Workspace as the Foundation for AI Agent Development

▶ Watch (2:19)

All AI solutions depend on three pillars: data, actions, and interfaces. Google Workspace provides all three at scale. Drive manages files across billions of users. Docs, Sheets, and Slides enable real-time collaboration. Chat and Meet connect devices. Gmail and Calendar organize communication. These applications are increasingly AI-assisted. Thread summaries in Gmail and the Gemini sidebar are two examples. Workspace serves millions of organizations daily.

Built-in AI Tools: NotebookLM and Workspace Studio

▶ Watch (3:40)

NotebookLM is an AI research assistant grounded in user-provided documents like financial reports and strategy files. It generates overviews, slides, and infographics. Workspace Studio lets users build personal AI workflows with starters and steps. Pierrick demonstrated a workflow that sends a daily news summary to Google Chat using an Ask Gemini step powered by web search. Users can edit and adapt these flows over time.

Custom Agents with ADK and Workspace MCP Tools

▶ Watch (5:24)

Developers can build custom agents using no-code AppSheet, low-code App Script, or any pro-code stack. Workspace APIs and libraries access Gmail messages, create calendar events, and more. The ADK framework registers Workspace MCP tools and defines when agents use them. Google releases managed MCP servers for Cloud and Workspace that are safer than custom API wrappers. Everything is client agnostic. Developers can combine Gemini CLI with Workspace MCPs to bridge local development and organizational data.

Gemini Enterprise Platform for Production Agents

▶ Watch (7:15)

The Gemini enterprise agent platform provides 150+ pre-built models including Gemini models, plus custom tuning and training. Agent designer and runtime tools support collaborative agents with open-source frameworks, sessions, memories, MCPs, and code execution. Governance tools enable continuous observation and evaluation. Search tools support RAG, vector, and AI search. For Workspace, connectors are available for Gmail, Chat, Drive, Calendar, and People. Agents integrate with Workspace UIs through add-ons built with App Script and agent platform APIs.

Current Availability and Future Developments

▶ Watch (11:35)

All tools discussed are available now. Pierrick recommended monitoring four areas: connectors and MCP servers for expanding agent tools; A2A and agent-to-UI protocols for standardized communication; full duplex models like Gemini Live for new interaction types; and skills growing as complements to MCP tools. Two codelabs teach building the travel concierge agent that spans Gmail, Chat, and Calendar. Getting started links are provided.

Notable Quotes

At the end of the day, all AI solutions are still pieces of software. Pierrick Voulet · ▶ Watch (1:57)

I can now focus 100% on the discussion and catch up later if I have a meeting conflict. Pierrick Voulet · ▶ Watch (3:25)

Using Google’s AI tech stack is not a requirement to build custom solutions that rely on these features. Pierrick Voulet · ▶ Watch (6:18)

You are a user, use pre-built agents. You are a builder, build with agent designers. You are a developer, develop agents with frameworks and tools. Pierrick Voulet · ▶ Watch (1:34)

Key Takeaways

  • Workspace APIs and MCP tools let agents access Gmail, Drive, Calendar, and Chat data.
  • Gemini Enterprise provides 150+ models, governance, and RAG search for production agents.
  • Codelabs teach building a travel concierge agent that operates across Gmail and Chat.