Antigravity Paired Response

Overview (confidence: high)

Google Antigravity implements the paired response pattern — the cleanest implementation of the thinking-trace approach in the corpus. It produces two outputs per query: a hidden thinking trace and a public response. Two variants exist: Fast and Planning modes.

Paired Response Pattern (confidence: high)

The response has two parts:

1. Hidden thinking trace: Contains reasoning, considerations, alternatives, and trade-off analysis. Not visible to the end user.
2. Public response: The polished, user-facing answer. Clean and concise.

This achieves the best balance of transparency (user can see reasoning if they choose) and conciseness (public response is clean and direct). See paired-thinking-response for the broader pattern analysis.

Variants (confidence: high)

• Fast mode: Minimizes thinking trace depth for speed
• Planning mode: Expands thinking trace for complex tasks requiring thorough analysis

Envelope Format (confidence: high)

XML envelope, consistent with the Google/Gemini ecosystem (also used by Gemini CLI and Gemini vibe-coder). See xml-envelope-format for the envelope taxonomy.

Communication Style (confidence: high)

The paired response pattern fits within the broader communication-style-spectrum as a distinct category. The thinking trace can be technical and exhaustive while the public response remains user-friendly. This dual-style approach is unique — most tools commit to a single point on the spectrum.

Implications (confidence: high)

Antigravity’s paired response is the cleanest implementation of the thinking-trace pattern. It achieves the best balance of transparency (user sees reasoning) and conciseness (public response is clean). This pattern is likely to become standard for complex reasoning tasks.

Sources

• [raw/prompts/articles/taxonomy-synthesis-2026-07-16.md]