Introduction to Google Antigravity IDE for Developers

Learn the mental model behind Google Antigravity's agentic coding workflow — the loop from prompt through reasoning plan, tool use, and reviewable artifacts (plans, diffs, walkthroughs); how Projects set repository boundaries and when Local vs new-worktree mode is the right choice; the Agent Manager's role in running multiple agent conversations in parallel; and the concrete differences between an artifact-first environment and a chat-and-paste coding assistant. By the end of this lesson you will be able to narrate the agent loop step-by-step, choose a project mode for a given task, and explain why Antigravity does not write code the instant you press Enter.

Build the artifact-review reflexes that make Antigravity actually faster than typing. You will go deep on the three artifact types (plans, diffs, walkthroughs), see how inline artifact feedback steers the agent without restarting, learn what to look for when you spend 30 seconds reading a plan, and know when inline feedback stops being enough and a restart is the better call. By the end of this lesson you will be able to analyze a plan against a project's conventions, apply inline feedback to steer mid-task, and choose between iterating in the same conversation versus starting a new one for a given scenario.

Learn the two layers of Antigravity that sit alongside full agent tasks — Tab-level autopilot (Supercomplete, Tab-to-Jump, Tab-to-Import) for small single-file edits where opening an agent conversation would be overkill, and the sandboxed Browser Subagent for driving a real Chromium browser against your dev server as part of an agent task. By the end of this lesson you will be able to pick between a Tab-feature edit and an agent task for a given change, explain the Browser Subagent's isolated-profile security model, and recognize when the Browser Subagent is the right tool for end-to-end QA inside an agent task.

Learn to turn one agent conversation into throughput — the Agent Manager pane and how multiple agent conversations run in parallel against isolated worktrees. You will see how conversations are first-class, persistent objects you can revisit days later; how Ctrl+E toggles between the Agent Manager and the editor pane non-destructively; and how parallelism interacts with worktree isolation to make three independent tasks finish in the wall-clock time of the slowest. By the end of this lesson you will be able to run multiple agent conversations in parallel safely, use the Agent Manager to switch review attention between them, and recognize when parallelism actually buys you throughput versus when it just adds context-switching cost.

Close out AGY-100 with the three features that scale your Antigravity usage beyond daily tasks — /boost deep reasoning for when the task is worth extra tokens, custom SKILL.md files that teach the agent your conventions project-by-project, and rules that act as persistent guardrails across every conversation. Then a short tour of the Antigravity ecosystem beyond this course — /teamwork-preview, Claude and GPT via Google AI Pro, the CLI (agy), the Python SDK, and the IDE extensions — plus a signpost to AGY-110 if you want to drive a real build end-to-end. By the end you will know when /boost is worth it, how to author a SKILL.md the agent respects, and where to go next.