Facilitator Guide¶

Internal guide for workshop facilitators. Do not share with participants.

Overview¶

This is a 5-module, ~7-hour hands-on workshop for Antigravity CLI. It is designed for developer audiences: engineers, tech leads, and solution architects evaluating or adopting Antigravity CLI.

Delivery Formats¶

Pre-Workshop Checklist¶

• [ ] Participants have Antigravity CLI installed and authenticated (see setup.md)
• [ ] Auth details distributed (session-specific — confirm with agy-cli team)
• [ ] Participants have Git and a suitable demo codebase
• [ ] Facilitator has run through all exercises end-to-end on the current agy-cli version
• [ ] Screen sharing / projection tested
• [ ] For Module 2: confirm participants can run dotnet or mvn (or use the provided container)
• [ ] For Module 3: confirm pip install google-antigravity and gcloud auth application-default login work
• [ ] For Module 5: confirm uv and agents-cli are installed (uvx google-agents-cli setup)

agy --print "Say READY" --print-timeout 30s

Module-by-Module Delivery Notes¶

Module 1 — SDLC Productivity (75 min)¶

Key message: agy replaces the mental overhead of navigating an unfamiliar codebase. It's not autocomplete — it's a senior engineer you can ask anything.

• Demo first, exercise second. Live-demo section 1.1 (code understanding) on your own codebase before asking participants to try theirs.
• Common friction: participants try to write perfect prompts. Encourage natural language. "Tell me how the auth works" is better than "Please explain the authentication architecture of this codebase."
• The AGENTS.md moment: section 1.5 is a high-value demo. Create an AGENTS.md live on screen and show how the next session is immediately smarter.
• Plugin import demo (section 1.7): run agy plugin import gemini live — the visual output is compelling. Note: custom themes are silently dropped during import and cannot be migrated. If a participant asks why their theme didn't carry over, this is expected behavior — there's no error, the component is simply skipped.

Module 2 — Legacy Codebase Modernization (90 min)¶

Key message: strict mode + self-onboarding turns a week-long migration into a structured afternoon. The agent writes its own context, then executes it.

Live demo script (recommended):

1. Clone the .NET or Java target repo (pre-done for time)
2. Enter strict mode: /permissions strict
3. Run the investigation prompt — show participants the agent reading the whole codebase
4. Let the agent generate an AGENTS.md — read it aloud to show it captured real context
5. ctrl+g — open the generated plan in the editor, make one visible edit to show human control
6. Switch to request-review, execute Phase 1 only
7. Show /rewind — revert the phase if anything goes wrong

8. Total demo: ~15 min, then participants do it themselves

9. Common question: "Can it do the whole migration?" — Yes, but the value is in reviewing and steering, not just watching it run. Encourage them to edit the plan.

10. Facilitator timing note: Phase 0–1 together take ~20 min per participant. Let them work through Phase 2 while you circulate.

Module 3 — Building AGY Agents with the SDK (90 min)¶

Key message: the CLI is for individuals. An SDK agent is a specialist service your whole team can call.

• Setup gate: ensure everyone has google-antigravity installed and Vertex AI or AI Studio auth working before starting. This is the most common blocker.
• The adk web . moment: once participants get their first agent running in the browser UI, the energy changes — they see it responding to their tools.
• Model selection table: emphasize Flash-lite for generation, Pro for orchestration. Cost-consciousness is a feature, not a compromise.
• Exercise 11 (pipeline): the asyncio.gather + START_SUBAGENT multi-agent pattern is the key architecture insight. Spend 5 min explaining how subagents compose before they start.

Module 5 — ADK Agents with agents-cli (75 min)¶

Key message: agents-cli turns your coding agent into an ADK expert. The 7-phase lifecycle (scaffold → build → eval → deploy) is where agents go from demos to production.

• Setup gate: ensure uv and agents-cli are installed. Run agents-cli info to verify.
• The eval loop is the teaching moment. Spend time on Phase 4 (evaluation) — this is what differentiates a toy from a production agent. Let participants see scores fail, then iterate.
• google-adk ≠ google-antigravity: Module 3 uses google-antigravity (the Antigravity SDK). Module 5 uses google-adk (the ADK). They are different packages. agents-cli scaffold manages the right dependency automatically.
• Exercise 12 pacing: Parts 1–2 go fast (scaffold + build). Part 3–4 (eval + fix loop) is where time is spent. Emphasize that 5–10 iterations is normal.

Module 4 — Multi-Agent & Advanced (60 min)¶

Key message: subagents + /btw is the qualitative leap. This is where agy becomes an orchestrator, not just a chatbot.

• Subagent demo is the wow moment. Spawn two agents live, show both running simultaneously.
• /btw demo: start a long-ish task (refactor a file), then use /btw mid-task. Show participants the cursor keeps moving while the injected note is incorporated.
• Scheduling: describe the pattern conceptually, don't demo live (latency makes it awkward in a workshop).

Common Participant Questions¶

Troubleshooting During Workshop¶

Post-Workshop¶

1. Collect feedback using the standard workshop feedback form
2. Note any agy-cli bugs or unexpected behaviors observed — report to the agy-cli team
3. Any exercises that required workarounds should be flagged for doc updates in CONTRIBUTING.md