Start here · Agents Pro ~9 min read Free survey

Gemini agents: Google, at full strength.

Google's AI agent story used to be scattered — Project Mariner here, Gems there, Deep Research separately, Workspace integrations in a different category. In May 2026 Google consolidated. Project Mariner was retired and absorbed into Gemini Agent. Now there's one umbrella agent + Gems for custom + Deep Research for analyst work + NotebookLM for source-grounded research + the Workspace integrations across Docs/Sheets/Gmail. This free lesson is the map.

The mental model

Google's bet: AI inside your Workspace, not separately.

Google's competitive position is the Workspace — billions of Docs, Sheets, Gmail accounts. Their agent strategy reflects that: instead of just chat, AI lives where you already work.

Gemini Agent is the umbrella autonomous mode. Gems are your custom agents. Deep Research handles long-running analyst work. NotebookLM is the grounded-research workspace. And the Workspace integrations in Docs, Sheets, and Gmail give in-app AI everywhere.

The advantage: zero context-switching. AI lives where your work lives. The cost: figuring out which surface to use when. This lesson sorts that out.

Workflow 01 Gemini Agent

1

Gemini Agent — the unified autonomous mode

Gemini Agent absorbed Project Mariner in May 2026. It handles multi-step tasks across the web and Workspace, deep research, and Canvas in one unified interface.

The prompt that works

What Gemini Agent doesA real Gemini Agent task: "Find 3 new restaurants near our office that serve a private dining room for 12-15 people, are open for dinner Tuesday-Thursday, have a 4.3+ Google rating with 100+ reviews, and aren't on our recently-used list. Verify availability for next Tuesday at 7pm at each. Don't book — just produce a comparison table." The agent works across Google Maps, OpenTable, restaurant websites, and your Drive (for the recently-used list). Returns a finished comparison.

Best use cases

  • Multi-site research with verification
  • Travel and event planning
  • Workspace-spanning project setup
  • Long-running background tasks
Time savings: Multi-site research + action: hours → 10-20 min.

Workflow 02 Gems

2

Gems — your custom Gemini agents

Gems are Google's answer to Custom GPTs — but with native Workspace access baked in. Build agents that know your Drive, your Gmail, your calendar.

The prompt that works

What Gems enableA Gem example: an "Email triage Gem" with your voice. Configured to check your calendar before suggesting meeting times. Connected to your Drive so it can pull project context automatically. You don't paste context anymore; the Gem already knows.

Best use cases

  • Personal assistants with Workspace context
  • Team-shared agents for consistent voice
  • Onboarding helpers for new hires
  • Custom-built agents that replace single-purpose tools
Time savings: A well-built Gem: 5 min/use × 20 uses/week = 100 min/week saved.

Workflow 03 Deep Research + NotebookLM

3

Deep Research and NotebookLM — analyst work, different modes

Deep Research is the long-running analyst agent. NotebookLM is the grounded research workspace where AI only references the sources you upload — useful for high-stakes research where source provenance matters.

The prompt that works

Analyst-grade researchDeep Research example: "Compare the top AI agent observability tools (LangSmith, Arize, Helicone) and 3-5 newer entrants. For each: what they do, funding, pricing, customers, recent product moves. 2,000-word report with citations. Show me the research plan before running." NotebookLM example: upload 12 academic papers on a topic. Then ask "what do these sources agree on" and "where do they disagree" — getting answers strictly grounded in the uploaded material, with quote-level citations.

Best use cases

  • Market research and competitive analysis (Deep Research)
  • Literature reviews from a curated source set (NotebookLM)
  • Multi-document synthesis with source-grounding
  • Audio briefings from research notebooks (NotebookLM's killer feature)
Time savings: Analyst-quality research: days → 30 min.

Workflow 04 Workspace agents

4

Gemini in Docs, Sheets, and Gmail — AI where you already work

The in-app Workspace integrations are the most-used Gemini surface — and the most uneven. Some are real time-savers; others are 'help me write' buttons that don't change what's possible. The Pro lessons separate signal from noise.

The prompt that works

In-app patternsUseful Workspace patterns: • Docs: "Summarize this 20-page doc in 5 sentences" — works well • Docs: "Find inconsistencies in this doc" — catches contradictions you missed • Sheets: "Help me build a formula that does X" — including complex XLOOKUP • Gmail: "Summarize this email thread" — long-thread compression • Gmail: "Find emails about X" — semantic search that beats keyword search Less useful: • "Help me write" from blank — vague output • Auto-categorizing email — too many false positives

Best use cases

  • Summarizing long Docs for meetings
  • Excel formula construction from natural language
  • Gmail thread summarization
  • Semantic email search
Time savings: Email triage and document summarization: 20-30 min/day saved.
Mariner is dead, long live Gemini Agent

If you were using Project Mariner — Google's browser-agent prototype — it was officially retired on May 4, 2026. Its technology is now part of Gemini Agent. Same capabilities, more polished experience, integrated into the broader Gemini app.

Want to learn each Gemini agent in depth?

Pro members get 13 hands-on Gemini lessons: Gemini Agent multi-step tasks, Gems custom builds, Deep Research with Workspace, NotebookLM workspace patterns, Workspace agent shortcuts that actually work, and the deep workflows across Docs, Sheets, and Gmail. Founding members lock in $9/mo forever (vs. $19 standard).

Back to Gemini Mastery → Try a prompt in the playground →

What you can do now

  • You know the five Gemini agent surfaces (Agent, Gems, Deep Research, NotebookLM, Workspace)
  • You understand Google's strategy: AI inside your Workspace, not separate
  • You've seen concrete examples of each agent's value
  • You can pick which surface to learn first based on how you work
  • Next step: Pro lessons walk you through each one with real workflows
Pro
Up next in Gemini Mastery

Lesson 02 · Gmail mastery — Gemini in your inbox

First hands-on Pro lesson: Gemini in Gmail — the patterns that turn email triage into a 15-minute morning ritual instead of a 90-minute reactive session. See the track →