Google Antigravity's /boost Turns Hard Bugs Into Multi-Agent Reasoning Missions

Google's agentic IDE just picked up a new gear. The team behind Google Antigravity has rolled out /boost, a slash command that hands off your prompt to a multi-agent reasoning pipeline built for the kind of problems that make single-shot coding assistants freeze up: race conditions, algorithmic optimization, deep refactors, and root-cause hunts across unfamiliar code.

The standard Antigravity agent handles feature work and pair programming well, but tasks that genuinely require planning, exploration, and verification benefit from prefixing the prompt with /boost, which triggers a heavier workflow that spends more tokens to think harder.

Three phases, one command

Under the hood, /boost activates an on-demand reasoning pipeline that breaks the problem apart, delegates focused workstreams to specialized subagents, and independently verifies solutions across iterative rounds. The docs describe three distinct phases:

1. Goal and strategy formulation. A Primary Orchestrator receives your prompt, inspects workspace context, and formulates an execution strategy, splitting the challenge into discrete verifiable subtasks.
2. Parallel execution and verification. The orchestrator dispatches work to subagents divided into implementation workstreams that build candidate solutions and generate unit tests, investigation workstreams that trace call graphs and analyze dependencies without modifying files, and local verification that runs build targets and test suites to validate hypotheses before reporting back.
3. Synthesis and delivery. The orchestrator validates the combined solution against full test suites and edge cases, feeds failed assertions back into the next iteration for automated correction, and only delivers a summary once all tests and requirements pass.

Each subagent runs in clean, isolated scopes, and the workspace model uses ephemeral isolated worktrees rather than the shared working tree that the default agent uses. That isolation prevents half-baked experiments from polluting your main chat context or your files.

Where /boost actually helps

Google is being explicit about the sweet spot. /boost targets the middle band between everyday autocomplete-style help and multi-day autonomous campaigns. The four use cases it calls out:

• Concurrency and race conditions: reproducing intermittent deadlocks in things like connection pools, where reproduction requires careful trace analysis and isolated verification.
• Algorithmic problem solving: implementing lock-free data structures, custom graph traversals, or SIMD-vectorized routines with stress tests against formal bounds.
• Non-trivial refactoring: jobs like migrating synchronous APIs to asynchronous patterns across multiple files while keeping callers green.
• Deep root-cause investigation: tracing why something breaks in a large unfamiliar codebase without touching any files.

Invocation is boring in a good way. In Antigravity 2.0 or the CLI you just type something like:

/boost Investigate the race condition in the session cache and implement a thread-safe fix with tests.

How it slots next to the other modes

Antigravity now has three execution modes. /boost sits between the default single-agent loop and the heavier /teamwork-preview mode designed for autonomous multi-day agent teams. The distinguishing traits:

Mode	Horizon	Architecture	Verification
Default Agent	Seconds to minutes	Single-agent direct loop	Single-pass tool check
/boost	Seconds to hours	3-phase reasoning hierarchy	Multi-round independent verification
/teamwork-preview	Hours to days	Multi-role agent teams	Adversarial falsification and audit

Default mode takes a single prompt and starts working. /teamwork-preview runs a two-phase scoping interview first. /boost jumps straight into execution with no scoping conversation, just deeper reasoning.

Availability, cost, and safety

The feature is available on paid plans across Google Antigravity 2.0 and the Antigravity CLI, meaning Pro and Ultra subscribers. There's no separate per-call price, but the token spend is meaningfully higher than a default agent turn since you're paying for orchestration plus multiple subagents plus verification rounds.

On safety, subagents inherit the file access rules and command permission policies configured for your workspace. When a worker proposes a protected terminal command or file edit outside trusted scopes, the authorization prompt surfaces in your interface for confirmation. Debug logs and scratch diffs stay in the isolated subagent context rather than cluttering your primary chat.

The bigger picture

Antigravity itself is Google's bet on agent-first development, with Gemini 3 Pro as the main reasoning engine, tuned for reasoning and tool use with strong performance on benchmarks like Terminal-Bench. The harness also exposes other frontier models through Vertex AI, including Claude Sonnet, so /boost can in principle route work to whichever model you have selected.

What /boost really adds is a productized version of the pattern power users have been hand-rolling for months: split a hard problem into pieces, run parallel exploration and implementation, verify against tests, iterate on failure. Making that a single slash command lowers the bar for reaching for deep reasoning when you actually need it, instead of burning tokens on it for every trivial task.