Saga City: generative AI in daily municipal work
自治体×生成AIの伴走支援リアル——佐賀市の現場から見る
B · Useful, limited A Saga City DX official describes a year of generative AI use with survey figures and concrete staff-built apps; useful municipal practice, but the numbers come from a self-reported staff survey and the co-presenter is the city's paid DX partner.
Summary based on the full transcript·Original: about 39 minutes to watch
In short
A 39-minute Code for Japan Summit 2026 session (published 22 October 2026). Kenji Fukuda of Saga City's DX Promotion Division, who says he had never touched a CSV file before the posting, describes how the city of about 224,000 people rolled out generative AI from October 2025. Code for Japan, which supports the city as its DX partner, frames the problem: most municipalities stop at question-and-answer chat. Saga has started building workflows and staff-made apps, but has no way to reward the people who deliver them.
Key numbers
- ~70%share of Saga City staff using generative AI (ChatGPT) in a June 2026 survey
- ~77,000 hoursannual time saved, converted from self-reported median savings (7.7万時間)
- 200volunteers at the first all-staff AI training, against about 50 for typical voluntary sessions
- 5 daystime for a staff member with no programming experience to build a migratory-bird viewing app
As reported by the source.
Japanese·Video·opens youtube.com
What it reports
- Structure: a mayor-chaired DX headquarters, one DX promoter per division, Code for Japan as support team. The first all-staff training drew 200 volunteers.
- Support channels: case examples with prompts and cautions, on-demand learning, an internal forum, and an after-hours "AI muscle-training club" of about 11 self-selected staff working on vibe coding, Dify, local LLMs and RAG manuals, who present failures as well as successes. The club was the mayor's idea.
- June 2026 survey: about 70% of staff use generative AI; about 98% of users feel an effect; about 90% report time savings; median-based conversion gives about 77,000 hours a year. Some staff say AI increased their work because they dig deeper.
- Main uses remain chat-level: drafting documents and email, summarising, ideas. Accounts were shared at first; individual accounts are being contracted.
- Workflows: commuting-allowance distance checks automated in n8n with AI anomaly detection (previously measured by hand in two map services); event data extracted from PDFs by Dify into the city super-app.
- A staff member with no programming background built a tide-aware migratory-bird viewing app in five days at the deputy mayor's request, then a disaster-duty schedule app in half a day. A health-department staff member published a cancer-screening finder. All DX division staff now vibe-code and take prototypes to other divisions.
- No evaluation or reward path for staff who deliver results; they are directed to the staff-suggestion scheme. Fukuda names the risk that builders become the office's on-call fixers.
- Outlook: AI raises what residents ask for, which can make staff workloads heavier; Code for Japan argues saved time should go into designing roles for residents.
How they got it
Spoken account by the city official with slides; figures come from an internal staff survey. Transcript is from auto-generated captions with frequent errors in names.
Limits
- Time savings are self-reported and converted from a median; no baseline measurement.
- No figures for the workflow time savings ("substantial" only).
- The DX partner co-hosts the session and is not neutral.
- Auto-caption transcript.
Why it matters for companies
The adoption pattern matches what firms see: wide chat use, high self-reported value, and slow movement into actual workflows. Two points transfer directly. Volunteer demand for training was four times normal, so interest is rarely the constraint. And staff who build useful tools get praise but no formal recognition, which is a gap HR teams should close before those people burn out or leave.
Key passages in translation
Japanese original
職員のですね、利用率としては大体7割ぐらいで
As for the staff usage rate, it is roughly 70%.
Japanese original
成果を出した職員っていうのがなかなかあのフィードバックというのなかなかできてないというのが現状
The current reality is that we are hardly giving any feedback to staff who deliver results.
Japanese original
AIを使うことによって少しあの、より深掘りをすることによって業務が増えたっていうこともありますけども
There are also cases where using AI, by letting people dig deeper, actually increased their workload.
Summary based on the full transcript. Read the original for the full argument and data.Watch the original on youtube.com