Sheets canvas puts prompt-built, write-back mini-apps into eligible domains on a schedule admins did not set

Google announced Sheets canvas on August 13, a Gemini-powered capability that turns spreadsheet data into custom, interactive read-write applications from natural-language prompts ^[1]. The rollout to Rapid Release domains had already begun on August 10 ^[2], three days before the announcement that explained it ^[16], and the Scheduled Release rollout starts August 31 and can take up to 15 days ^[3], which puts the tail of that wave around September 15 ^[17].

What is being generated is not a chart. Google describes the canvas as a dynamic visualization layer connected directly to the underlying spreadsheet, and says edits inside a canvas, including dragging task cards or adding entries, immediately update the source Sheet, while changes to source data appear in the canvas in real time ^[4]. Users produce these layouts without coding or formulas, then revise layout, design and functionality through further Gemini prompts ^[5]. Android Authority reported that creation runs through the Gemini side panel and that users can request further changes to the generated interface ^[6].

Google's own examples are Kanban boards, dashboards with financial-scenario planning, whiteboards with sticky notes, study trackers, seating charts, budget-analysis tools and regional field-assignment trackers ^[7]. That is a list of internal tooling. Eric Birnbaum, director of product management for Google Sheets, described the capability as transforming rows and columns into interactive dashboards and custom layouts through a simple prompt ^[8].

The governance surface around it is thin. Eligibility runs by plan: Business Standard and Plus, Enterprise Standard and Plus, Google AI Pro and Ultra, Google AI Pro for Education, and AI Expanded Access ^[9]. The feature is web-only at launch and requires an account set to English with Gemini in Sheets enabled ^[10]. That enablement setting is the main documented lever. Google's launch material documents the interaction model and the rollout but publishes no accuracy or failure-rate measurements for generated canvases ^[11], and the material contains no description of a review step, an app inventory, or an audit trail for the interfaces individual users create ^[15].

A canvas inherits the spreadsheet's sharing settings, so collaborators use the same access model as the source Sheet ^[12]. That is continuity, not control. It means the write scope of a generated interface equals whoever already had edit rights on the sheet, including people who have never inspected how the canvas maps its fields. Lets Data Science notes that inheritance does not remove the need to test whether a generated layout exposes the intended data and writes changes to the correct cells ^[13], and that practical value will depend on behaviour under mixed data types, formulas, protected ranges, concurrent edits and changing schemas ^[14].

The consequence for anyone running Sheets as a lightweight operational database is that the prompt-to-app path now creates write paths nobody specified in advance. Lets Data Science recommends validating edit paths, field mappings, permissions and the consequences of write-back before generated interfaces carry consequential workflows ^[13]. Protected ranges and formula-bearing cells are where to start, because those are the cells a generated layout is most likely to treat as ordinary fields, and because a write-back error there propagates to every downstream consumer of the sheet.

Watch the Scheduled Release wave: if your domain has Gemini in Sheets enabled and English-language accounts, treat the second half of September as the point at which canvases exist whether or not anyone asked for them ^[3][10][17]. Watch also for Google publishing anything on failure behaviour, since the launch material does not ^[11].