Google Antigravity can now turn one request into a small team of coding agents — and instead of letting the model improvise the coordination, the agent first writes a Python script you can read, approve, save, and run again. Why it matters: multi-agent coding work becomes reviewable and repeatable, rather than a one-off chat you hope goes right.
November 14, 2026
What happened
Google’s November 13 Antigravity release adds Dynamic Workflows. Type /run-workflow and a task, and the main agent loads Antigravity’s built-in workflow skill and generates a Python script for the job. Antigravity shows you the script and waits for approval before running it in a separate background process, while your original conversation stays available. When it finishes, the main agent reads the logs and summarizes the results. Google’s changelog describes the feature as letting “agents orchestrate parallel subagents and multi-step tasks in Python using the runtime bundled with the app.”
How it works
The generated script uses a small set of building blocks: phase() to group work into named stages, agent() to define a worker with its own prompt, role, model, and workspace, parallel() to run a step across many items at once, pipeline() to process items one by one, and access to earlier steps in the conversation. Each workflow agent gets its own conversation and cannot see the script or its peers, so the script has to hand it everything it needs. Results come back through a finish step that validates them against a JSON Schema before the script can use them — which keeps the main conversation’s context clean during big jobs. Parallel editing agents can each get an isolated Git branch workspace so they do not overwrite one another, and each step can pick a cheaper or stronger model (Flash-class for high-volume checks, Pro-class for deeper reasoning). Saving the script as a .py file in your workspace makes the same orchestration reusable for recurring audits, refactors, and test runs.
Plans and availability
Dynamic Workflows are available with Google AI Pro and Google AI Ultra, in both the Antigravity App and the Antigravity CLI. There is no separate price — workflows consume your plan’s existing model credits, so how many agents a workflow fans out to is the main cost control. No local Python install is needed; the workflow runs on the runtime bundled with Antigravity. Dynamic Workflows sit alongside Antigravity’s other multi-agent options: /swarm, where agents negotiate over a shared message board, and plain subagents for a single delegated task.
Why it matters
Every major coding tool is racing toward parallel agents, but the coordination is usually hidden inside the model — hard to inspect, hard to reproduce, and hard to trust on a large codebase. Putting the plan in a plain Python script you approve before it runs, with schema-checked handoffs between agents, is a more engineering-minded answer: the workflow itself becomes code you can review, version, and re-run. If you already pay for Google AI Pro or Ultra and use Antigravity, this is the feature to try on a repetitive job first — a review pass over every file changed on a branch is the textbook example.
Frequently asked questions
What is a Dynamic Workflow in Google Antigravity?
A Python script the agent generates from your /run-workflow request that orchestrates parallel and sequential subagents, with results validated against a JSON Schema. You approve the script before it runs.
Who can use Dynamic Workflows?
Google AI Pro and Google AI Ultra subscribers, in the Antigravity App and Antigravity CLI. There is no extra fee; usage draws on your plan’s model credits.
How is this different from /swarm?
/run-workflow uses a fixed script with deterministic control flow for repeated checks and ordered stages. /swarm lets agents coordinate and review each other through a shared message board, which suits open-ended, long-horizon tasks.
Sources: Google Antigravity changelog (November 13, 2026), AlphaSignal, Google Antigravity on LinkedIn