Same ideas, two stacks

Every row in these tables is one skill wearing two outfits. The Google column is what you have free for a year; the Internet Menace column is the same idea with the branding sanded off. Learn the idea instead of the product and month thirteen becomes a pricing decision. Learn the product instead of the idea and you start over every time a rebrand ships.

Tool equivalence: learn the idea once, run it in either stack

Portable ideaGoogleInternet Menace
A file the agent re-reads every sessionGEMINI.mdCLAUDE.md
An agent with a filesystem and a shellgemini-cliClaude Code
A saved persona with standing instructionsGemsSystem prompts and playbooks (§2.6)
Review the plan before executionDeep Research "Edit plan"; Antigravity Implementation PlanSpecifications, reviewed (§2.2)
A reviewable record of what changedAntigravity WalkthroughGit discipline and pull requests (§6.2)
Answers anchored to sources you control@Drive grounding, NotebookLMContext engineering (§2.4); RAG
A protocol for adding tools to an agentMCP servers via npx add-mcpMCP, the same protocol
Compute you do not own vs. compute you doColab GPU tiers, Vertex AISelf-hosting (§4.4, §4.5)
Knowing what a run costsCompute units, credits, budget alertsCost literacy (§4.6)

Where each part of the guide goes deeper

Source: Willson (2026), Introduction to Google AI Pro for Graduate Data Science Students.

Figures and setup steps live on the toolkit page.