Gemini Workspace Upgrades: New Features, Access, and What's Unproven

Gemini Workspace Upgrades: New Features, Access, and What's Unproven

Google expanded Gemini availability across qualifying Workspace and Google One AI plans, Android Police reported today. Until this week, the deeper Gemini Workspace upgrades were initially limited to select subscribers and Workspace testing programs. The shift means far more people will encounter Gemini inside Docs, Sheets, Drive, and Gmail without opting in a change in distribution as much as capability.

The beta launched six weeks ago for Google AI Ultra and Pro subscribers and select Workspace customers in the Gemini Alpha program, per Google's own announcement at the time. On the same day as today's wider rollout, Google announced the Gemini Enterprise Agent Platform at Google Cloud Next: infrastructure for building and governing autonomous agents capable of running multi-step business workflows over hours or days, with a partner marketplace spanning Oracle, Salesforce, ServiceNow, Adobe, and Workday, TestingCatalog reported today.

Below: what changed in the core apps, who can actually access it, what the performance evidence shows, and what the enterprise tier signals about Google's direction.

What actually changed in Docs, Sheets, Drive, and the rest

In Docs, Gemini's "Help me create" feature generates first drafts by pulling context from Workspace files, Gmail threads, and chat history, then stays available inside the document for live section-level edits. Two new options, "Match writing style" and "Match doc format," let users standardize voice and structure against a reference document without manual reformatting, per Google's blog.

Sheets gets the most substantial upgrade. A natural-language prompt can now instruct Gemini to build a formatted spreadsheet from scratch, apply conditional formatting scoped to the correct columns, or convert raw data into a dashboard with visualizations, pulling from Gmail, Drive, Chat, and Google Search as needed, according to Google's admin documentation. Google VP Yulie Kwon Kim described the pitch six weeks ago as eliminating "the digging through emails to get a first draft on the page," Computerworld reported. Google also commissioned a 95-participant study comparing manual data entry against its "Fill with Gemini" feature on a 100-cell task, framing the tools around time savings rather than novelty, according to the Google blog.

Drive and Gmail see meaningful changes. Drive now surfaces an AI-generated summary with citations at the top of natural-language search results; "Ask Gemini in Drive" lets users query across documents, emails, calendars, and the web simultaneously, per Google's blog. Gmail's "Help me write" pulls from existing threads and Drive files to draft personalized replies without requiring explicit instructions. Both features carry geographic and language restrictions: Drive's AI Overview launched in the U.S. first, and "Help me write" is currently English-only, per Google's admin documentation.

Meet, Voice, and Slides round out the set. Meet's note-taking feature now generates structured outputs with distinct Decisions and Action Items sections; Google is testing a setting that would automatically activate note-taking for meetings with three or more participants, per Google's admin documentation. Voice can produce a full recording, transcription, and summary after any call. In Slides, Gemini can generate a new slide matching an existing deck's theme from a prompt, pulling from files, emails, and the web, per Google's blog.

One claim in Slides warrants scrutiny. Google's blog describes generating "entire presentations from scratch" as a current capability, per the March announcement. Six weeks ago, however, Computerworld reported Google saying full from-scratch presentation generation would be possible "in the future." That discrepancy hasn't been independently resolved.

The pattern across every app is consistent. Gemini is designed to absorb the preparatory friction that precedes real work: the blank page, the raw table, the inbox archaeology, the post-meeting cleanup. That's where the tools perform best and where errors matter least.

Who gets it now, and what the reliability evidence actually shows

"Out of beta" describes part of this rollout accurately. Not all of it.

The March beta launched specifically for AI Ultra and Pro subscribers and Workspace customers in the Gemini Alpha program, Computerworld reported at the time. Alpha features remain off by default in enterprise environments and must be activated by a Workspace admin, who can enable them for specific users, groups, or organizational units but cannot toggle individual features within the alpha bundle, per Google's admin documentation. Alpha features may not be available in all Workspace editions. Some features are rolling out to Android first, with web and iOS following later this month. Others remain English-only or U.S.-limited.

The practical advice for anyone expecting these features to appear automatically: check your Workspace edition, and if you're in an organization, ask your admin whether Gemini Alpha is enabled for your account.

On performance, the clearest publicly cited independent benchmark covers Sheets. Gemini scored 70.48% on SpreadsheetBench, a public benchmark for complex multi-step spreadsheet tasks, Android Police reported today. Hands-on reliability for everyday formatting sits somewhere between 70 and 80 percent, with error risk dropping substantially for simpler operations. No equivalent published benchmark exists for Gemini's performance in Docs, Slides, Drive, or Meet.

IDC senior research manager Amy Machado put the underlying problem plainly six weeks ago: "Accuracy is a top challenge for genAI, and user trust is the 'final boss,'" Computerworld reported. Google's own blog still notes that "generative AI is experimental," per the March announcement. These tools are most reliable as a starting point, for first drafts, initial formatting, search summaries, and meeting notes, and least reliable as final output that goes out without review.

What the enterprise agent platform reveals about Google's next move

Six weeks ago, Google avoided calling Workspace Gemini an "agent," framing it instead as a "collaborative partner," Computerworld noted. Today, "agentic" is the central term in Google's enterprise announcement. The Gemini Enterprise Agent Platform evolves and consolidates Vertex AI capabilities into Google's broader Gemini Enterprise platform for enterprise agent development, with all Vertex AI services and roadmap updates now flowing through it rather than continuing as a separate standalone service, per TestingCatalog.

The platform supports agents that run autonomously for hours or days inside cloud sandboxes, handling multi-step workflows like financial reconciliations, with optional human-approval checkpoints configurable at defined stages, Computerworld reported today.

Several tools sit on top of that infrastructure, per TestingCatalog and Computerworld:

• Agent Designer: a no-code builder for schedule- or trigger-based agents, operable via natural language or a visual interface

• Projects: a shared memory layer connected to Workspace files, Microsoft 365, and team chats, persisting context across sessions

• Canvas: a co-editing environment for documents and slides inside Gemini Enterprise, with export to Microsoft Office formats

• Inbox: a central monitoring location for tracking active and long-running agents

Agents also support integrations with third-party tools including Asana, Mailchimp, Slack, and Salesforce, per Google's admin documentation.

On governance, each agent is assigned a cryptographic identity that creates an auditable trail tied to defined authorization policies, per TestingCatalog. Agent Registry indexes approved agents for IT teams; Agent Gateway enforces connectivity policies across environments. "This creates a clear, auditable trail for every action an agent takes, mapped back to defined authorization policies," Google Cloud VP Michael Gerstenhaber said, per Computerworld. Those controls are Google's structural answer to the trust problem Machado identified: enterprise infrastructure that operates well below the user layer.

Gemini Enterprise is priced at $30 per user per month for large organizations, $21 for smaller ones, Computerworld reported. Pricing for the Agent Platform itself was not disclosed at launch.

What's verified, what's still pending

The Docs, Sheets, Drive, and Gmail upgrades are available now for qualifying users and testable today. They're most reliable for prep work: drafts, formatting, search summaries, meeting notes. The 70.48% SpreadsheetBench score for Sheets is the only third-party performance datapoint available across any of the Workspace apps. Comparable data for Docs, Drive, Meet, and Slides does not exist yet.

Google's broadest claims remain thin on independent evidence. The discrepancy between what Google's blog says Slides can do today and what Computerworld reported six weeks ago about future capability has not been resolved. Google's blog still characterizes the technology as experimental, per the March announcement.

The enterprise agent tier will roll out over the coming months with Agent Platform pricing still undisclosed, TestingCatalog reported today. Whether user trust catches up to the roadmap, which Machado identified as the hardest problem to solve, is the open question neither today's Workspace rollout nor the enterprise announcement resolves.