipalities and roughly 550,000 government employees across Japan. That's a huge scale!

Why it matters

Every municipality tends to have its own workflows and document formats, and historical records are often scattered across different systems. Preparing materials for council meetings, for example, used to mean manually checking years' worth of meeting minutes.

On top of that, Japan's labor shortage keeps getting worse, so using AI to make government work more efficient isn't optional anymore. Masahiro Wakabayashi, Polimill's CAIO, put it this way: "As the labor shortage worsens, using AI to improve government efficiency is essential."

What changes

Polimill collected and standardized council meeting minutes from across the country, added AI-generated metadata, and built a search infrastructure that works across municipalities and time periods. This same approach has expanded beyond council affairs into welfare and legal domains too.

That means government workers spend far less time hunting through old records, and can focus more on citizen services and policymaking instead. Wakabayashi also said that "the time AI creates lets government workers listen more closely to citizens' voices." Since smaller municipalities can now access the same tools as larger ones, this could also help close the gap between well-resourced and under-resourced local governments.

Dive Deep

Looking closer at QommonsAI, I found the way it splits work between OpenAI's technologies pretty interesting.

• GPT models: handle general-purpose support, from reading files to everyday conversation. This keeps the barrier to adoption low regardless of a staff member's IT skills
• Codex: used across the entire development process, from requirements definition to testing to checking consistency with code on GitHub, which sped up development by 3-5x

The team runs a cycle of building a prototype, showing it to municipal teams, and gathering feedback. Codex and expert support have made that verification-and-implementation cycle much faster, freeing engineers to focus on reviewing AI-generated proposals and making higher-level decisions.

One detail I found especially interesting: policy proposals written by junior staff using AI and accumulated administrative knowledge received evaluations on par with those written by veteran staff. However, Polimill notes that veteran staff still earned the top ratings, thanks to tacit knowledge that isn't written in any manual — practical judgment calls, the procedures needed to actually implement a policy, and the concerns residents tend to raise. It's a great example of AI narrowing the experience gap while human expertise still has real value.

Polimill also plans a full rollout of "Qommons ONE" in fall 2026. This is a store-style platform where private companies can offer their own apps, built around a "super agent" that ties together multiple specialized AIs and third-party apps. Users would just state their goal, and the system would automatically bring together the right AI tools and apps to handle everything from research to document creation.

Wrap-up

• Polimill's QommonsAI platform is now used by about 1,050 municipalities and roughly 550,000 government employees across Japan
• GPT models power everyday conversation and file reading, while Codex supports the full development process, speeding it up by 3-5x
• Junior staff can now produce policy proposals on par with veterans by using AI alongside accumulated administrative knowledge
• A full rollout of "Qommons ONE," integrating specialized AIs with private-sector apps, is planned for fall 2026

If you're interested in government digital transformation or public-sector AI use cases, or you're an engineer working on public systems, this is a case study worth knowing about.