Gemini in Google Workspace: Announcements Gallop Forward, Rollout Trots Along

Linguistic localization, not an easy feat with generative AI services?

Two years ago, during Cloud Next ’23, Google introduced in Docs and Gmail a writing-assistance feature (Help me write) powered by Gemini — then still branded as Duet AI. It took until May 2024 for languages other than English to be supported. Spanish and Portuguese were added. French has just joined the list. It appeared on Gmail at the end of March and on Docs in early April.

Within Google Workspace, other Gemini integrations have faced substantial localization delays. For example:

• In the Slides side panel (June 2024 in English; March 2025 in French)
• In Drive’s PDF viewer (July 2024 – March 2025)
• In Meet for note-taking (August 2024 – March 2025)

The delays were shorter for, among others:

• Integration into the Chat side panel (November 2024 – March 2025)
• Creation of enriched documents in Docs (December 2024 – March 2025)
• Integration in the same panel across all suite apps to generate images (November 2024 – January 2025)
• Interaction with Google Calendar from Gmail (November 2024 – April 2025)

Toward an additional automation layer in Google Workspace

The Chat integration will go beyond the side panel in the coming weeks. It will be experimentable in Labs.
In the meantime, organizations using Gemini within Google Workspace and enrolled in the alpha program can try natural-language automation design (“Flows”). It can leverage Gems (custom Gemini versions). Open integration with third-party apps is not yet live.

Should soon follow, in Docs, a writing assistant (“Help me refine”) and a voice-synthesis option (“Listen up”) inspired by what NotebookLM already offers. Additionally, in the Google Vids app, video creation with the Veo 2 model is on the roadmap.

Independent of its integrations within Google’s ecosystem, Veo 2 expands its capabilities with inpainting (filling) and outpainting (extending), composition management and interpolation (choosing the first and last images).
The Chirp 3 audio model likewise welcomes diarization (in preview) and the possibility to create a personalized voice “with 10 seconds of speech” (globally available…on demand).
The Google-made LLM family is growing in parallel, with Lyria, a music generator. It is in preview on Vertex AI.

Agent2Agent: how Google aims to complete the MCP protocol

Vertex AI, of course, is not untouched by the agent wave. Google announces, in particular, that they have added an agent development kit (ADK); currently in Python. Focused on integration with the American group’s ecosystem, it underpins already the Agentspace offering, centered on enterprise search*. Its level of compatibility with other agent frameworks varies. It is fully compatible with LangChain and LangGraph; partial for AG2 and LlamaIndex (provision of managed templates in the SDK) as well as for CrewAI (ability to adapt templates).

Rather than a new product, Agent Engine is a new brand that takes over from Vertex AI Reasoning Engine (or LangChain on Vertex AI). It remains a runtime aimed at deploying agents in production. Google positions it within an AI Agent Builder suite. It also includes ADK, an agent library (Agent Garden, in preview) and various tools (code interpreter, RAG engine, Google Search anchoring, Apigee connectivity…).

To this, an application-layer protocol initiative called A2A (Agent2Agent) is added. Seen as a complement to MCP, it uses the same message format but enables communication between agents. Each would declare its capabilities and authentication requirements via a JSON file in the /.well-known folder.

Major players are already following the approach. Notably, on its roadmap is support for agent registries and dynamic UX negotiation (triggering different conversation modes during task execution).

* Launched in December, Agentspace is in early access. Google now makes it accessible in Chrome’s search bar. It adds a no-code designer intended to complement the “developer” approach of Agent Builder.

To consult in addition:

What is the MCP protocol, rising in the world of agentic AI?
From UX to AX: designing interfaces for AI agents
Gemini integrated into Google Workspace: prices on the rise
With LLMs, can quality and diversity of responses be reconciled?
ChatGPT, Gemini, Perplexity… The race for “in-depth” search