Google Antigravity introduced Dynamic Workflows: type /run-workflow and the main agent writes a Python script that schedules subagents with parallel() and pipeline(), each returning a structured result checked against a JSON Schema. You approve the script before it runs and can save it for reuse. Available on Google AI Pro and Ultra in both the App and the CLI.
Key Takeaways
- ✓New /run-workflow command: the agent writes a Python script that runs subagents concurrently with
parallel()or one by one withpipeline()(docs) - ✓Each subagent works in its own conversation and returns a result through a
finishtool validated against a JSON Schema, keeping file reads and test logs out of the main context - ✓You approve each script before it runs (presets like Turbo can allow it) and can save it to the project to rerun
- ✓Available on Google AI Pro / Ultra in the Antigravity App and CLI; ships its own Python, or set
ANTIGRAVITY_PYTHONto a 3.8+ interpreter - ✓The official X post had about 1.9k likes and 650+ bookmarks at write time
Key Decision Metrics at a Glance
Turn your technical choice into a development budget
Compare 40 dev plans & simulate token costs vs $20/mo subscriptions
Project Links & Resources
Direct AccessIn-Depth Technical Analysis
Background
A common problem with multi-agent work is prompt drift: a single long-running agent decides how to split, assign and merge work, so the flow is unpredictable and the context fills up with logs. Tasks like reviewing each changed file, auditing each package, or writing a failing test before the fix are really fixed procedures that are better controlled by code.
Architecture highlights
Per the official docs, the main agent reads the built-in run-workflow skill, writes a Python script for the task, and runs it in a separate background process. The script defines agents with agent(prompt, schema), runs them concurrently with parallel(items, fn) (one failure cancels the rest) or one at a time with pipeline(items, fn). Subagents return results validated against a JSON Schema, and a single agent can be kept across turns, so the reviewer that found bugs can later confirm the fixes. You can follow each phase and agent live.
Benchmarks and data
No benchmark or efficiency numbers were published. The official X post had about 1.9k likes and 142 reposts at write time.
Developer quick start
On Google AI Pro or Ultra, type /run-work in the Antigravity App or CLI, press Tab, describe the task, then review and approve the script. It fits running the same check across many files, staged steps, and reviewer-and-fixer loops; if agents need to coordinate with each other, the docs recommend /swarm instead. Ask the agent to save a working script to your project to reuse it.
Benchmark side-by-side against alternatives, or calculate monthly token cost vs subscription break-even.
Discussion & Comments
0Sign in to join the discussion
Connect with AI developers to exchange benchmark insights.