Agents · Lesson 01 Pro ~11 min read Multi-step + Workspace

Gemini Agent: Google's answer to autonomy.

In May 2026 Google consolidated its agent story. Project Mariner was retired as a standalone product; its technology now powers Gemini Agent — the unified agentic mode inside the Gemini app. It handles multi-step tasks across Deep Research, Canvas, Workspace integration, and live web browsing. One mode, one interface, end-to-end.

Workflow 01 Multi-step web tasks

1

Tasks that span planning, research, and action

The Project Mariner heritage shows up in Gemini Agent's web automation. Tell it an outcome; it plans the steps, executes across multiple sites, and produces the result.

The prompt that works

Multi-site taskFind me three new restaurants near our office that: • Serve a private dining room for 12-15 people • Are open for dinner Tuesday-Thursday • Have at least a 4.3 rating on Google Maps with 100+ reviews • Aren't already on our 'recently used' list (which I'll paste below) For each candidate, verify availability for next Tuesday at 7pm by checking their reservation system (OpenTable, Resy, or direct). Don't make a booking — just confirm availability and pricing for the private room. Produce a comparison table I can share with the team.

Best use cases

  • Travel and event planning
  • Vendor research with availability checks
  • Comparison shopping across platforms
  • Cross-site data verification
Gemini Agent asks for confirmation before any irreversible action. If it didn't ask, double-check what just happened.
Time savings: Multi-site research + verification: hours → 10-20 min.

Workflow 02 Workspace deep integration

2

Tasks across Docs, Sheets, and Gmail in one go

Where Gemini Agent really shines is the Workspace integration. Tasks that touch Docs, Sheets, and Gmail in one workflow happen as one prompt.

The prompt that works

Workspace kickoffCreate a project kickoff document for [Project Name]: 1. Search my Drive for any prior docs about this project — incorporate what's already decided 2. Search my Gmail for the original kickoff thread — pull goals and stakeholders 3. Create a new Doc with: project summary, goals, stakeholders, key dates, open questions, RACI chart 4. Build a tracking spreadsheet with: task name, owner, due date, status, dependencies 5. Set up a calendar event for the kickoff meeting (next Tuesday 10am if attendees are free) 6. Draft (don't send) a kickoff email to the stakeholders Share the Doc with the stakeholders in view-only mode.

Best use cases

  • Project kickoff packages
  • Customer onboarding doc generation
  • Event planning across multiple Workspace files
  • Hiring loop coordination
Sharing permissions are a place to be cautious. Gemini Agent will share things if asked — verify the permission level (view, comment, edit) matches your intent.
Time savings: Project kickoff setup: 2-3 hours → 15 min including review.

Workflow 03 Long-running background tasks

3

Tasks that take time — let Agent work in the background

Gemini Agent handles long-running tasks (research, comparison, multi-step automation) as background jobs. You're notified when done; the result is in your account.

The prompt that works

Background researchRun a deep competitive analysis on [Competitor]: • Pull their public messaging, pricing, recent product launches • Read every analyst report you can find from the last 12 months • Search Reddit, Hacker News, and product-review sites for user sentiment • Compare their messaging to ours (use our website + recent press releases as input) • Identify 5 strategic vulnerabilities we could exploit This is going to take 30-60 minutes. Run it in the background. When complete, save the report to Drive and ping me.

Best use cases

  • Strategic competitive analysis
  • Market research over long time horizons
  • Document synthesis from large corpora
  • Multi-source consolidation projects
Background tasks are stateful — they run for a while. Don't kick off 20 of them at once; pick the most valuable tasks and let them complete.
Time savings: Background processing: hours of compute, zero hours of your attention.
Mariner heritage

Project Mariner was Google's standalone browser-agent research prototype, shut down on May 4, 2026. Its technology is now embedded in Gemini Agent — same core capabilities, integrated experience. If you used Mariner before, expect a similar but more polished experience.

Hand Agent a real Workspace project

Pick a project that requires creating multiple Workspace files. Use Gemini Agent to set up the whole package in one prompt. Notice where it's faster than doing each file individually, and where you'd still rather build by hand.

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

What you can do now

  • Use Gemini Agent for multi-step tasks, not single questions
  • Verify share permissions match your intent
  • Run long tasks in the background; don't watch them work
  • For irreversible actions, expect confirmation prompts
  • Combine Workspace integration with research for richer outputs
Pro
Up next in Gemini Mastery

Lesson 02 · Gems — your custom Gemini agents

Gems are Gemini's custom-agent feature. Like Custom GPTs but with Google's integration into Workspace, Drive, and Gmail. The build pattern that works. See the track →