Polimill QommonsAI Is Becoming Japan's Public OS

Polimill QommonsAI now runs roughly 1,050 municipalities and about 550,000 public employees in Japan on OpenAI technology, effectively serving as a public operating system.

It launched in October 2024 with a clear goal: turn fragmented assembly, welfare and legal records into searchable knowledge.

Why infrastructure beats point tools in government

For builders, the lesson is blunt. Infrastructure wins.

Polimill standardized assembly minutes nationwide and added metadata so cross-municipality precedent is searchable in one interface.

That data work matters more than model choice, and it explains why GPT familiarity lowered adoption friction.

GPT is the most selected option for daily chat in QommonsAI, covering file reading and routine drafting in a single model.

Masahiro Wakabayashi, Polimill's CAIO, says Codex lifted development speed three to five times by handling implementation and testing.

Engineers now review AI-generated plans while Codex executes. OpenAI also helped design where humans must stay accountable.

The next test is Qommons ONE in fall 2026, an app store where a super agent orchestrates specialized AIs to produce research and presentations.

If it ships, procurement shifts from buying tools to governing a platform.

What Polimill QommonsAI still has to prove

The announcement claims junior staff nearly matched veteran proposal quality with AI, yet offers no method or sample size.

It touts shared knowledge across 1,050 tenants without detailing isolation for sensitive welfare data, or audit controls beyond usage history.

Capturing veteran tacit knowledge by logging prompts and edits is compelling, but unproven without longitudinal results.

External developers will watch whether Qommons ONE truly opens or stays a curated bundle.

Dependence on GPT and Codex brings breadth and velocity, but ties municipal operations to API pricing and policy set in San Francisco.

Polimill began with Surfvote to amplify citizen voices. Its bet is that efficiency now buys time for listening later.

The metric that matters isn't coverage. It's whether response times and policy quality visibly improve.