Google Antigravity Dynamic Workflows: Run a Team of Coding Agents From One Script

Photo: Hippopx (public domain). Illustrative image.
What happened. Google shipped Dynamic Workflows in Antigravity v2.32.0 on 13 November 2026, a feature that turns multi-agent coding from a conversation into a script. The centrepiece is the /run-workflow command: instead of asking one agent to do a big job, you describe the job and the agent writes a short Python script that coordinates many agents — running some in parallel, others in sequence — and you approve that script before anything runs.
How it works. The script uses parallel() and pipeline() phases to organise the work: independent tasks run side by side, dependent ones wait their turn. Each agent returns a JSON-Schema-structured result, so the outputs are machine-readable and the workflow can check them, pass them on, or branch on them. The key design choice is the approval step. Multi-agent systems are powerful and unpredictable in equal measure; making the coordination plan a readable script that a human signs off on is Google’s answer to the control problem — the plan is inspectable before the agents start spending tokens and touching files.
What it costs and who gets it. Dynamic Workflows is available in the Antigravity App and CLI, on Google AI Pro and Ultra plans only. There is no separate charge: workflows consume the plan’s existing model credits. It runs on Gemini 4 Carbon, Argon and Hydrogen. It is distinct from two neighbouring features: Orchestrator, which arrived on 12 November and switches models mid-session, and /swarm, which puts agents on a shared board. Dynamic Workflows is the scripted, repeatable one — closer to a build pipeline than a brainstorm. Antigravity lead Varun Mohan also announced a Gemini quota reset the same week, though its duration was not stated.
Why it matters. Coding agents are splitting into two philosophies: the single agent you chat with, and the team of agents you orchestrate. Dynamic Workflows is Google’s bet that the second model wins for real projects — but only if the orchestration itself is transparent and reviewable. The same week, GitHub announced it will retire six Copilot models on 14 December, a reminder that the model underneath any of these tools changes constantly. Our three-way comparison of Antigravity vs GitHub Copilot vs Claude Code weighs the three against each other, and What Is an AI Agent? explains the category in plain English.
Sources
- Google — “Dynamic Workflows”, Antigravity documentation: https://antigravity.google/docs/dynamic-workflows/
- FreshSynapse — “GitHub Copilot Is Retiring 6 AI Models on December 14”: https://freshsynapse.com/articles/github-copilot-retires-six-models-december-14-2026-11-16/
- FreshSynapse — “Google Antigravity vs GitHub Copilot vs Claude Code: Which AI Coding Agent Should You Pick?”: https://freshsynapse.com/articles/antigravity-vs-copilot-vs-claude-code-2026-11-16/
- FreshSynapse — “What Is an AI Agent? A Plain-English Guide to How AI Agents Work”: https://freshsynapse.com/articles/what-is-an-ai-agent/