Google adds multi-agent '/boost' reasoning to Antigravity coding assistant
The new slash command triggers a three-phase pipeline to tackle complex software engineering tasks that exceed standard AI capabilities.
Google has launched "/boost," a new slash command for its Antigravity coding assistant designed to handle high-complexity software engineering tasks. The feature activates a multi-agent reasoning pipeline to solve problems that typically exceed the limits of single-turn AI interactions.
According to Google Antigravity documentation, the /boost command triggers a structured three-phase workflow: Goal & Strategy Formulation, Parallel Execution & Verification, and Synthesis & Delivery. To execute these phases, Boost deploys specialized subagents dedicated to implementation—handling coding and testing—and investigation, which focuses on tracing and root-cause debugging. These agents operate within isolated worktrees to ensure stability and precision during the process.
The gap in AI engineering
Antigravity serves as Google's AI-powered coding environment, but until now, a gap existed between "everyday engineering"—which is fast and interactive—and "long-horizon campaigns" that are autonomous and can span multiple days. Google positioned /boost to fill this void, specifically targeting high-intensity tasks such as intricate refactoring and the resolution of concurrency bugs.
A shift toward orchestration
This move signals a strategic shift from simple chat-based code generation toward multi-agent orchestration. By using a primary orchestrator to manage specialized workers who iteratively verify solutions, Google is attempting to move toward autonomous, verifiable engineering workflows. This approach reduces the reliance on a single large model to get everything right in one pass, instead favoring a system of checks and balances between agents.
Availability and reception
Access to /boost is currently restricted to users on paid plans and is available through both the Antigravity CLI and Antigravity 2.0. While the technical framework is now live, the rollout has met with mixed reactions. On Hacker News, some community members have described the feature as "half baked" and questioned the marketing copy, with some suggesting the promotion may be "astroturfed."
Industry observers will now be watching to see if this multi-agent approach significantly reduces the hallucination rates common in complex refactoring tasks, or if the overhead of the three-phase pipeline slows down the development cycle for professional engineers.