Status
The system is in production and was used at a live shareholder meeting in June 2026. It runs in a browser, in Japanese or English, on cloud infrastructure designed so that meeting data stays within Japan. The executive always has the final word; the system's job is to inform it.
Next, we are preparing to extend the assistant to other high-stakes meetings where executives take unscripted questions, beginning with investor relations sessions.
Shareholder meeting Q&A assistant
Summary
A shareholder sits down after a long, unscripted question. Two seconds later, a plain summary of what was asked is on the executive's screen. Within ten, generative AI has turned the answers the company prepared in advance into a suggested response to the question actually asked.
The Shareholder Meeting Assistant AI takes the moments while the chairperson restates the question and turns to the right executive, and turns them into time to read, weigh and decide how to respond. At a live shareholder meeting in June 2026, executives spoke with the material already shaped in front of them rather than recalled from memory.
The problem
A shareholder meeting is one of the few occasions when executives answer unrehearsed questions in public, on the record. Companies prepare by building a document of hundreds of anticipated questions and answers, which executives then commit to memory. The preparation burden is substantial, and it still cannot be exhaustive: shareholders may raise something it does not cover. When that happens, the executive has only moments before the microphone reaches them to compose an answer that shareholders, analysts and the press will read closely. Preparation rests on individual recall, and the answer rests on whoever is holding the microphone.
Solution
A generative model does what no person could in the seconds available. It reads the question, finds what the prepared material says about it, and writes that into one suggested response rather than returning documents to search. What it removes is the scramble to recall. The executive reads a position the company has already agreed, shaped to the question in front of them, and answers in their own words.
A trained operator stays in control, marking where each question begins and ends. Because the model draws on several prepared entries at once, it also helps with questions no single entry covers, and it flags where the material runs thin. An executive at the microphone cannot afford lag, so several models run in parallel: delay in any one does not hold up what appears on screen.