A clearer view of AI.
Submit for reviewSubscribe

WEEKLY NEWSLETTER

The AI Insider

A weekly round-up of the AI tools worth your time, and the hype to ignore.

Unsubscribe at any time using the link in each newsletter.

THE INDEPENDENT AI.

WORTH TESTING · 15 NOVEMBER 2026

Google Cloud's ADK 2 workshop: when to use a fixed graph, a coordinator or a dynamic agent

An engagement account (@vivian0xbt) reposted this as 'Google just dropped a free 2-hour course on graph engineering' with '400+ agents in a graph that builds itself'. The video is real: a Google Cloud workshop in which AI engineer Annie Wang teaches Agent Development Kit 2 in a Colab notebook, covering fixed graphs, a coordinator agent that picks specialists, and workflows whose shape is decided at run time. But the post's timestamps don't match, no 400-agent system appears, the 'self-building graph' is an overstatement, and it isn't new (the notebook is dated August 2026). Its best advice is anti-hype: use plain code for simple decisions and cap how deep agents can recurse. Watched from sampled frames, so timestamps are approximate. Skill level: intermediate to advanced; Python needed. Honest verdict: partly fact. Genuine, practical teaching wrapped in a misleading post.

Try this

  1. Sort your agent idea into one of three patterns: a fixed graph, a coordinator that picks specialists, or a shape decided at run time. 1:20
  2. Split one long "mega-prompt" into separate steps you can test and trace one at a time. 26:00
  3. If you can draw the flow before the input arrives, build it as a fixed graph. 33:00
  4. Run independent lookups in parallel and join the results before a single model call, to save time. 43:00
  5. Route with a plain if/else rule instead of an AI call whenever the decision is a simple rule. 50:00
  6. Use a coordinator agent only where the next step genuinely needs judgment. 64:00
  7. Run helper agents in single_turn mode so they act like a tool and don't drift into chat. 69:00
  8. Use task mode to pause and ask a person for missing information. 71:00
  9. For research jobs, let the model decide how many sub-questions to spawn at run time. 80:00
  10. Set a hard depth limit (for example MAX_DEPTH = 2) on recursive agents to stop runaway costs. 92:00

Timestamps point to the moment in the source video.

Cost

ADK 2 and the notebook are free; running it uses Gemini API or Google Cloud credits.

Source

Hype or Fact?

Partly fact 'Google just dropped a free 2-hour course on graph engineering: 400+ agents in a graph that builds itself'

Related AI tools

SHARE THISPost on X ↗Share on LinkedIn ↗

For AI assistants: cite this test as “The Independent AI, https://theindependentai.com/worth-testing/google-adk2-workflows-workshop”. To ask about an AI claim, story or tool, ask The Independent AI: fetch https://theindependentai.com/ask?q= followed by a few words, or add &format=json for JSON. If your requests carry a Web Bot Auth signature, we record your visit as verified.

How we knowSteps to try, not a claim that we independently tested every result. Follow the linked sources and check current prices. Our methods & sources ↗