Setagaya Ward cuts 約2900時間 with Genspark, but AI checks add burden

Many organizations that bring in generative AI find it does not stick, stays with individual users, or never gets woven into daily work, and local governments face the same pattern. Setagaya Ward had already built its own AI environment on Azure OpenAI Service and adopted Microsoft Copilot, yet staff who were new to the tools were unsure what instructions to give.

To test whether people without deep AI skills could produce usable work from vague instructions, the ward brought in Genspark, which is built around AI agent functions for research and document creation. In the trial it fed the tool other municipalities' plans and published standards and procedures, then checked how far it could support the ward's own research and planning. It also used the AI Document, AI Slide, AI Image Generation and AI Design Generation features to make internal briefing materials, flyers and publicity items.

The trial confirmed three points: whether an individual staff member can use the tool on the job, whether the task can be repeated, and whether the work holds up as a division task when the person in charge changes. Sho Miyama of the DX promotion division said the trial showed staff spend time comparing and organizing materials and adjusting formatting around document creation, and that having AI handle these steps would be significant. At the same time, the burden of checking and correcting AI output has grown, a trade-off the ward will need to address as it moves from individual use toward organization-wide use.