Case studyKokuyo's engineers use Notion agents to review documents and find files
ITmedia Keyman's Net reported on 9 November how Japanese office-products maker Kokuyo uses Notion and Notion AI Notion's built-in AI assistant that can answer questions, write or edit pages, and take actions inside a workspace. in an engineering-led organisation of about 100 people. AI Meeting Notes A Notion tool that records a meeting and writes a transcript and summary. Notion also uses the name Notetaker for a wider version of this feature., requirements, designs and tasks now sit in one place, and AI agent An AI program that can take actions for you, like writing code, filling forms or moving items, not just answering questions. help check documents. The article describes the setup but gives no measured results.
What we learned
- The engineering team, formed in July 2024, first kept minutes, requirement documents and designs across several Cloud computing Renting computers in someone else's data centre over the internet instead of buying your own. services and its own file servers in mixed formats, so people had to ask colleagues where files were. It adopted Notion the same month, citing ease of use for staff new to IT tools.
- A Notion Custom agent In Notion, an AI helper you set up to do a job on its own, often started by an event such as a meeting ending or a new database entry. does a quick first review of documents. The review and its results go into a feedback database; people re-check them, and what they confirm is fed back to the agent as new knowledge for the next review.
- Developers also connect Claude to Notion through Notion MCP Notion's connector that lets AI tools like ChatGPT and Claude read and write your Notion pages, if you allow it., so the AI can fetch Notion information and help with the searching, copying and updating that people used to do across several tools. Next, Kokuyo wants AI to prompt people to fill gaps and to tidy duplicated notes.
- The headline says scattered documents dropped to 'zero', but the article gives no before-and-after figures. Treat it as a description of how the system is set up, not proof of time saved.
Why it matters: The reusable idea is the loop: store an agent's review results, have a person confirm them, and feed confirmed lessons back in. A team can start with one document type and a simple feedback database before trusting an agent with wider edits.
Sources (1)
- コクヨが開発部門の「資料散乱」をAIでゼロに 「Notion」をどう使ったか? · ITmedia Keyman's NetRead ↗