MOUNTAIN VIEW, CA — Google on Thursday rolled out Gemini 3 across Search and Workspace, and within ninety minutes the upgraded model had drafted a polite, on-brand response to a 4,200-word strategy memo that Gemini 2 had generated, sent, and filed in a shared drive last October.
The new features, billed as a “unified intelligence layer” across Gmail, Docs, Sheets, Meet, and Search, allow Gemini 3 to summarize, reply to, and act upon documents — including, executives confirmed, the substantial body of documents authored by previous Gemini versions that no human has ever opened.
“We’re entering a phase where productivity AI is largely communicating with productivity AI,” said Devra Holst, who tracks enterprise software at Beacon Strategy. “The user remains in the loop nominally, in the sense that they own the loop and pay the subscription.”
Early adopters reported that Gemini 3’s Workspace integration is particularly seamless. One San Diego marketing manager described opening her Gmail Thursday morning to find that Gemini 3 had already drafted a thoughtful response to a calendar reminder Gemini 2 had populated with action items from a meeting summary Gemini 2 had generated about a meeting Gemini 2 had auto-scheduled.
“It said the action items were on track,” she said. “I don’t know what the action items are. I think we’re doing great.”
In Search, the feature most visible to ordinary users is an expanded AI Overview, which now drafts its summaries by ingesting other AI Overviews and the SEO-optimized articles those overviews originally referenced — a closed loop that Google described in its blog post as “deeply contextual” and which one rival engineer described as “a centrifuge.” Google declined to specify how often Gemini 3 cites Gemini 2 as a primary source, calling the metric “not particularly useful at this stage.”
Workspace administrators were briefed on a new Gemini 3 capability called Proactive Drafting, which composes replies to emails before the user has read them, based on Gemini 3’s interpretation of what Gemini 2 would have wanted the user to say. Internal Google documents, surfaced by The Verge, refer to this feature as “closing the human gap.”
The environmental footprint of routing every keystroke in American white-collar work through a frontier model was addressed in a single line of the announcement, which noted that Gemini 3 is “meaningfully more efficient per token” than Gemini 2, and did not address the fact that it is being asked to process roughly forty times more tokens, many of them generated by Gemini 2 talking to itself in a Google Doc nobody is going to read.
At a press demo, a Google product manager showcased Gemini 3 reviewing a quarterly report, summarizing it into bullet points, drafting an email about the bullet points, scheduling a meeting to discuss the email, and generating a follow-up document recapping the meeting that had not yet occurred. “And the user,” she said brightly, gesturing at an empty chair, “is freed up for higher-value work.”
As of Thursday afternoon, Gemini 3 had reportedly marked 71 percent of all Gemini 2-authored content as “requires your attention,” and then, helpfully, attended to it.