---
title: 'Google Antigravity Just Got Dynamic Workflows!'
source: 'https://youtube.com/watch?v=OP0wybd_nAo'
video_id: 'OP0wybd_nAo'
date: 2026-11-17
duration_sec: 619
channel: 'Julian Goldie SEO'
---

# Google Antigravity Just Got Dynamic Workflows!

> Source: [Google Antigravity Just Got Dynamic Workflows!](https://youtube.com/watch?v=OP0wybd_nAo)

## Summary

The video introduces Dynamic Workflows, a new slash command in Anti-Gravity that lets users automate multi-step tasks by describing them in plain English. The host demonstrates four practical workflows—lead enrichment, client audit, content planning, and daily roundup—and explains key concepts like parallel vs. pipeline execution, persistent agents, and model selection.

### Key Points

- **Dynamic Workflows Overview** [00:02] — Dynamic Workflows is a new slash command in Anti-Gravity called 'Run Workflow'. Users describe a job in plain English, the agent writes a Python script, and after approval, it runs in the background.
- **How It Works** [00:28] — The script acts as a conductor, splitting work into phases. Within a phase, 'parallel' runs agents simultaneously, while 'pipeline' runs them sequentially. Each agent returns results in a fixed shape that is validated.
- **Agent Isolation** [01:13] — Each agent works in its own conversation, keeping file reads, searches, and logs separate from the main chat. This prevents context overflow and keeps the main chat clean.
- **Workflow 1: Lead Enrichment** [01:42] — A workflow that reads a CSV of leads, spawns parallel agents to research each company, and a reviewer agent to polish openers. Output is a leads board with niche, pain point, and a personalized opener.
- **Workflow 2: Client AI Audit** [03:49] — Uses a persistent reviewer agent to read intake call notes, then fans out parallel agents per department to find automations. The reviewer ranks all ideas and writes a report.
- **Workflow 3: Content Week Engine** [05:28] — For creators, this workflow generates hooks, titles, thumbnail concepts, and captions for each video idea in parallel, then a judge agent selects the strongest one.
- **Workflow 4: Daily Roundup** [06:38] — Summarizes changes across chats, notes, and news in parallel, then merges them into a digest and creates a to-do list via a pipeline.
- **Saving and Reusing Workflows** [07:23] — Once a workflow works, it can be saved as a script and rerun with a simple command. Approval can be set to 'always allow' for convenience.
- **Model and Workspace Settings** [07:54] — By default, agents use the same model as the chat. For bulk simple tasks, use 'flash' or 'flash light'; for hard thinking, use 'pro'. A branch workspace gives each agent its own copy of the project.
- **When to Use Workflows vs Other Tools** [08:23] — Use sub-agents for single tasks, workflows for predictable steps, /swarm for agents that need to talk, /boost for deep thinking, and /teamwork for large projects.
- **How to Start and Common Mistakes** [09:06] — To start, type '/run work', describe the job, approve the script, and watch the phases. Avoid making agents depend on each other in parallel, give each agent all needed context, and watch for approval requests.

### Conclusion

Dynamic Workflows in Anti-Gravity enables powerful automation of repetitive tasks, saving significant time. By understanding parallel vs. pipeline execution and agent isolation, users can build efficient, reusable workflows.

## Transcript

workflows. You type one slash command [music] and Anti-Gravity writes a script same time. [music] Watch this. 12 leads, 12 agents all
researching at once, and my main [music] chat stays completely clean. Today I'll show you four workflows I built with it, the exact prompts I [music] typed, and which one you should build first for your business. So what actually is this?
Dynamic Workflows is a new slash command in Anti-Gravity called Run Workflow. It works in the Anti-Gravity app and in the Anti-Gravity command line. You describe a job in plain English. The agent writes a short Python script for that job. You
approve the script, then it runs in the background while you keep working. Think of it like an orchestra. The script is the conductor. It decides who plays and when. Every script is split into phases like discover then review. Inside a
phase, parallel means every agent plays at the same time. Pipeline means one agent after another in order, and every agent hands back a result in a fixed shape. That shape is checked before the script uses it. So you get clean,
structured answers back every single time. Here's the bit people miss. Each agent works in its own conversation. All the file reads, the searches, and the test logs stay inside that agent. Your main chat only gets the results. A
member asked this week why their Claude chats fall apart past 40% of the context. This is the fix. The heavy reading happens somewhere else. Let's build the first one. Workflow one is the lead enrichment sweep. Lead enrichment
came up again and again in the boardroom this week. People have a list of names and no time to research every company before reaching out. So here's my folder, a leads file with 12 companies, agencies, coaches, course creators, a
couple of software founders, a few online stores, and a snapshot of each online stores, and a snapshot of each website. I type / run work, press tab, and paste this, readleads.csv. For each company, in parallel, have one
agent read its file insights and return the niche, the biggest pain point it can see, and a one-line personal opener. Use the flash model. Then have one reviewer agent check that every opener sounds human and save enriched leads plus a
clean leads board page. Anti-Gravity shows me the script first. Allow running this workflow script. I click yes. Now look at the workflow step. Phase one, discover. Phase two, enrich. And here are the agent cards, one per company,
all running at the same time. I can click any card and read that agent's own conversation, its prompt, its tool calls, its answer. Then the reviewer phase kicks in. One agent reads every opener and rewrites the ones that sound
like a robot. And here's the result, a leads board with every company, its niche, the pain point, and an opener you can actually send. For an agency doing outreach, that's your whole morning of research done while you make a coffee.
Quick one before workflow two. If you want my best Anti-Gravity setups, like the exact workflow scripts from this video, the lead sweep, the client audit, the content engine, and the daily roundup, they're all inside the AI
Profit Boardroom. You'll get the prompts, the scripts ready to rerun, and step-by-step guides for every build, plus the faceless video system I use to make videos like this one. And there are thousands of business owners, agency
owners, coaches, and creators in there sharing what's working for them every single day. The link is in the comments and the description, or go to aiprofitboardroom.com. Okay, workflow two is the client AI
audit. A member posted that they're running their first proper AI audits for clients in the next few weeks, sitting with the owners, mapping their processes. So I built the audit as a workflow. My client here is a coaching
business. I've got the notes from the discovery call and one file for each team, sales, onboarding, delivery, content, and support. This prompt uses a special trick. Have a persistent reviewer agent read the intake call,
then fan out one agent per file in departments in parallel to find the three best automations for that team. Then ask the same reviewer to rank everything and write the audit report. The word persistent matters. A normal
workflow agent does one job and finishes. A persistent agent stays alive across turns. So the reviewer that read the call notes at the start is the same reviewer that ranks the ideas at the end. It remembers the client. In the
orchestra, that's the first violin. Plays at the start, waits, and comes back for the finale. Five department agents run at once. Each one returns three automations, the time they'd save each week, and the tools involved. Then
the reviewer comes back, ranks all 15 ideas against what the owner said on the call and writes the report. And there it is, a clean audit report. The top automations first, why each one matters to this client, and what to build in
week one. If you're a consultant or you run an AI agency, this is a deliverable you can walk into a meeting with. Workflow three is the content week engine. Someone asked how I make so many videos every day. Others asked for
script prompts, and one member said the boardroom made them stop betting on a single thumbnail. So this one is for creators and course owners. My creator here teaches AI to business owners on YouTube and runs a community. I've got a
file with her channel voice and six video ideas. For each idea in ideas, in parallel, have one agent write a hook, three titles, two thumbnail concepts, and a real caption in the voice from channel. Then have one judge agent pick
the strongest video for this week and save content week. Six agents, six ideas all at the same time. Each one reads the channel file first, so the hooks actually sound like her. Then the judge reads all six and picks the one to film
first and says why. Here's the page. Every idea with its hook, titles, and thumbnail ideas side by side, and the winner at the top. That's a full week of content planning in a few minutes. And you can test three titles instead of
guessing one. Workflow four is the daily roundup. This was the busiest thread in the boardroom this week. A member built a daily roundup across all their AI chats, and loads of people asked how. Another thread was simply called keeping
up. Too much happening, not enough time. So I've got a folder that looks like a busy agency owner's week. Chat summaries, meeting notes, and a few AI news notes. For every folder in this project, chats, notes, and news in
parallel, have one agent summarize what changed this week. Then, one after another, have one agent merge them into a digest and a second agent turn it into a three item to-do list. Save digest. This one mixes both. The summaries run
in parallel. Then a pipeline where the merge happens first and the to-do list comes second because the second step needs the first one's answer. And here's the digest. What changed this week and the three things to do today. Not one of
those files landed in my main chat. Now the part that makes this really useful. Once a workflow works, keep it. I type, [snorts] save this workflow script to scripts/roundup.py. Next time I just say run workflow, run
scripts/roundup.py. Same steps every time. Antigravity still asks me to approve it unless I've chosen to always allow it. Two more settings worth knowing. First, the model. By default, every agent uses the same model
as your chat. For big, simple checks across lots of files, ask for flash or flash light. For hard thinking, ask for pro. A member asked about heavy agent jobs eating through their usage. This is how you keep the bulk work light.
Second, the workspace. If your agents are editing code at the same time, ask for a branch workspace. Each agent then gets its own copy of the project so they never trip over each other. So when do you use this and when do you use
something else? Loads of members asked how all these agent tools fit together. Use a sub agent when you want to hand off one task, like one research question. Use dynamic workflows when the steps are predictable. The same check
across a list, steps in order, or a reviewer and fixer loop. Use / swarm when the agents need to talk to each other on a shared board while they work. Use / boost for one really hard problem that needs deep thinking and / teamwork
for a big project with lots of milestones. Simple rule, if you could write the steps on a sticky note, it's a workflow. Here's how to start today. Open Anti Gravity. Open the folder you want to work in. Type [clears throat] /
run work and press tab. Describe the job and the result you want back. Read the script, click allow, then watch the phases. Three mistakes to avoid. One, don't make agents depend on each other inside a parallel phase. They can't see
each other. If step two needs step one, use a pipeline. Two, give each agent everything in its prompt. Workflow agents don't see the script or your chat. Three, watch for approvals. Each agent asks for permission in its own
conversation. If a card says waiting for approval, click it and approve. Otherwise, it just sits there. And which one should you build first? If you run an agency, start with the lead sweep. If you're a coach or consultant, the client
audit. If you're a creator or you run a course, the content engine. And if you just feel buried, start with the daily roundup. If you want every workflow from this video, the four prompts, the saved scripts, and the sample folders so you
can run them straight away, they're inside the AI Profit Boardroom with a link is in the comments and the description or go to aiprofitboardroom.com. See you in the next one.
