跳转至

Facilitator Guide

How to deliver this workshop to enterprise customers and internal teams.


Pre-Session Checklist

  • [ ] Environment tested: Run ./setup.sh on a clean clone within the last 24 hours
  • [ ] Auth verified: gemini -p "Hello" returns a response (no auth errors)
  • [ ] Demo app running: cd demo-app && npm run dev starts without errors
  • [ ] Conductor installed: gemini extensions install https://github.com/gemini-cli-extensions/conductor
  • [ ] Terminal setup: Font size 16+, dark theme, no transparent background
  • [ ] Backup plan: Have pre-recorded terminal sessions for network failures
  • [ ] MCP tokens: GitHub Personal Access Token configured if demoing MCP
  • [ ] Rate limits: Check free tier quotas — Vertex AI recommended for full-day workshops

Timing Options

Quick Demo (2 hours)

Block Time Content
Setup 15 min Environment setup + first prompt
UC1 (SDLC) 45 min GEMINI.md + Conductor + Policy engine
UC3 (DevOps) 30 min Headless mode + hooks demo
Q&A 30 min Open discussion

Standard Workshop (4 hours)

Block Time Content
Setup 15 min Full environment setup
UC1 (SDLC) 60 min Complete walkthrough
Break 15 min
UC2 (AppMod) 60 min Plan Mode + subagents
Break 15 min
UC3 (DevOps) 45 min Headless + GitHub Actions
Wrap-up 30 min Q&A + next steps

Full Day (8 hours)

Block Time Content
Morning Session 3.5 hrs All 3 use cases with hands-on exercises
Lunch 1 hr
Extensions + Outer Loop 1 hr Extensions Ecosystem (30 min) + UC1 Part 2 Outer Loop (20 min)
Afternoon Session 2.5 hrs Customer-specific exercises using their own repos

Audience-Specific Delivery

Audience Start With Emphasize Skip/Skim
Backend Developers UC1 (SDLC) GEMINI.md, Conductor, subagents Enterprise config
Platform/DevOps UC3 (DevOps) Headless mode, hooks, GitHub Actions Conductor
Tech Leads/Architects UC2 (AppMod) Plan Mode, model routing, governance Hook scripting details
Security/Compliance UC1 §1.5 (Governance) Policy engine, sandboxing, tool isolation Conductor, batch ops
Engineering Leadership Quick Demo All 3 use cases at overview level Hands-on exercises

Live Demo Tips

Terminal Setup

  • Font size: 16pt minimum (18pt recommended for conference rooms)
  • Theme: Dark background, high contrast
  • Terminal width: At least 120 columns
  • No transparency: Solid background so projectors work
  • Split panes: Keep the walkthrough doc visible alongside the terminal

Pacing

  • Show, don't tell: Run the command first, explain after
  • Wait for output: Let the agent complete before narrating
  • Acknowledge latency: "This is thinking — it's analyzing the entire codebase right now"
  • Have fallbacks: If the agent gives an unexpected response, use it as a teaching moment: "See how we can steer it?"

Common Issues During Live Demos

Issue Recovery
Rate limit hit Switch to Vertex AI auth, or use a different Google account
Agent gives wrong answer Use model steering: "That's not quite right. The pattern we use is..."
Hook fails silently Check /hooks panel — usually a permissions or JSON issue
Conductor not installed gemini extensions install https://github.com/gemini-cli-extensions/conductor
Slow response Normal for first prompt (cold start). Subsequent prompts use token caching.

Customizing for a Customer

Before the Workshop

  1. Clone their repo (or a representative sample) into demo-app/
  2. Write a custom GEMINI.md encoding their coding standards
  3. Write a custom PRD targeting a real feature they want to build
  4. Configure MCP servers for their tool chain (Jira, GitHub Enterprise, etc.)

During the Workshop

  • Replace ProShop examples with their codebase
  • Use their coding standards in GEMINI.md demos
  • Demo MCP against their actual Jira/GitHub instance
  • Have them write policy rules for their security requirements

After the Workshop

  • Share the repo link for self-paced follow-up
  • Offer to review their GEMINI.md files
  • Connect them with the Google Cloud AI team for Vertex AI setup

Bonus: gemini-cli-scanner

The gemini-cli-scanner is a TUI tool that scans a developer's Gemini CLI installation and generates a maturity report.

Use it as: - Pre-workshop audit: Scan participant environments to identify setup issues - Closing activity: Have participants scan their setup after the workshop to see their "maturity score" - Team assessment: Aggregate scans across a team to identify standardization opportunities


Key Talking Points

On Context Window

"Other tools only see the files you manually open. Gemini CLI reads your entire codebase — architecture, dependencies, test suites, configs — and reasons across all of it at once. Your agent doesn't just autocomplete lines. It understands how your systems connect."

On Governance

"Claude Code and Cursor don't have a policy engine. They don't have hooks. They don't have tool isolation for subagents. When your CISO asks 'how do we prevent the AI from running rm -rf?' — you need guardrails-as-code, not just a checkbox."

On Cost

"Gemini CLI with a personal Google account is free. Free. With token caching, your enterprise Vertex AI costs are a fraction of what you'd pay for comparable tools. Full-codebase awareness means fewer sessions and less back-and-forth — you stop explaining context and start shipping code."

On Conductor

"Most AI tools generate code one prompt at a time. Conductor gives you persistent specs, phased plans, and progress tracking that survives across sessions. It's the difference between giving a contractor verbal instructions and handing them architectural blueprints."