▶️YouTube

DON'T build AI automations, build agentic workflows! (Google Antigravity)

Duncan Rogoff | AI Automation·youtube.com·21 min read·Mar 28

Select text to create a clip

Transcript — click timestamp to jump, select text to highlight

387 segments

Stop building AI automations, start building agentic workflows.

Google recently released their vibe coding platform, Antigravity,

and it's completely changing the way AI systems are built.

I run a six-figure AI agency, I've built hundreds of AI automations

using no-code platforms like N8N,

and I've sold these systems to multi-billion dollar companies.

But agentic workflows are fundamentally changing how I run my business.

No-code platforms eliminated the need to code with their visual interface,

but now you don't even need that.

With tools like Antigravity, you just describe what you want in plain language.

You no longer have to think about multi-node workflows or even troubleshooting.

The system figures it out and corrects itself when something breaks.

In this video, I'll show you the best way to set up Antigravity using the DOE framework,

which stands for Directive, Orchestration, and Execution,

and then we'll use that to build a Reddit news scraper which delivers a daily digest to you

of all the trending topics in your niche,

complete with infographic-style visuals in just a matter of seconds.

If you want access to the agent we built today and want to learn how to make money with AI and

automations, I'll leave a link in the description.

Let's dive in.

Here we are inside of Google Antigravity,

and I wanted to take this automation that I built,

which basically scrapes a bunch of Reddit feeds to get all of the trending topics relevant to my audience,

and then just delivers them into this really clean email format for me.

And I wanted to see if we could do this all today, not knowing how to code,

not knowing how to do anything technical, and letting Antigravity do all of the work for us.

If you've never used Antigravity before and you want to know how to install it,

you can just come over to antigravity.google.

Go ahead and click download down here or download up here in the top right.

And then what you're going to do, you're just going to install this on your system.

You're going to log in with your Google account, and you are ready to go.

In order to start using this, you just want to make sure that you are inside of a folder

somewhere on your own machine.

So I'm just going to go ahead and click open folder,

and I'm just going to go over to my desktop, and I'm just going to create a brand new folder,

and I'm just going to call this Reddit News.

It really doesn't matter what you call this.

And now we have a folder.

Go ahead, click open, and we can just go ahead and click.

Do you trust the authors?

Yes, I'm the author, so I trust anyone on the folder desktop.

Click trust, and now we have Antigravity set up.

It's ready to get to work for us.

There are a couple of main things that you should know if you've never used Antigravity before.

The left side here is the Explorer, which will show you how to have Antigravity set this up

entirely for you so you don't have to do anything.

This is where all of your files and folders are kept.

This right side is the agent where you actually talk to Antigravity in normal language and tell

it what you want to do.

And then if you come up here, Terminal, you can go ahead and click New Terminal.

And it may seem fancy, but if you just go ahead and type Terminal,

this Terminal window has been around forever,

and it's basically just a fancy way for you to talk to your system.

You really don't need to know all that much about it, but just know that this is here.

So at this point, most people get lost, but let me show you exactly how to set this up

so you never have to worry about it again.

And you can literally use this format to create anything you want.

So I'm going to go over here and this is this agents.md file,

and this is how you want the system to behave.

Truth be told, I got this from my man, Nick Sarayev.

He gives this out for free, so I'm sure he won't mind sharing with you,

especially if I'm hyping up his channel right here.

But this is a really effective format.

What this is, is these are instructions for Antigravity

that tells it exactly how you want to structure all of your files and folders

and exactly how you want it to behave.

So I'm just going to go ahead and I'm just going to highlight all of this

and copy this over into Antigravity.

I'll leave a link in the description where you can get access to this yourself.

So you can just go ahead and come into file, make a copy to your own Google Drive,

or go ahead and download this in whatever format that you want.

So if we take this over here and we come back to Antigravity,

I can come into Reddit News and I can go ahead and click the plus on new file.

We can just call this agents.md.

And .md is basically just in markdown format, which Antigravity will recognize.

And then I can just paste in all of this and we can see

it turned all of those asterisks and stuff like that into nice kind of language.

And just go ahead and hit command S.

This tripped me up the first time.

You actually have to save this file.

So just hit command S on your system.

So I'm actually going to have Antigravity set up all of our files for us.

And you can see it cooking in the background while we read through these instructions.

And I can tell you exactly what they do.

If we come over into our agent here, I can just say instantiate based on agents.md.

And instantiate means like go, like get started.

Fancy word, simple meaning.

So what this is going to do is it's going to read through all of the agent instructions.

What I love about Antigravity is you can actually see it thinking.

This is the entire thought process that it went through.

So you can just see how it's beginning to process this information.

And so Antigravity is going through here.

It's figuring out how it wants us to think.

And the main part down here that it's looking at right now is this directory structure.

Temporary files to store all of our intermediate files.

Execution scripts.

So any Python scripts which show us how to execute.

Any directives, which is basically us telling the system what to do.

An environment file which will store all of our API keys or anything like that that we might need.

In this case, we may not need it, but we might.

It's just good to have.

And then any credentials or things like that.

And you can see here that this agent is cooking up for us over here.

And on the left hand side is beginning to create all of these files and folders for us.

There are two main principles of this that make it really, really powerful.

One is this three-layer architecture.

And two is this idea that this is self-annealing or self-healing or really self-correcting.

Which is like a major upgrade from building automations inside of a platform like N8N.

So if we look over here, you can see this is already done.

It instantiated the three-layer architecture based off of the file that we set.

And it's created all of our files for us.

So now we don't have to worry about this ever again.

And now it understands this agents.md folder.

And it basically knows exactly how we want it to behave.

And really from here, you can just tell this thing to build whatever you want in plain language.

And it will go out and do it.

The first thing to understand is this three-layer architecture.

Layer one is directive or what to do.

Basically just standard operating procedures written in Markdown.

And so again, anti-gravity will create these for you or for itself, really, which is even crazier.

It's just going to define the goals, any inputs, any tools or scripts to use.

And again, it's going to figure out this information for you.

It's going to figure out the best way to do it.

And we can actually tell it to improve itself over time.

Layer two is the orchestration layer.

This is the job of anti-gravity.

Basically, its job is to figure out all of these pieces of the puzzle and figure out how they talk to each other and connect with one another.

And then layer three is execution, writing the code or the Python scripts that actually tell the system what to do and how to behave or execute on the task that we have given it in the directive files.

That's the first principle to understand.

The next big thing to understand is this self-annealing factor.

When things break, read error messages and stack trace.

So basically read the error messages, figure out where there are problems and fix those errors.

Fix the script, test it again until it works.

Update the directive.

So update any of the instructions to accommodate these changes or the errors that we fix.

If we, you can see if we've hit an API rate limit, look into the API, figure out how to fix it and then rewrite the scripts to accommodate.

So this is incredibly powerful.

This system is just going to improve itself over time.

Really crazy.

So again, this panel on the left is all your files and folders.

You mostly never have to look into these.

This here is like where you can actually get access to any of the individual files.

This panel on the right is where you talk to anti-gravity and tell it what you want to do.

And then the terminal on the bottom is basically just how you run whatever it is that you've built once you're finished with it and ready to test.

So remember I said I want to recreate this kind of workflow.

So at a high level, what this automation is doing is it's just scraping a couple of different subreddits.

It's filtering out posts from the last six hours.

It's kind of using AI or chat TPT to analyze them to make sure that they're relevant for my audience.

And then it's sending me an email.

That's it.

This is a great way to stay on top of trends, know what's going on in your industry and get ideas for your content based off what people are already talking about.

So I just created this really simple action plan.

If you have this, great.

If you don't have it, no worries.

But this is me just telling anti-gravity what I want it to do.

So I just want to pull the latest posts from one or more Reddit feeds.

I want to normalize the data so every entry has the same basic fields like the title, the link, any upvotes, the actual text, the number of comments.

I want to evaluate, analyze each post to make sure it's right for my audience, summarize all that, and then save this as a .txt file somewhere.

So I'm just going to copy this.

And if you have this, again, great.

And if not, you can literally just say I want to create a system that does XYZ and anti-gravity will go figure it out for you.

So what's cool is you can type to this or you can use a little dictation mode down here just to talk to the system.

The more information you can give the system, the better.

One thing I know is that Reddit RSS feeds come in this really predictable format.

If you don't know what an RSS feed is, I realize I've been saying it, it's basically just this is what's going on behind the scenes of any subreddit.

If you look really closely, you can see actually the text and the titles and everything that has to do with the individual posts.

But this is just a really quick way to process all the information.

It's literally the same as coming over into like RN8N and filtering by new.

It's just the code way to actually look at all of these posts.

So I can come back into anti-gravity and I can say RSS feeds, follow this predictable format.

Just paste this in.

So again, the more information you can give it, the better.

We can just see that the model down here that we're using is Gemini 3 Pro.

You can select different models if you want.

Quad is incredible at this.

I'm just going to use the built-in Gemini model for now.

And we're just using this in planning mode.

So it's actually going to create a plan for us before executing any tasks.

And that way we have control and approval over it and kind of think through how to do this.

One thing you can also do here is give me three options on how to approach this task.

So now it's just going to figure out three options or three best ways to approach what we're asking it to do.

It's going to present them to us as the human.

We're going to decide which path we want to move forward with.

So we can see here, this has been thinking, designing the Reddit workflow options.

I'm devising three distinct approaches for the Reddit Pro's workflow.

These strategies are centered around collecting, cleaning, evaluating, summarizing, and distributing Reddit content from RSS feeds.

It's keeping the three-layer architecture that we devise.

And it's just going ahead.

And it's going to start creating all of the files and folders for us.

Pretty crazy.

So we can see here that this just went ahead.

Here are three options for implementing your Reddit RSS workflow within our three-layer architecture.

Option one, the daily batch pipeline.

This is the simplest.

This is just a single workflow that runs everything in sequence.

So it's going to fetch an RSS feed.

It's going to filter it all out.

It's going to use an LLM like ChatTBT to evaluate it.

It's going to create a summary and then write the file.

This is really great.

It gives us the pros and cons.

Option two is the most robust option, which is going to use a separate SQL database to track every post that is seen.

And option three is clustered inside agents.

I don't know what this means, but let's go ahead.

We can see this focuses on smarter summarization instead of evaluating and summarizing list items sequentially.

It's going to group them first and then evaluate them.

So that's pretty cool, actually.

For today, I'm just going to go with this recommendation of option one just to get something up and running quickly for the purpose of this video.

And we can just talk to this.

Let's go ahead with option one.

Just sent this off as a voice note.

It's going to analyze my voice note.

Okay, so I'm diving into the daily batch pipeline.

It's going to craft all the markdown directives, again, which is it's writing directions for itself on how to behave, which is totally crazy.

I'll open them up for you when it's done.

What did this take?

We can see this took 10 seconds to do this.

It created this whole implementation plan for us.

So it wants to create a system to fetch RSS feeds.

And then it proposed changes.

So we're going to actually create a new markdown file, which is a new direction to define the five-step process that we've given it.

We want validation rules.

So we give it a 72-hour window, so we want to filter any posts in the last three days.

And then here we have an execution layer.

So this is what's going to run.

Once we run this execution, it's going to perform all of these steps.

It's going to fetch the feeds.

It's going to clean the posts, evaluate them, summarize them, and then save the posts to our system.

And so basically it just says, I created a plan for you.

You can click this button to open it like this.

And if you're ready to get to work, go ahead and click proceed.

And anti-gravity is going to start building this out for us and doing all of the coding without us knowing how to do anything.

And so, again, what's really powerful about this is if you're someone coming from the automation world or if you're not, you don't have to learn all of these nodes.

You don't have to learn what each one does.

You don't need to learn how to basically like fill out all of these different fields or all of these different text boxes, right?

Or how to like dynamically link content.

To me, this is simpler than coding, but this is significantly more complex than just telling anti-gravity what you want it to make.

So this saves you a massive step.

So this like accelerates the learning curve like a hundred times.

It's crazy.

There are a couple of challenges with it if you are new to code or you haven't coded before is that you are really just trusting the system to do the best job.

And these AI tools are getting better and better every single day.

Anyway, I'm not a coder myself, so I'm not the right person to go ahead and look into this Python script that it is writing and tell it whether it's doing a good job or not.

But like, look, in just a couple of seconds, it's written this entire script for us.

I don't know what any of this means.

And so I can't check it myself.

But again, with the self-annealing factor, if it's not working, we can say, hey, anti-gravity, this isn't working or I want to make these improvements here.

And it's going to know how to do that and it's going to make the improvements for us.

And so we can see here that's exactly what it's doing.

It's debugging the RSS structure and it's running this back.

And so it's already encountering some errors with the way that it's written the code.

And it's just going ahead and fixing them for us.

I've told people about automation that 90% of the job is troubleshooting.

And now if anti-gravity or these vibe coding platforms can take away that 90% of the job that's troubleshooting, think about how much freaking time you're going to save.

And now we can see here that in a couple of minutes, the script ran successfully.

It fetched 303 relevant, unique posts.

It does say the Google API key is missing, which to me stands out to me because I don't know why we need to use Google at all for this.

So I can go ahead and ask the system.

But it basically went ahead and created all of the directives for us, all of the directions, all of the execution scripts.

It says, and then it gives us the next steps as the human, right?

So add your Google API key.

If you have it, you can add it here.

Again, I don't know what we need this for.

So I just said, I don't know what we need the Google API key for.

Can we do this without?

So I'm going to send this off and anti-gravity is going to research.

It's going to think through my question.

It's going to figure out why we're using the API key in the first place.

And then it's going to see if there's some sort of workaround.

My guess is why it wants to use the Google API key is to actually have the LLM process the data.

So what we can do is we can come over here and we can go to Google AI Studio.

So AI studio.google.com.

And if we come down here, you can go into get API key.

You literally just log into this with your own Google account.

Go ahead, click create API key.

And I can just call this anti-gravity and December 31st.

Happy New Year's, everybody.

Just add this to a single project that you've created.

Go ahead, click create key, and then just copy and paste this over into anti-gravity.

So now we've created this new key here.

I'm going to go ahead and click copy API key.

I'm going to come over into anti-gravity.

I just said, here's my Google Studio API key to use for the LLM.

And then paste in your API key.

What's crazy is this actually set.

It updated the system to work without the Google API key.

It's going to filter this based off a couple of different things.

And then it actually even generated a report for us.

So I can just click open on this and we can see that this generated a report for us.

We can see here, built my first complex N8N workflow, a PR crisis detection engine.

Would love to hear your feedback.

So it just went ahead and brought in that text.

And we can just go ahead here.

And this is direct link to the post itself.

Now, this obviously had a lot of posts.

So my guess is what this did is this just grabbed everything and put it all into one

because we actually didn't have AI to process the data.

One other thing that's cool is these are real files on your computer.

So we can go to our desktop.

We can open Reddit News and you can see all of these folders here.

So these are all of the folders and files that the system has created.

So you can package these up.

You can upload these to GitHub.

You can share this with friends and family and they can run the system themselves.

So what you can do is that's cool.

You can say, I want to end up with 10 of the most interesting relevant posts for my audience.

And you can actually just send this off as a comment.

So anti-gravity is basically going to get to your feedback, like when it has a pause in the process.

So you can actually make improvements while it's building.

Also, because I thought it would be fun.

I said for each of the 10 articles, I want to create an infographic using Gemini 3 Pro Preview

or Nano Banana Pro in a style relevant to the text.

Now I'm going to show you the real power of these agentic workflows.

There are two main things that I'm showing you here.

One is if we wanted to create images now for each of these posts in an NNN workflow like this,

we'd have to figure out how to do that.

So we'd have to attach more nodes to this.

We'd have to figure out all the properties.

We'd have to figure out all the different text fields and all the links and any of the API calls.

And it just gets confusing and cumbersome really quickly, especially if you've never done this before.

What's cool about anti-gravity is it's just going to go and research the internet to figure out how to do that.

If you want to give it a little help, what's also even cooler about anti-gravity is I can literally just go ahead

and copy a link to the Nano Banana Pro API documentation.

And I can say, here is the Nano Banana Pro API docs.

And I can just paste a link in here.

And it's now going to read through all of the documentation, figure out what works and what doesn't work,

how to actually access these systems.

And it's going to build it automatically for you without you having to know what any of this means.

So that right there is a massive, massive time saver.

One of the other things that's really powerful about these agents is their ability to parallel process.

So I'm going to show you how we can make this go even faster.

You can literally just tell the agent, can we make this faster?

Can we optimize this?

Can we parallel process or actually do multiple things simultaneously?

Generally, the answer is yes.

And anti-gravity will figure out how to make your system as fast as it possibly can be,

which, again, saves you hours and hours of troubleshooting inside of a platform like N8N,

trying to optimize and make it faster, which you may or may not know how to do.

So this is really cool.

We can see what the system has done now.

It's basically implemented a top 10 workflow to pull out the top 10 most relevant, most engaging posts.

It links to them here.

It creates a little insight about them.

And then it goes ahead and creates an infographic image for each one.

Very, very cool.

So I just said, the image generation seemed to take a long time.

Can we process them in parallel?

And then infographics should also have text relevant to the post.

Because what I found is that most of these infographics here are really just images

and don't include a ton of text explaining what they are.

And then I just went ahead and said, how else can we make the system faster?

I don't know the answer to this, but I bet anti-gravity does.

So we can see here that the system is thinking its primary concern is the slow image generation.

It's exploring parallel processing techniques to significantly improve speed.

It wants to improve the efficiency by identifying key bottlenecks,

hindering system performance, which is pretty cool.

So it's actually going to be able to parallel process a lot of these different posts all at once

and then improve the image prompts to specifically reference the text from the actual posts.

So this finished in just a minute or two.

And again, you can go ahead and like do some other work in the background while this system is processing.

So basically it went ahead and now it's processing all of our feeds in parallel simultaneously.

It's evaluating them in parallel.

It's enriching them in parallel.

And it's generating all of the images also in parallel.

And now you can see it actually took our advice about including some of the text inside of these infographics.

So these are much more meaningful.

And now I just went ahead and said, clean up our temp file so we can run the system fresh

so that we can test this out from end to end and see how fast we can do this.

And so now if you don't know how to actually run this,

you can just say, what do I type to run this in terminal so I can run this down here without having to use my credits.

So now we can see we've actually opened up our project, which we have right here.

And then we just have this, which I can copy over here.

And this is where the terminal comes into play because I can paste this in and go ahead and click enter.

And this is just going to go ahead and it's running all of the code for us and doing the entire process for us.

So we can see that it found all of the relevant posts in the last 72 hours.

It found 21 that have the most upvotes.

It's processing all of the batches for us in parallel.

It's going to summarize them and it's going to generate images for everything that it found.

So we can click through here on all of our brand new images and infographics that we generated.

Again, it's up to you the type of graphics that you want to make.

And it went ahead and created this markdown file for us with all the insights and the links to each one of the posts and even a little score for us.

So now anytime you want, you can just go ahead and run the script to get the latest news and some cool infographics made for you.

The two biggest takeaways are the speed and the really lack of technical knowledge that you need in order to build pretty much anything.

Remember the DOE framework, the directives, orchestration and execution layers, as well as using this agent's MD file.

Again, there's a link in the description to basically set up your entire anti-gravity environment and build anything that you want pretty much on autopilot.

If you thought this video was cool, please subscribe to the channel.

If you want to learn how to use agentic workflows to build an audience and drive traffic to your offer, just check the link in the description to join me in school.

Clips

No clips yet

Select text above to create your first clip

Loading connections...

Reflect

What part of this is most worth remembering a month from now?

Press Cmd+Enter to submit