Choose Orchestrator in Antigravity for handoffs, not a higher thinking level

Choose Orchestrator in Antigravity for handoffs, not a higher thinking level

If you use Antigravity to read files and discuss a proposed solution, the new Orchestrator option is not an automatic upgrade for every conversation. Google's documentation says the planner can answer questions, reviews and explanations directly when no changes are needed. Hydrogen does not have to take over.

Introduced in Google's November 12 announcement, the option is more useful when a task needs both planning and implementation, with a way to revisit the plan when execution goes wrong.

Check the planner before starting

Open the model menu below the prompt box. The model documentation lists Carbon and Argon planner options for Ultra subscribers. Paid Pro subscribers get the Argon version, with an additional eligibility exception for trials tied to qualifying premium devices. Hydrogen performs implementation in either combination; the name in parentheses identifies the planner.

For work requiring edits, the planner examines the project in a read-only phase, clears up ambiguities and sets out a plan. Hydrogen then changes files, builds and tests. An unexpected problem can send control back for a revised plan. For straightforward work whose steps are already clear, Google also recommends standalone Hydrogen as an option.

Extra is a separate control

Hydrogen defaults to High. Selecting Extra raises its reasoning budget to the level called xhigh in the command-line interface. Orchestrator instead uses a fixed configuration: its thinking level cannot be adjusted. Select an individual model if you want control over reasoning depth or want the same model to handle every step.

Check overages before comparing runs

The plans page ties quota consumption to work performed, rather than a fixed number of prompts. Complex tasks can consume more. Pro and Ultra subscribers can use AI credits after exhausting their baseline allowance. Credits are consumed at standard Gemini Enterprise consumption pricing, so automatic delegation should not be mistaken for free execution. Set AI Credit Overages to Never if you do not want credits used automatically. With that setting, once a model's baseline quota runs out, you must wait for it to refresh before using that model again.

Before changing your default, compare Orchestrator with your existing single-model workflow on a small task with explicit acceptance tests. Record elapsed time, test results and quota movement. A quick first response is not the same thing as a completed, verified change.

Share this article