This is the clickable demo, with a linked page of three signal decks. The layers diagram, the surfaces document and the Univer analysis are not included.
Agent office suite is a conceptual model for an office suite that presumes agents, not people, are the primary users. Apps become surfaces over shared material and intent, and renderings are produced on demand. An analysis of Univer, an existing open-source office suite, finds it is the closest existing substrate but has no intent layer.
Office suites are built around fixed documents that people operate, so agents working in them have to go through files and screens made for humans.
People who use agents to do office work, and the teams who build office software, have this problem. Documents, spreadsheets and slides are fixed files made for people to open and edit, and the intent behind them is not part of the software. Agents have to work through those files and screens.
Additionally, information across files can quickly be out of sync, leading to confusion and misrepresentation by AI tools. It matters because if agents become the main users of office software, the document-first design adds friction, and nothing records what people or agents were trying to do.
Instead of fixed documents, the office suite keeps shared material and intent as the source. Apps become surfaces over that material, and renderings such as documents, spreadsheets and slides are produced on demand. Because renderings come from one shared source, information can't drift out of sync across files.
The model presumes agents are the primary users, and people work through the same surfaces. A design essay and demos show the idea, and an analysis of Univer shows what an existing office suite lacks.
If an office suite keeps shared material and intent as its source, and produces renderings on demand, then agents and people can work from one current source instead of separate files that drift apart. If that holds, the document stops being the unit of work, and office software becomes a set of surfaces over what people and agents are trying to do.