BUILDING WITH ANTIGRAVITY 2.0: THE UPDATES YOU NEED TO KNOW

Building With Antigravity 2.0: The Updates You Need to Know

Transcript

Dalton (00:00:00.000)

But anti-gravity is a really interesting tool and something

Dalton (00:00:02.400)

that I have become very fond of.

Dalton (00:00:04.960)

It was very fond of Gemini command line.

Dalton (00:00:07.520)

So these updates of anti-gravity, it's really interesting how good the

Dalton (00:00:14.350)

tool is and the new model update of 3.5 flash, it's phenomenal.

Dalton (00:00:21.220)

It's one-shotted some very difficult tasks that I'd given it to just

Dalton (00:00:26.580)

trial things out before we had this episode.

Dalton (00:00:28.740)

And I was thoroughly impressed with some of the stuff that it had

Dalton (00:00:31.820)

built pretty much from scratch with limited feedback.

Dalton (00:00:38.700)

Welcome to BenchStep Podcast.

Dalton (00:00:39.980)

We discuss entrepreneurship, mission trends, and the occasional

Dalton (00:00:43.220)

book review.

Dalton (00:00:44.300)

I'm your host Dalton Anderson.

Dalton (00:00:46.980)

In this episode, we're going to be doing a deep dive into

Dalton (00:00:50.780)

anti-gravity 2.0 and anti-gravity command line.

Dalton (00:00:54.900)

These were recently announced products from Google's Google

Dalton (00:00:59.140)

Cloud Next that I talked about a couple of weeks ago in a

Dalton (00:01:03.500)

podcast episode.

Dalton (00:01:04.660)

If you would like a full deep dive breakdown of what was announced

Dalton (00:01:08.060)

and why is it important, please listen to that episode.

Dalton (00:01:10.980)

In this episode, we're going to be discussing these agentic

Dalton (00:01:13.180)

coding tools and how to get the most out of them and why I'm

Dalton (00:01:16.900)

so fond of them.

Dalton (00:01:18.470)

So first off, before we dive into the tool and share my

Dalton (00:01:21.990)

screen and get into a live demo of some stuff that has been

Dalton (00:01:24.790)

built, I first want to backtrack and discuss anti-gravity 1.0

Dalton (00:01:30.870)

and 2.0 and then Gemini command line and then anti-gravity

Dalton (00:01:35.910)

command line.

Dalton (00:01:36.510)

First, let's start with anti-gravity.

Dalton (00:01:39.140)

So anti-gravity 1.0 is a fork, which is basically a copy of

Dalton (00:01:44.500)

this popular coding idol called VS Code.

Dalton (00:01:47.900)

VS Code is an open source coding editor, idol, whatever

Dalton (00:01:52.700)

you want to call it.

Dalton (00:01:54.220)

That was originally created by Microsoft and has been supported

Dalton (00:02:00.100)

by Microsoft forever.

Dalton (00:02:02.460)

Like when I was in college, I used it and it's a phenomenal

Dalton (00:02:05.580)

tool.

Dalton (00:02:05.940)

It's got the best resources.

Dalton (00:02:08.180)

Everybody knows how to use it.

Dalton (00:02:09.220)

It's standard, it's open source as well.

Dalton (00:02:11.260)

And so there has been some popular takes where people have

Dalton (00:02:14.060)

been taking this coding editor and taking that and flipping it

Dalton (00:02:20.050)

to make cursor, which is another agentic coding platform.

Dalton (00:02:26.740)

And then there is anti-gravity, which was a fork of VS Code.

Dalton (00:02:30.260)

So same thing, a copy of VS Code to create this new

Dalton (00:02:33.180)

product.

Dalton (00:02:34.220)

And the innovation of anti-gravity was really this

Dalton (00:02:39.010)

thing called mission control.

Dalton (00:02:41.130)

And mission control was an extension of this idol.

Dalton (00:02:45.370)

And an extension in this instance is a new window that

Dalton (00:02:48.650)

opens up when you say open mission control and mission

Dalton (00:02:52.370)

control allowed you to manage multiple instances of agents.

Dalton (00:02:57.170)

And these agents could create sub-agents and one workflow

Dalton (00:03:01.130)

that you had ongoing could have 20 or so agents running

Dalton (00:03:06.290)

at the same time.

Dalton (00:03:07.170)

And you could have up to 15 workflows or instances open

Dalton (00:03:10.970)

at the same time, as long as you had enough local or

Dalton (00:03:13.890)

cloud compute and enough tokens to support that.

Dalton (00:03:17.940)

That's either your usage or your wallet, either or, because

Dalton (00:03:22.380)

it can get expensive if you're not optimized correctly.

Dalton (00:03:28.900)

So anti-gravity 1.0 is a extension.

Dalton (00:03:34.200)

Whereas anti-gravity 2.0 is a platform for agent

Dalton (00:03:38.160)

orchestration, whereas before it was a new window.

Dalton (00:03:41.680)

Now it doesn't even have an idle part of its instance.

Dalton (00:03:45.920)

It's a separate thing.

Dalton (00:03:47.520)

It's a separate product.

Dalton (00:03:48.720)

You have to download a separate anti-gravity idol.

Dalton (00:03:51.080)

Whereas anti-gravity 2.0, it's just the orchestration

Dalton (00:03:54.720)

part, it does everything on its own.

Dalton (00:03:56.240)

You don't see it, it can provide you demos and

Dalton (00:03:58.600)

screenshots and videos, but you're not actually in

Dalton (00:04:02.260)

the code base if you don't want to be.

Dalton (00:04:05.390)

Which is completely different than how things would

Dalton (00:04:08.870)

work typically to get the most out of these tools.

Dalton (00:04:12.780)

There is terminal based coding, which has been

Dalton (00:04:15.540)

around for, I don't know, a couple of years now.

Dalton (00:04:19.540)

There is cloud code, there's codecs, and then there

Dalton (00:04:22.620)

was Gemini command line until it was retired.

Dalton (00:04:26.060)

And now it's anti-gravity command line.

Dalton (00:04:29.270)

The main difference between anti-gravity command

Dalton (00:04:31.950)

line and anti-gravity I think is the speed and

Dalton (00:04:36.870)

architecture of the tool.

Dalton (00:04:38.990)

So anti-gravity command line is built in Golang.

Dalton (00:04:43.070)

Golang is world-class at information orchestration.

Dalton (00:04:47.510)

So sending a package of data from one place to

Dalton (00:04:51.270)

another place and receiving it back and sending

Dalton (00:04:53.350)

it however the pipeline is, it's the best.

Dalton (00:04:58.460)

This whole reason Google built it was to be able

Dalton (00:05:00.820)

to send data from one place to another place.

Dalton (00:05:03.100)

And that's all it's really good at.

Dalton (00:05:04.340)

I mean, it's good at other things, but that's

Dalton (00:05:06.020)

the main thing.

Dalton (00:05:06.780)

It's world-class at data orchestration and sending

Dalton (00:05:11.750)

things back and forth and receiving information.

Dalton (00:05:15.320)

It's by far the best.

Dalton (00:05:16.280)

And that's what anti-gravity command line is

Dalton (00:05:19.600)

built in, which is great.

Dalton (00:05:22.220)

And then there's some other stuff that's really

Dalton (00:05:23.580)

cool about anti-gravity command line is it has a

Dalton (00:05:28.940)

full-on toolkit built into it.

Dalton (00:05:30.820)

I mean, Gemini command line had some of this

Dalton (00:05:33.140)

stuff, but it wasn't as straightforward as

Dalton (00:05:35.820)

anti-gravity.

Dalton (00:05:36.860)

I know people like Gemini command line, but

Dalton (00:05:39.540)

people just don't like change.

Dalton (00:05:40.940)

And there are some features that aren't within

Dalton (00:05:44.580)

anti-gravity command line that are built out

Dalton (00:05:46.340)

in the 1.0 release.

Dalton (00:05:47.940)

But in a general sense, the process is faster.

Dalton (00:05:52.870)

It's better.

Dalton (00:05:53.390)

The UI is better.

Dalton (00:05:54.310)

The functionality is more smooth.

Dalton (00:05:56.590)

It's more straightforward.

Dalton (00:05:57.910)

It's more user-friendly overall.

Dalton (00:05:59.470)

It's a better product.

Dalton (00:06:00.430)

So I think it's a win-win for everybody.

Dalton (00:06:04.580)

So those are the two tools.

Dalton (00:06:07.420)

So why don't I just share my screen and show

Dalton (00:06:10.380)

you what anti-gravity 2.0 looks like.

Dalton (00:06:15.220)

And then we could talk about something that I

Dalton (00:06:18.100)

decided to build and walk through some of the

Dalton (00:06:22.260)

features because the features set is kind of

Dalton (00:06:24.980)

similar within the anti-gravity platform, I

Dalton (00:06:30.340)

call it, and then anti-gravity command line.

Dalton (00:06:34.050)

So share my screen.

Dalton (00:06:36.330)

Okay.

Dalton (00:06:36.570)

So within here, you have all your projects and

Dalton (00:06:40.170)

your projects are structured wherever you want

Dalton (00:06:43.620)

them to be, but typically people put them in

Dalton (00:06:46.740)

their home directory, which is depending on where

Dalton (00:06:50.060)

your system lives.

Dalton (00:06:51.980)

Like if you're in Linux, Mac, or Windows, it's

Dalton (00:06:54.580)

in your user.

Dalton (00:06:55.580)

So it's in like your root of your user, which is

Dalton (00:06:57.700)

referred to as your home directory.

Dalton (00:06:59.860)

And so it would be whatever your username is, a

Dalton (00:07:02.460)

folder, it's lowercase, dev, and then you put

Dalton (00:07:06.900)

your project in there.

Dalton (00:07:09.000)

The historical context of that is back in the

Dalton (00:07:12.520)

day, the way that you ran code was that you'd

Dalton (00:07:16.630)

have the source and then the file.

Dalton (00:07:19.510)

And then it was like the, like this user slash

Dalton (00:07:22.950)

source in within your computer, because not

Dalton (00:07:26.910)

everyone had home computers.

Dalton (00:07:27.910)

So this really wasn't a thing until Linux came

Dalton (00:07:30.310)

out and then Linux had users.

Dalton (00:07:33.330)

And then within there, you had sandbox your

Dalton (00:07:37.130)

system.

Dalton (00:07:37.930)

And so you needed to switch to this user split

Dalton (00:07:41.930)

out because everything was sandbox.

Dalton (00:07:43.610)

So that's why it moved to the home directory.

Dalton (00:07:47.960)

It's very interesting.

Dalton (00:07:49.000)

And then the dev folder thing is just like a

Dalton (00:07:53.600)

natural progression of development.

Dalton (00:07:55.960)

And so it went from source, it went to dev, and

Dalton (00:07:59.600)

that's, that's kind of where it lives.

Dalton (00:08:01.400)

And then all of your coding stuff and your

Dalton (00:08:03.640)

agentic stuff lives in your home directory.

Dalton (00:08:06.120)

And that's just how it is because it's easier

Dalton (00:08:09.040)

to sandbox and provide a ease of permissions

Dalton (00:08:13.460)

within that folder.

Dalton (00:08:14.940)

And so it's just much easier.

Dalton (00:08:17.830)

Okay.

Dalton (00:08:18.230)

So this is the UI in here.

Dalton (00:08:20.630)

So I was working on something that I was

Dalton (00:08:23.070)

inspired up, inspired from me trying to build

Dalton (00:08:27.670)

content for the podcast venture step.

Dalton (00:08:29.950)

And I was thinking about how can I create

Dalton (00:08:33.780)

something that's useful to myself, which

Dalton (00:08:36.540)

isn't wasteful, but also interesting to you.

Dalton (00:08:40.900)

And I was like, thinking about all these

Dalton (00:08:43.060)

things that last time I did some kind of

Dalton (00:08:44.820)

claims thing related to insurance, which maybe

Dalton (00:08:48.780)

is a little abstract for a wide audience, but

Dalton (00:08:52.260)

people can get behind building a system

Dalton (00:08:54.780)

for the podcast.

Dalton (00:08:56.780)

You can reflectively look at it and think

Dalton (00:08:58.900)

about your hobby or things that you're

Dalton (00:09:00.780)

trying to do.

Dalton (00:09:01.460)

And it's like, wow, like if you could just do

Dalton (00:09:03.580)

that by asking questions, like how can

Dalton (00:09:05.260)

I do my thing?

Dalton (00:09:06.660)

Like I've always been interested in X.

Dalton (00:09:08.780)

How can I take the content that Dalton

Dalton (00:09:11.620)

created here and follow this framework?

Dalton (00:09:15.220)

In this episode, I will be providing the

Dalton (00:09:18.340)

same framework.

Dalton (00:09:21.550)

I just can't help you do the questions.

Dalton (00:09:23.510)

You've got to do the what I'll can give

Dalton (00:09:27.950)

you the how, and I can give you the do

Dalton (00:09:30.750)

kind of thing where I can give you the

Dalton (00:09:32.430)

framework and the execution.

Dalton (00:09:34.550)

You just have to download it from the

Dalton (00:09:35.630)

internet and installing your computer, which

Dalton (00:09:38.030)

I'll give you the commands on GitHub and

Dalton (00:09:41.150)

I'll link it in the show notes.

Dalton (00:09:42.310)

But that's a thought process.

Dalton (00:09:44.070)

And the way that I was inspired from this

Dalton (00:09:46.510)

and maybe if I backtrack a little bit, the

Dalton (00:09:48.950)

way that I was inspired to build this

Dalton (00:09:50.950)

platform was I was building out this

Dalton (00:09:54.550)

LinkedIn carousel and on LinkedIn, the

Dalton (00:09:57.710)

go-to to really push for engagement is

Dalton (00:10:01.830)

these LinkedIn carousels.

Dalton (00:10:03.030)

People really like that because when you

Dalton (00:10:05.230)

put a PDF on LinkedIn, it splits out the

Dalton (00:10:08.390)

pages and so you basically create a

Dalton (00:10:10.870)

slide deck and you have people swipe

Dalton (00:10:13.990)

through that slide deck.

Dalton (00:10:15.990)

But since you're swiping on it, there's a

Dalton (00:10:17.990)

lot of engagement reading from this

Dalton (00:10:20.310)

carousel and people find it interesting.

Dalton (00:10:22.390)

And a lot of times if you're making the

Dalton (00:10:24.470)

time to make a carousel and put it out

Dalton (00:10:28.090)

there and post it and name the file, it's

Dalton (00:10:30.650)

normally high value content, which is

Dalton (00:10:33.730)

where it thrives on LinkedIn.

Dalton (00:10:35.650)

Cause LinkedIn people want to feel like

Dalton (00:10:38.490)

they've got some edge or they're on the

Dalton (00:10:41.130)

forefront and this carousel might

Dalton (00:10:43.330)

make them feel that way.

Dalton (00:10:45.430)

So let me share my screen of what I

Dalton (00:10:47.110)

was able to build in Gemini.

Dalton (00:10:49.950)

So this is Gemini Canvas.

Dalton (00:10:52.670)

And so Canvas can build apps.

Dalton (00:10:54.630)

It can build Word docs.

Dalton (00:10:56.790)

It can make documentation.

Dalton (00:10:58.350)

It could do a whole bunch of stuff.

Dalton (00:10:59.830)

And I think it's a really good tool

Dalton (00:11:02.030)

just to prototype something to really

Dalton (00:11:03.630)

think about where you're at and what you

Dalton (00:11:06.950)

want to do with your idea and just

Dalton (00:11:09.710)

prototype before you do anything serious.

Dalton (00:11:12.630)

And so I was trying to create a carousel

Dalton (00:11:16.830)

and I was thinking about the prompts and

Dalton (00:11:19.590)

the thought process and I threw it

Dalton (00:11:21.470)

into a couple of things and I was like,

Dalton (00:11:23.510)

okay.

Dalton (00:11:24.190)

And then I was like, all right, let's

Dalton (00:11:25.390)

try to just make it like, who knows.

Dalton (00:11:27.470)

And then it just made this.

Dalton (00:11:29.310)

And so I thought it was pretty good.

Dalton (00:11:31.430)

There's not as much editing as I'd

Dalton (00:11:32.990)

like, but this is a really good start

Dalton (00:11:35.230)

to where you can add slides, you can

Dalton (00:11:39.670)

change the template of your slides.

Dalton (00:11:43.580)

So you can make it a quote.

Dalton (00:11:45.060)

You can split it out for contrast.

Dalton (00:11:48.260)

You can make it a primary hook.

Dalton (00:11:50.140)

You can change the header and title,

Dalton (00:11:52.260)

but you just can't change the body.

Dalton (00:11:54.060)

You can change the colors, but there's

Dalton (00:11:56.940)

not as much detail as I'd like.

Dalton (00:11:59.500)

But overall, this is really nice

Dalton (00:12:02.340)

because then you can pretend like what

Dalton (00:12:04.220)

it would look like on LinkedIn.

Dalton (00:12:06.500)

And it's just great.

Dalton (00:12:08.660)

And you can flip through.

Dalton (00:12:09.740)

So this is what it would look like to

Dalton (00:12:11.100)

a user when they're on the platform.

Dalton (00:12:13.940)

So I thought that was awesome.

Dalton (00:12:15.260)

I was like, wow, what if I could do

Dalton (00:12:17.740)

this, but on higher scale?

Dalton (00:12:19.980)

And one thing I have an issue with

Dalton (00:12:22.700)

is some of the things that I'm trying

Dalton (00:12:24.060)

to do with venture step, acquire too

Dalton (00:12:26.180)

many tokens, so on this site or on

Dalton (00:12:29.460)

a website like chat.com or for a

Dalton (00:12:32.580)

chat GPT or anthropic or Gemini,

Dalton (00:12:36.890)

you only get a certain amount of

Dalton (00:12:39.170)

tokens on the website.

Dalton (00:12:40.890)

If you want a lot of tokens, you need

Dalton (00:12:43.570)

to go onto the API and the API costs

Dalton (00:12:47.970)

money, it's not part of your normal

Dalton (00:12:51.290)

subscription, et cetera, depending on

Dalton (00:12:53.250)

your subscription, like Google has a

Dalton (00:12:54.850)

set up to say that you get $100 of

Dalton (00:12:58.370)

Google cloud credit and your Google

Dalton (00:13:00.210)

cloud credit is also part of the API.

Dalton (00:13:02.810)

So you do technically get $100 in

Dalton (00:13:05.130)

credits. I think they got rid of that.

Dalton (00:13:06.930)

I'm not sure now that I'm saying it

Dalton (00:13:08.770)

because I know that they lowered

Dalton (00:13:09.890)

this cost of the subscription by like

Dalton (00:13:12.130)

$150. So I don't know if they're

Dalton (00:13:14.450)

still providing that large of a

Dalton (00:13:15.610)

credit. Regardless, what I'm saying

Dalton (00:13:19.500)

is if you want to do a lot of token

Dalton (00:13:22.180)

work, whether it requires a lot of

Dalton (00:13:24.220)

context, then you have to use the

Dalton (00:13:26.720)

API. So it's like, okay, well,

Dalton (00:13:28.280)

that's awesome. And then I'm getting

Dalton (00:13:29.920)

help. Like venture step is not just

Dalton (00:13:32.520)

me anymore. And it's easy for me to

Dalton (00:13:37.040)

manage this content, but now I'm

Dalton (00:13:40.200)

trying to make LinkedIn care cells.

Dalton (00:13:41.960)

I'm trying to make thumbnails.

Dalton (00:13:43.520)

I'm trying to upgrade the content.

Dalton (00:13:45.760)

And so for that to work, you need

Dalton (00:13:49.480)

to have some kind of prompts and

Dalton (00:13:52.560)

then you need to have that person

Dalton (00:13:55.560)

who's helping you. They need to have

Dalton (00:13:57.360)

access to AI, and that costs money.

Dalton (00:14:01.380)

And so you either have the option

Dalton (00:14:03.060)

to make an account and give them a

Dalton (00:14:05.540)

subscription, or you can create a

Dalton (00:14:07.620)

platform that has this API

Dalton (00:14:10.860)

integrated and they can use the

Dalton (00:14:12.460)

platform and call on the API, which

Dalton (00:14:14.740)

I think is easiest because I needed

Dalton (00:14:16.420)

anyways, because some of the work

Dalton (00:14:17.540)

I wanted to do wasn't available

Dalton (00:14:19.860)

because I needed more tokens.

Dalton (00:14:21.900)

And just to write an example, like

Dalton (00:14:24.420)

a blog, a blog is 2000 to 8,000

Dalton (00:14:29.900)

words. And then from there, it takes

Dalton (00:14:32.860)

the transcription. What I would

Dalton (00:14:34.620)

really want is a blog and then

Dalton (00:14:36.940)

underneath the blog, it's a

Dalton (00:14:38.060)

transcription of the podcast

Dalton (00:14:39.620)

episode. What that will do with

Dalton (00:14:41.540)

that will provide a massive, massive

Dalton (00:14:44.860)

SEO, slash geo push, where it's

Dalton (00:14:47.980)

just a massive amount of content.

Dalton (00:14:49.940)

Users don't want to read the

Dalton (00:14:51.820)

transcription, the machines do.

Dalton (00:14:54.460)

Users want to read the blog, but

Dalton (00:14:55.820)

if you can put them all into one,

Dalton (00:14:57.820)

then it makes a lot of sense.

Dalton (00:14:59.580)

But I personally don't want to

Dalton (00:15:00.980)

format the transcription.

Dalton (00:15:02.460)

AI can do that just fine.

Dalton (00:15:03.820)

And I personally think that if

Dalton (00:15:05.780)

it could take the transcription,

Dalton (00:15:07.100)

format it, and provide some

Dalton (00:15:08.580)

context, it can create a good

Dalton (00:15:11.140)

enough blog, and then it can be

Dalton (00:15:13.780)

edited by a human, where it gets

Dalton (00:15:16.420)

it to 90%, 85% or something, and

Dalton (00:15:18.700)

the human could spend an hour

Dalton (00:15:20.060)

getting the blog in a good state.

Dalton (00:15:21.460)

So the implementation plan, it

Dalton (00:15:23.740)

describes what it wants to do

Dalton (00:15:26.140)

with what you set.

Dalton (00:15:27.860)

So it's taking all your

Dalton (00:15:28.940)

information and say, okay,

Dalton (00:15:30.020)

here's what I'm thinking.

Dalton (00:15:31.700)

And it's similar to Antigravity

Dalton (00:15:33.100)

1.0, you can put comments and

Dalton (00:15:35.700)

it will change the approach and

Dalton (00:15:37.500)

it will change it live.

Dalton (00:15:38.580)

So even if it's already running,

Dalton (00:15:40.460)

it's already done that, you can

Dalton (00:15:41.420)

make a comment and it will

Dalton (00:15:42.380)

submit it and it will make the

Dalton (00:15:44.140)

change. One thing that's a

Dalton (00:15:46.570)

little different with Antigravity

Dalton (00:15:49.130)

2.0 is you can manage your

Dalton (00:15:51.050)

permissions by project.

Dalton (00:15:53.330)

And so I have it in turbo mode,

Dalton (00:15:55.850)

which means it just does what it

Dalton (00:15:58.930)

does, disables all safety

Dalton (00:16:00.970)

barriers for maximum iteration

Dalton (00:16:03.650)

velocity. So I'm good with it.

Dalton (00:16:10.160)

I've always been a YOLO guy.

Dalton (00:16:13.080)

What can I say?

Dalton (00:16:13.880)

But this is their version of

Dalton (00:16:15.680)

YOLO and they feel like this

Dalton (00:16:17.760)

was new because prior it was

Dalton (00:16:20.560)

something else. And then I was

Dalton (00:16:23.080)

saying, hey, I wanted to do

Dalton (00:16:25.560)

YOLO and I set up all the

Dalton (00:16:27.400)

settings and just wasn't doing

Dalton (00:16:29.000)

it on the first couple days it

Dalton (00:16:30.360)

was out. And I was like, wow,

Dalton (00:16:32.040)

this sucks because I've got to

Dalton (00:16:33.400)

approve everything, which is a

Dalton (00:16:35.680)

lot when you've got multiple

Dalton (00:16:37.120)

things running at the same

Dalton (00:16:38.360)

time. Like if you had three

Dalton (00:16:39.840)

different projects going at the

Dalton (00:16:40.760)

same time and you've got to

Dalton (00:16:41.640)

approve everything, you were

Dalton (00:16:43.240)

just sitting there like

Dalton (00:16:44.280)

nonstop, just do, do, do, do.

Dalton (00:16:47.280)

And you can't get up like you

Dalton (00:16:48.400)

just, you're just a clicking

Dalton (00:16:49.680)

monkey. And I don't like that.

Dalton (00:16:51.600)

And it doesn't seem very

Dalton (00:16:52.680)

efficient because I want this

Dalton (00:16:54.200)

to kick off. So one of the

Dalton (00:16:57.020)

things that's special about

Dalton (00:16:58.060)

Antigravity 2.0 is that you can

Dalton (00:17:00.140)

set up your permissions by

Dalton (00:17:02.980)

project. So this project could

Dalton (00:17:04.660)

be different than another

Dalton (00:17:05.980)

project. It has its own

Dalton (00:17:08.060)

customizations and then you

Dalton (00:17:10.660)

can just specify the different

Dalton (00:17:15.230)

importances of what you think.

Dalton (00:17:17.350)

If you want it to have access

Dalton (00:17:19.270)

to certain things, you can

Dalton (00:17:20.270)

remove it. But then for a

Dalton (00:17:21.390)

different project, it could be

Dalton (00:17:22.310)

contributed completely

Dalton (00:17:23.270)

differently. So that's cool.

Dalton (00:17:25.110)

And then you can set up your

Dalton (00:17:28.430)

permissions and different

Dalton (00:17:29.910)

things. And it's similar to

Dalton (00:17:31.750)

Antigravity 1.0 where you can

Dalton (00:17:33.630)

edit the MCPs available. All

Dalton (00:17:36.940)

that stuff is pretty similar.

Dalton (00:17:38.100)

And then you can create custom

Dalton (00:17:39.460)

stuff on your own, which we'll

Dalton (00:17:41.380)

talk about later on in the

Dalton (00:17:42.420)

episode. If this thing ever

Dalton (00:17:44.460)

stops running, and we might

Dalton (00:17:46.140)

have to flip over and flip

Dalton (00:17:47.660)

backwards and just go terminal.

Dalton (00:17:50.140)

Alright, so while this is

Dalton (00:17:51.020)

running, we're going to walk

Dalton (00:17:52.020)

through terminal. So within

Dalton (00:17:58.020)

the terminal, you have access

Dalton (00:18:01.740)

to the Antigravity command

Dalton (00:18:04.460)

line. You've got to download

Dalton (00:18:05.580)

it though. So there is AGY and

Dalton (00:18:08.980)

I'll pull up. So this is

Dalton (00:18:09.780)

going to be Antigravity command

Dalton (00:18:10.940)

line. I have an alias set up

Dalton (00:18:12.660)

for it to be automatically set

Dalton (00:18:14.900)

up to be YOLO. So I have AGY

Dalton (00:18:19.530)

hyphen you only live once.

Dalton (00:18:23.810)

And this is going to be the

Dalton (00:18:26.620)

same thing as AGY, but this is

Dalton (00:18:30.740)

going to be AGY with YOLO

Dalton (00:18:33.940)

because I'm a YOLO guy. And

Dalton (00:18:35.460)

you could set it up however

Dalton (00:18:36.460)

you want. I figured it'd be

Dalton (00:18:38.220)

better to make it very clear

Dalton (00:18:40.180)

that I am in YOLO mode at

Dalton (00:18:41.660)

all times. So let's set it

Dalton (00:18:43.620)

up to look like it's in YOLO

Dalton (00:18:45.340)

mode. So YOLO has a couple

Dalton (00:18:47.880)

things that you need to be

Dalton (00:18:49.640)

aware of. YOLO has full

Dalton (00:18:52.500)

rights. It just does

Dalton (00:18:53.740)

whatever it wants to do. That

Dalton (00:18:55.420)

can be dangerous in certain

Dalton (00:18:57.460)

scenarios. So you have to be

Dalton (00:18:58.660)

careful on what you're doing

Dalton (00:19:00.940)

and what it has access to.

Dalton (00:19:02.660)

And so I give it YOLO

Dalton (00:19:03.740)

permission, full access to

Dalton (00:19:06.020)

the dev folders and within

Dalton (00:19:09.180)

the dev folders is full

Dalton (00:19:10.180)

access. And it's sandboxed.

Dalton (00:19:12.740)

Otherwise it doesn't have

Dalton (00:19:14.940)

full reign to do whatever it

Dalton (00:19:16.780)

wants. So that's important.

Dalton (00:19:19.180)

And then another thing that's

Dalton (00:19:20.140)

important about YOLO is you

Dalton (00:19:21.980)

can get caught up into these

Dalton (00:19:24.380)

agentic loops. And if you're

Dalton (00:19:26.140)

in YOLO, it doesn't

Dalton (00:19:27.820)

necessarily know when to

Dalton (00:19:28.660)

stop. So you need to set up

Dalton (00:19:32.220)

maximum iterations and I have

Dalton (00:19:34.020)

it set to 25. But if it

Dalton (00:19:36.980)

gets stuck, you need

Dalton (00:19:38.180)

somebody to kill, like have a

Dalton (00:19:39.580)

kill switch. And if you

Dalton (00:19:42.260)

aren't there to kill the

Dalton (00:19:43.420)

switch, you've got to have

Dalton (00:19:44.580)

some kind of automated kill

Dalton (00:19:45.780)

switch, which that's what

Dalton (00:19:46.980)

the max iterations does.

Dalton (00:19:48.540)

But that's YOLO. But this

Dalton (00:19:49.900)

is almost a Gemini

Dalton (00:19:50.820)

command line. Antigravity

Dalton (00:19:51.860)

command line, you can access

Dalton (00:19:53.740)

things with the forge slash.

Dalton (00:19:56.020)

And so there's a couple of

Dalton (00:19:57.580)

cool things that we'll go

Dalton (00:19:58.660)

through. You can see the

Dalton (00:19:59.980)

agents that you have

Dalton (00:20:00.700)

corrected or have fun. If

Dalton (00:20:02.820)

you have agents made, you

Dalton (00:20:05.100)

won't be able to see your

Dalton (00:20:06.100)

agents within here unless you

Dalton (00:20:08.420)

have something going and

Dalton (00:20:09.740)

not all instances will they

Dalton (00:20:11.260)

activate your agents. And if

Dalton (00:20:13.380)

they're sub agents, you

Dalton (00:20:14.500)

can't see them at all. So

Dalton (00:20:17.550)

you won't be able to see

Dalton (00:20:18.230)

the director slash like

Dalton (00:20:19.630)

coordinator agents. And it

Dalton (00:20:21.350)

kind of depends on what

Dalton (00:20:22.270)

you're doing. The way that

Dalton (00:20:24.190)

I have it set up, it has

Dalton (00:20:25.790)

three agents. And we'll go

Dalton (00:20:28.790)

over that, but you can't

Dalton (00:20:30.190)

see them unless they're

Dalton (00:20:30.950)

running. So just keep that

Dalton (00:20:32.230)

in mind. And then there is

Dalton (00:20:35.590)

the artifact. If you have

Dalton (00:20:36.870)

something ongoing, you can

Dalton (00:20:38.470)

view the artifacts similar

Dalton (00:20:40.230)

to how you can view the

Dalton (00:20:41.070)

artifacts within here. And

Dalton (00:20:43.190)

then by the way, it's

Dalton (00:20:44.470)

quite cool. And then the

Dalton (00:20:48.070)

context is nice. And I

Dalton (00:20:50.030)

also like the quota, you

Dalton (00:20:51.750)

can type stuff, but there's

Dalton (00:20:54.020)

quota and you can see how

Dalton (00:20:55.180)

there's grill me or

Dalton (00:20:56.020)

schedule. It's all that

Dalton (00:20:56.940)

stuff, but the quota or

Dalton (00:20:58.970)

usage, you can see how

Dalton (00:21:01.810)

many tokens you have

Dalton (00:21:02.890)

available. So I recently

Dalton (00:21:04.970)

ran out of tokens earlier

Dalton (00:21:06.170)

today, like 30 minutes

Dalton (00:21:08.160)

before we had this podcast

Dalton (00:21:10.560)

episode. And so you can

Dalton (00:21:11.880)

see how my tokens are

Dalton (00:21:12.800)

slowly filling up like a

Dalton (00:21:14.240)

little gas bar, but I

Dalton (00:21:15.440)

was bummed out when I

Dalton (00:21:16.200)

ran out of tokens with 3.5

Dalton (00:21:17.440)

flash because it's such a

Dalton (00:21:18.400)

good model. But you can

Dalton (00:21:20.220)

see your usage and your

Dalton (00:21:21.620)

tokens. And then if you

Dalton (00:21:23.100)

run out of tokens, you

Dalton (00:21:24.020)

just type model, but you

Dalton (00:21:25.580)

can just switch to a new

Dalton (00:21:26.660)

model. Just fine. No

Dalton (00:21:28.220)

problem. Right now I'm

Dalton (00:21:29.220)

using Claude Sonnet 4.6

Dalton (00:21:31.900)

thinking and you can

Dalton (00:21:33.700)

just toggle in between

Dalton (00:21:34.660)

models. And if you want

Dalton (00:21:35.980)

to look at what that

Dalton (00:21:38.130)

model is doing when

Dalton (00:21:39.210)

you're running something,

Dalton (00:21:40.250)

you can just use

Dalton (00:21:41.210)

context as a command and

Dalton (00:21:43.170)

then it will show you

Dalton (00:21:43.890)

what's going on. So right

Dalton (00:21:44.690)

now it says zero to

Dalton (00:21:47.010)

250,000 tokens and this

Dalton (00:21:49.490)

and that. So hopefully

Dalton (00:21:51.090)

it doesn't crash my

Dalton (00:21:51.690)

computer. So with this

Dalton (00:21:52.690)

going, you can ask a

Dalton (00:21:54.370)

command, say for such by

Dalton (00:21:55.850)

the way, what are you

Dalton (00:21:56.530)

doing? And then it'll

Dalton (00:21:58.020)

ask you, and then you

Dalton (00:21:58.700)

could say context and

Dalton (00:22:01.140)

then you could see what

Dalton (00:22:02.420)

is taking up the context

Dalton (00:22:03.420)

window. So you could

Dalton (00:22:04.100)

see right now it is

Dalton (00:22:05.740)

filling up. It's right

Dalton (00:22:07.020)

now it's thinking it has

Dalton (00:22:08.220)

41,000 tokens out of

Dalton (00:22:09.940)

the 250,000. It is

Dalton (00:22:13.100)

starting to kick off

Dalton (00:22:14.100)

some of these processes.

Dalton (00:22:15.460)

It's got user messages,

Dalton (00:22:16.700)

user messages are not

Dalton (00:22:17.980)

user messages from you.

Dalton (00:22:19.580)

User messages are the,

Dalton (00:22:22.720)

it's weird the way that

Dalton (00:22:23.960)

they have it, but like

Dalton (00:22:24.800)

user messages and

Dalton (00:22:25.840)

agent responses are part

Dalton (00:22:28.040)

of their correspondent

Dalton (00:22:29.600)

with other agents, if

Dalton (00:22:31.970)

that makes sense. So it

Dalton (00:22:33.050)

just goes and I like

Dalton (00:22:34.890)

this little view and then

Dalton (00:22:36.690)

you can ask it what it's

Dalton (00:22:37.690)

doing and it will tell

Dalton (00:22:38.570)

you and from there it

Dalton (00:22:40.970)

will just run. And so

Dalton (00:22:43.210)

this is just a goal that

Dalton (00:22:44.570)

will run until it's

Dalton (00:22:45.530)

finished. I gave it some

Dalton (00:22:47.090)

pretty big projects. It's

Dalton (00:22:48.530)

probably going to run out

Dalton (00:22:49.570)

of tokens, unfortunately,

Dalton (00:22:50.770)

but it is what it is, you

Dalton (00:22:53.660)

know, but this is a

Dalton (00:22:54.460)

really cool tool and I

Dalton (00:22:56.420)

like it. It's very user

Dalton (00:22:58.140)

friendly. It's super easy

Dalton (00:23:00.870)

to know what's going

Dalton (00:23:02.150)

on and how to use it

Dalton (00:23:03.790)

and it change and it

Dalton (00:23:05.710)

makes a lot of sense. I

Dalton (00:23:07.070)

think it's great. I

Dalton (00:23:09.370)

think that this is a

Dalton (00:23:10.810)

really good first release.

Dalton (00:23:14.160)

I mean, it's an iteration

Dalton (00:23:15.320)

on a previous product.

Dalton (00:23:16.520)

Okay. So with that

Dalton (00:23:17.240)

running, this is done. So

Dalton (00:23:19.040)

let me go back and let's

Dalton (00:23:21.040)

share the website. So the

Dalton (00:23:23.400)

website, remember, it was

Dalton (00:23:24.960)

inspired by this. So if

Dalton (00:23:27.230)

we look at this website

Dalton (00:23:28.510)

and if I reload this,

Dalton (00:23:30.030)

maybe it won't log me

Dalton (00:23:30.990)

out. Okay, cool. So

Dalton (00:23:32.670)

you can come in here and

Dalton (00:23:33.870)

you can put in an

Dalton (00:23:34.430)

episode. And so, of

Dalton (00:23:35.990)

course, I'll select the

Dalton (00:23:37.310)

episode that's related to

Dalton (00:23:38.350)

this episode, which is

Dalton (00:23:40.190)

Google Cloud Next

Dalton (00:23:41.430)

scaling autonomous

Dalton (00:23:42.870)

workflows. And

Dalton (00:23:44.750)

unfortunately for

Dalton (00:23:45.550)

myself, it doesn't look

Dalton (00:23:46.870)

like I can copy and paste

Dalton (00:23:48.990)

I guess we will just do

Dalton (00:23:54.780)

this podcast. I know Apple

Dalton (00:23:56.540)

can allow me to copy and

Dalton (00:23:57.900)

paste. All right. So then

Dalton (00:23:59.580)

we'll take this and so

Dalton (00:24:00.580)

I'm just gonna fill this

Dalton (00:24:01.580)

information out quickly.

Dalton (00:24:04.660)

Boom. And then I'll put

Dalton (00:24:06.380)

in the Apple link course,

Dalton (00:24:08.420)

I'll put in the Spotify

Dalton (00:24:11.100)

link. And then I will

Dalton (00:24:13.380)

take the YouTube link, I

Dalton (00:24:16.400)

guess let's skip it for

Dalton (00:24:17.840)

now. And then I'll take

Dalton (00:24:19.200)

the transcription, which

Dalton (00:24:21.120)

I do not have. All right,

Dalton (00:24:23.690)

let me get that. So once

Dalton (00:24:24.730)

I get the transcription, I

Dalton (00:24:25.770)

put this in here, it's

Dalton (00:24:27.050)

going to do some cool

Dalton (00:24:28.570)

stuff, which I want to

Dalton (00:24:29.970)

share. And this took me

Dalton (00:24:31.450)

about an hour and a half,

Dalton (00:24:33.290)

nothing serious, just vibes.

Dalton (00:24:36.210)

And it's really cool. It

Dalton (00:24:37.730)

probably would take you

Dalton (00:24:38.850)

longer if you were

Dalton (00:24:40.170)

inexperienced. But if

Dalton (00:24:41.170)

you're super experienced

Dalton (00:24:42.090)

with this kind of stuff,

Dalton (00:24:42.890)

then it wouldn't take

Dalton (00:24:45.190)

very long at all. But

Dalton (00:24:46.430)

it's definitely just a

Dalton (00:24:48.390)

viewpoint of like what's

Dalton (00:24:49.630)

possible in a short

Dalton (00:24:51.600)

amount of time and just

Dalton (00:24:53.320)

giving you inspiration of,

Dalton (00:24:55.680)

hey, I've got this side

Dalton (00:24:56.760)

project or this thing

Dalton (00:24:57.840)

that I've liked or I

Dalton (00:24:58.880)

wanted to do. And here's

Dalton (00:25:00.640)

the opportunity to do

Dalton (00:25:01.480)

something and it doesn't

Dalton (00:25:02.480)

necessarily need to make

Dalton (00:25:03.440)

money. It could make your

Dalton (00:25:05.680)

passion easier, right to

Dalton (00:25:08.200)

do, which is what I'm

Dalton (00:25:09.720)

doing. And so download

Dalton (00:25:12.240)

this, this is a

Dalton (00:25:13.000)

transcription and it is

Dalton (00:25:15.320)

only 28,000 characters.

Dalton (00:25:17.560)

The ones with guests could

Dalton (00:25:18.680)

be like 70,000. So I

Dalton (00:25:20.920)

create the episode, episode

Dalton (00:25:22.200)

has been created. I'm

Dalton (00:25:23.560)

missing one of the links,

Dalton (00:25:24.960)

but it's fine. So then

Dalton (00:25:26.080)

we'll go back to the

Dalton (00:25:26.800)

studio desk and then we

Dalton (00:25:28.800)

will change this episode

Dalton (00:25:30.200)

and now it will populate.

Dalton (00:25:32.840)

So I click show notes and

Dalton (00:25:35.850)

then the show notes will

Dalton (00:25:36.730)

populate within here. Boom.

Dalton (00:25:39.090)

And then it's going to

Dalton (00:25:39.770)

pop up on the YouTube

Dalton (00:25:40.730)

description. Boom. And

Dalton (00:25:43.540)

so these platforms have

Dalton (00:25:45.780)

different takes. So YouTube

Dalton (00:25:48.580)

is more of a search

Dalton (00:25:50.820)

engine than it is a social

Dalton (00:25:52.700)

media platform. When

Dalton (00:25:53.860)

people interact with the

Dalton (00:25:54.620)

content, of course, but

Dalton (00:25:55.620)

how do you find the

Dalton (00:25:56.500)

content that's really

Dalton (00:25:57.540)

through searches and

Dalton (00:25:59.260)

Google aggressively uses

Dalton (00:26:01.420)

the YouTube content on

Dalton (00:26:04.170)

Chrome and other search

Dalton (00:26:06.010)

engines. Whereas Spotify

Dalton (00:26:08.800)

and these other types of

Dalton (00:26:10.080)

platforms, majority of

Dalton (00:26:11.000)

people listen to the

Dalton (00:26:11.560)

podcasts on their phone.

Dalton (00:26:14.000)

And so you need

Dalton (00:26:14.840)

something that's phone

Dalton (00:26:15.800)

friendly. And for some

Dalton (00:26:17.200)

reason, this has

Dalton (00:26:17.800)

marked down, which is not

Dalton (00:26:18.840)

supposed to. So that's

Dalton (00:26:20.080)

interesting. That's a

Dalton (00:26:21.120)

violation of the prompt.

Dalton (00:26:22.920)

It wasn't doing that

Dalton (00:26:24.080)

yesterday, so I'm not

Dalton (00:26:25.360)

sure what happened. But

Dalton (00:26:26.840)

this is done via

Dalton (00:26:28.600)

transcription and then

Dalton (00:26:30.360)

these prompts that are

Dalton (00:26:31.320)

in the background within

Dalton (00:26:32.840)

the prompt manager. And

Dalton (00:26:35.190)

then it populates

Dalton (00:26:35.910)

everything. So if you go

Dalton (00:26:37.960)

to social media posts,

Dalton (00:26:38.920)

it do the same thing.

Dalton (00:26:40.080)

And the first thing you

Dalton (00:26:40.920)

got to do is extract that

Dalton (00:26:42.240)

the main takeaways. And

Dalton (00:26:43.720)

so the main takeaways

Dalton (00:26:45.000)

would be, you know,

Dalton (00:26:48.170)

they're launching new

Dalton (00:26:48.850)

products, anti-gravity.

Dalton (00:26:50.290)

Let's see what it says.

Dalton (00:26:50.930)

So it says, a multi-layered

Dalton (00:26:53.760)

approach to agents.

Dalton (00:26:54.840)

Yes, that's true. Key

Dalton (00:26:57.620)

operational. So is its

Dalton (00:26:59.500)

own first customer?

Dalton (00:27:00.860)

Yes, it's different to

Dalton (00:27:01.900)

the live demos. I did

Dalton (00:27:03.140)

say that the company

Dalton (00:27:04.660)

aims to lead the race

Dalton (00:27:05.940)

in AI by focusing on

Dalton (00:27:06.900)

superior context,

Dalton (00:27:07.780)

availability and compute

Dalton (00:27:08.940)

infrastructure,

Dalton (00:27:09.620)

leveraging its extensive

Dalton (00:27:10.660)

data from Chrome,

Dalton (00:27:11.740)

YouTube, alongside new

Dalton (00:27:12.820)

specialized DPU chips.

Dalton (00:27:14.340)

Yep, I said that as

Dalton (00:27:15.180)

well. A practical

Dalton (00:27:16.820)

application was shown

Dalton (00:27:18.220)

with Google Workspace

Dalton (00:27:19.580)

Studio, automating

Dalton (00:27:20.700)

podcasts, guests, emails,

Dalton (00:27:21.900)

workflows, and summarizing

Dalton (00:27:23.020)

content. Yep, I did do

Dalton (00:27:24.060)

that. So all this stuff

Dalton (00:27:25.580)

is legit, right? And

Dalton (00:27:27.260)

so what I'm trying to do

Dalton (00:27:29.380)

is I'm trying to have a

Dalton (00:27:30.660)

full context window when

Dalton (00:27:32.500)

I'm creating social media

Dalton (00:27:33.620)

posts. So you have the

Dalton (00:27:34.740)

transcription, then you

Dalton (00:27:36.580)

have these key takeaways.

Dalton (00:27:39.420)

So your transcription is

Dalton (00:27:41.900)

the finite grounding

Dalton (00:27:43.540)

where it's like, this

Dalton (00:27:44.460)

is the foundation. And

Dalton (00:27:46.740)

then these takeaways are

Dalton (00:27:48.340)

the floor. And if

Dalton (00:27:50.420)

you're creating social

Dalton (00:27:51.340)

media posts, or you're

Dalton (00:27:53.180)

doing stuff, you need to

Dalton (00:27:55.220)

be grounded in each. So

Dalton (00:27:56.980)

you shouldn't have more

Dalton (00:27:57.780)

stories than you do

Dalton (00:27:59.740)

floors. And so this

Dalton (00:28:01.020)

instance, in this

Dalton (00:28:01.860)

analogy, we've got one,

Dalton (00:28:03.660)

two, three, four, five,

Dalton (00:28:06.300)

six floors. So your

Dalton (00:28:08.620)

building can't have

Dalton (00:28:09.340)

10 stories, if that

Dalton (00:28:10.900)

makes sense. And then this

Dalton (00:28:12.620)

section, you can toggle

Dalton (00:28:13.580)

between who your voice

Dalton (00:28:15.180)

is, but we're going to

Dalton (00:28:16.380)

do venture steps. So you

Dalton (00:28:17.900)

can do like this. And

Dalton (00:28:19.260)

then LinkedIn is really

Dalton (00:28:20.540)

about takeaways and

Dalton (00:28:22.220)

firm. Wow, this is this

Dalton (00:28:24.060)

is really good stuff, where

Dalton (00:28:26.330)

they feel like they're on

Dalton (00:28:27.450)

the cutting edge. And I

Dalton (00:28:28.930)

feel like LinkedIn is more

Dalton (00:28:33.130)

of a buckshot, where it's

Dalton (00:28:36.610)

got a small spread to it,

Dalton (00:28:37.810)

but it's got spread. Okay,

Dalton (00:28:39.090)

did a good job with

Dalton (00:28:39.770)

this. We should put more

Dalton (00:28:41.730)

information, but it was

Dalton (00:28:43.370)

providing more last time.

Dalton (00:28:44.690)

It's weird. But with X, I

Dalton (00:28:46.930)

would describe X as a

Dalton (00:28:47.890)

laser. Like X is as little

Dalton (00:28:51.850)

amount of words and the

Dalton (00:28:53.530)

most interesting takes in

Dalton (00:28:57.420)

the shortest context window

Dalton (00:28:59.020)

you can provide. And so

Dalton (00:29:00.260)

this is what it came up

Dalton (00:29:01.020)

with YouTube community. It

Dalton (00:29:03.340)

does everything within this

Dalton (00:29:04.380)

one click, it batches it,

Dalton (00:29:05.860)

which is nice. And then

Dalton (00:29:07.220)

this is the studio that

Dalton (00:29:08.940)

will create the threads.

Dalton (00:29:11.860)

So it automatically creates

Dalton (00:29:13.380)

the threads for you, which

Dalton (00:29:14.660)

is great. And so it creates

Dalton (00:29:16.980)

a thread to put on X, and

Dalton (00:29:20.180)

then you can specify for

Dalton (00:29:22.140)

Dolan or the brand. Right

Dalton (00:29:23.580)

now it's the brand, and

Dalton (00:29:25.140)

then you can do carousel

Dalton (00:29:26.420)

studio. And so you can

Dalton (00:29:28.140)

create a carousel similar

Dalton (00:29:30.220)

to the thing I showed you

Dalton (00:29:31.980)

before. And you can just

Dalton (00:29:33.900)

click, you know, region the

Dalton (00:29:35.340)

storyboard, and then it'll

Dalton (00:29:36.980)

create the LinkedIn

Dalton (00:29:38.660)

carousel studio, which is

Dalton (00:29:40.060)

more expensive. So I'm not

Dalton (00:29:41.140)

doing it. And then this is

Dalton (00:29:43.700)

the clip sections where if

Dalton (00:29:45.500)

you have a clip, you can

Dalton (00:29:46.500)

put in the transcription of

Dalton (00:29:48.140)

that clip and it'll help

Dalton (00:29:49.260)

create a caption for it

Dalton (00:29:51.500)

slash description. And

Dalton (00:29:54.380)

this is how you could

Dalton (00:29:55.100)

create thumbnails. And then

Dalton (00:29:57.060)

this is how it create

Dalton (00:29:57.900)

blogs. So this is an

Dalton (00:30:00.220)

easier way of getting a

Dalton (00:30:03.260)

first pass of what the

Dalton (00:30:06.470)

caption should be or

Dalton (00:30:07.910)

description should be

Dalton (00:30:09.150)

versus doing all this

Dalton (00:30:10.590)

stuff from scratch. Like

Dalton (00:30:11.710)

it just not, it's just

Dalton (00:30:12.750)

really not feasible for

Dalton (00:30:14.270)

someone to make all this

Dalton (00:30:16.390)

content easily without a

Dalton (00:30:19.470)

big struggle and do it

Dalton (00:30:21.950)

consistently without some

Dalton (00:30:23.670)

kind of structure. So this

Dalton (00:30:24.790)

is just a broad structure.

Dalton (00:30:26.590)

It's a work in progress. It's

Dalton (00:30:28.930)

something that I'm trying

Dalton (00:30:30.290)

to build out and provide a

Dalton (00:30:33.210)

structure for myself and

Dalton (00:30:34.450)

the team. So when we're

Dalton (00:30:36.600)

making stuff, the stuff is

Dalton (00:30:38.120)

consistent, same brand

Dalton (00:30:39.640)

voice. It is somewhat

Dalton (00:30:42.000)

automated. And then the

Dalton (00:30:43.200)

users are not the users,

Dalton (00:30:44.680)

but the user of the

Dalton (00:30:45.360)

platform creates the

Dalton (00:30:47.920)

right voice per se. Okay.

Dalton (00:30:50.940)

So that was venture step

Dalton (00:30:52.540)

footprint. Okay. So this

Dalton (00:30:54.020)

is what I built Friday

Dalton (00:30:55.140)

night, like an hour and a

Dalton (00:30:56.660)

half. The next thing I

Dalton (00:30:57.620)

want to talk about is

Dalton (00:30:58.580)

getting the most out of

Dalton (00:31:01.340)

these tools. So I created a

Dalton (00:31:04.060)

GitHub repository for

Dalton (00:31:05.940)

people that are interested

Dalton (00:31:07.140)

in learning more. It's

Dalton (00:31:08.820)

pretty structured. I

Dalton (00:31:10.340)

helped AI remember all

Dalton (00:31:14.180)

the things that we did

Dalton (00:31:15.460)

together and updated, and

Dalton (00:31:17.820)

then put it all together

Dalton (00:31:18.980)

within this repository so

Dalton (00:31:21.060)

other people could use it.

Dalton (00:31:23.030)

So the main thing that

Dalton (00:31:24.590)

this does is it

Dalton (00:31:25.990)

restructures the way that

Dalton (00:31:30.390)

it calls agents and sub

Dalton (00:31:32.150)

agents. So it has three

Dalton (00:31:35.740)

agents that it typically

Dalton (00:31:37.380)

would sign off on. It

Dalton (00:31:38.700)

would be the domain and

Dalton (00:31:41.420)

adapter agent, UI and docs,

Dalton (00:31:43.740)

UA and security. So UI

Dalton (00:31:46.220)

is user interface, and

Dalton (00:31:47.540)

then UA is quality

Dalton (00:31:48.980)

assurance. Within these,

Dalton (00:31:52.410)

they have ownership and

Dalton (00:31:55.130)

what they own in this

Dalton (00:31:56.810)

instance is domain and

Dalton (00:31:58.890)

adapter. So the domain and

Dalton (00:32:00.610)

adapter agent, it owns the

Dalton (00:32:04.330)

database, the Docker

Dalton (00:32:06.690)

instance, all that stuff,

Dalton (00:32:09.040)

the models, the business

Dalton (00:32:10.360)

logic, and then the UI

Dalton (00:32:12.360)

and docs owns the

Dalton (00:32:13.400)

components pages and the

Dalton (00:32:14.920)

documentation, and it's

Dalton (00:32:16.320)

spun up if there's some

Dalton (00:32:17.800)

kind of browser session

Dalton (00:32:19.120)

needed. And then the QA

Dalton (00:32:21.960)

and security, it's used

Dalton (00:32:23.920)

when there's multiple

Dalton (00:32:24.560)

repos involved. And so it

Dalton (00:32:28.380)

talks about how to

Dalton (00:32:29.500)

structure a goal. And the

Dalton (00:32:31.100)

main thing that you need

Dalton (00:32:31.940)

to do with the goal is

Dalton (00:32:32.860)

tell it the what, and

Dalton (00:32:34.340)

then you got to tell it

Dalton (00:32:35.140)

how should it meet these

Dalton (00:32:36.980)

conditions? Because a

Dalton (00:32:39.310)

goal will run

Dalton (00:32:40.270)

continuously until it

Dalton (00:32:41.670)

can't anymore. So you can

Dalton (00:32:43.750)

tell it what to do. And

Dalton (00:32:45.590)

then the last thing you

Dalton (00:32:47.430)

tell is how to measure

Dalton (00:32:49.030)

that what you did worked.

Dalton (00:32:51.790)

Because if you just tell

Dalton (00:32:52.710)

it what to do is I go,

Dalton (00:32:54.150)

make this thing and they

Dalton (00:32:55.470)

all make it, but it might

Dalton (00:32:57.190)

not be good because it's

Dalton (00:32:58.990)

just going to do

Dalton (00:32:59.750)

whatever it wants. So you

Dalton (00:33:01.550)

got to structure it and

Dalton (00:33:03.430)

tell it, okay, this is

Dalton (00:33:04.470)

how you'd validate that

Dalton (00:33:05.510)

what you did was what

Dalton (00:33:06.830)

I'm thinking that you

Dalton (00:33:07.630)

should have done. And

Dalton (00:33:09.220)

then from there you can

Dalton (00:33:10.540)

iterate. And then there's

Dalton (00:33:13.410)

some verification gates

Dalton (00:33:14.610)

that have been put

Dalton (00:33:15.290)

in place. So this is the

Dalton (00:33:16.810)

hex validate. So hex

Dalton (00:33:18.210)

validate is what I think

Dalton (00:33:20.450)

is the right way to think

Dalton (00:33:22.810)

about architectural pieces

Dalton (00:33:24.890)

within your platform. It

Dalton (00:33:26.090)

kind of depends on what

Dalton (00:33:26.690)

you're doing. But for a

Dalton (00:33:27.530)

lot of stuff I'm doing, I

Dalton (00:33:29.370)

want a hexagonal

Dalton (00:33:30.770)

approach, which is

Dalton (00:33:32.170)

similar to modular, but

Dalton (00:33:33.570)

modular is different. Like

Dalton (00:33:34.970)

modular is like, Hey, I

Dalton (00:33:36.890)

want to have multiple

Dalton (00:33:37.610)

services instead of a

Dalton (00:33:38.770)

monolithic code hexagonal

Dalton (00:33:40.650)

is saying, Hey, not only

Dalton (00:33:42.490)

do I want modular

Dalton (00:33:43.450)

services, but I want

Dalton (00:33:45.290)

inputs and outputs to be

Dalton (00:33:47.250)

these separate entities

Dalton (00:33:48.890)

to where I could switch

Dalton (00:33:50.050)

out something that I

Dalton (00:33:51.410)

didn't want before. And

Dalton (00:33:53.290)

I can plug and play

Dalton (00:33:54.090)

technology. So if I wanted

Dalton (00:33:55.970)

to change my infrastructure

Dalton (00:33:57.290)

from this API to a whole

Dalton (00:34:00.050)

different API, that's

Dalton (00:34:01.330)

fine, because it's not

Dalton (00:34:04.290)

integrated anywhere else.

Dalton (00:34:05.490)

It's just in that one

Dalton (00:34:06.930)

spot. That's a really

Dalton (00:34:08.010)

bad example, but it's

Dalton (00:34:09.210)

really easy to switch

Dalton (00:34:10.370)

technologies instead of if

Dalton (00:34:13.270)

it wasn't like that, it's

Dalton (00:34:14.870)

a lot more work to set it

Dalton (00:34:16.990)

up hexagonally, but it

Dalton (00:34:19.590)

provides a rigid structure

Dalton (00:34:21.110)

for AI agents to follow,

Dalton (00:34:22.790)

which is very important.

Dalton (00:34:24.070)

Because if it's not rigid,

Dalton (00:34:25.230)

then it's hard to verify

Dalton (00:34:26.990)

that it's working. And

Dalton (00:34:28.630)

then it can spire out

Dalton (00:34:30.310)

of control. And then

Dalton (00:34:33.010)

there's security audit. So

Dalton (00:34:34.930)

this is these audits, like

Dalton (00:34:36.850)

you don't want any of

Dalton (00:34:37.530)

your secrets. You don't

Dalton (00:34:38.850)

want any of these, like

Dalton (00:34:41.090)

leaks to happen within

Dalton (00:34:42.770)

your platform. When you're

Dalton (00:34:43.930)

posting stuff on your

Dalton (00:34:45.250)

repos. And then there's

Dalton (00:34:47.820)

the UI slash like lint.

Dalton (00:34:50.580)

And so it just reviews

Dalton (00:34:52.060)

the code. And then

Dalton (00:34:53.820)

there's some auto

Dalton (00:34:54.540)

verification. So I have

Dalton (00:34:55.740)

built in auto

Dalton (00:34:56.460)

verifications for GoLang,

Dalton (00:34:58.540)

for Rust, and then for

Dalton (00:35:00.660)

TypeScript. And then the

Dalton (00:35:01.660)

setup, this is the most

Dalton (00:35:02.660)

important part, you would

Dalton (00:35:03.700)

just copy and paste these

Dalton (00:35:05.460)

things into your terminal,

Dalton (00:35:07.180)

which is the terminal

Dalton (00:35:08.140)

commands. And then you

Dalton (00:35:09.860)

would just be on on the

Dalton (00:35:12.060)

go. One thing it might

Dalton (00:35:14.020)

have to do is I have to

Dalton (00:35:15.380)

put like what to put

Dalton (00:35:17.580)

there. But anyways, this

Dalton (00:35:20.730)

is my first attempt at

Dalton (00:35:24.170)

providing this information

Dalton (00:35:25.730)

in this way. I hope that

Dalton (00:35:27.370)

it's super useful and

Dalton (00:35:29.810)

that it provides a better

Dalton (00:35:33.710)

outcome for you. There's

Dalton (00:35:34.870)

a lot of extra stuff

Dalton (00:35:36.470)

that's built in here.

Dalton (00:35:38.030)

Like one of the most

Dalton (00:35:38.710)

important things that I

Dalton (00:35:39.750)

probably didn't touch on

Dalton (00:35:40.750)

is if the agents are

Dalton (00:35:42.550)

ran, they're only ran at

Dalton (00:35:44.550)

runtime. So what that

Dalton (00:35:46.710)

means is if you start a

Dalton (00:35:48.030)

project, you don't have

Dalton (00:35:51.330)

all your folks around

Dalton (00:35:52.730)

you. You know, if you're

Dalton (00:35:54.490)

starting a project and you

Dalton (00:35:55.730)

need accounting and you

Dalton (00:35:56.890)

need some technology folk,

Dalton (00:35:58.690)

you don't have all these

Dalton (00:36:00.050)

people gathered around the

Dalton (00:36:01.170)

watering hole waiting for

Dalton (00:36:03.370)

you. They're doing their

Dalton (00:36:04.770)

own thing. They're

Dalton (00:36:05.690)

working. And then you

Dalton (00:36:07.090)

call them over when you

Dalton (00:36:08.410)

need them. And so this

Dalton (00:36:09.930)

instance is you only

Dalton (00:36:12.200)

wake these agents up when

Dalton (00:36:13.800)

you need them at that

Dalton (00:36:14.800)

moment. Sometimes if you

Dalton (00:36:16.360)

don't have it set up that

Dalton (00:36:17.320)

way, if it knows it needs

Dalton (00:36:19.960)

that agent, it will just

Dalton (00:36:21.000)

turn it on. And then you

Dalton (00:36:23.760)

feeds it contexts. And

Dalton (00:36:25.720)

then from there, it's

Dalton (00:36:27.470)

wasting contexts. And

Dalton (00:36:28.790)

then it's just

Dalton (00:36:30.500)

inefficient. So you want

Dalton (00:36:31.620)

to make sure that you're

Dalton (00:36:32.220)

not wasting your tokens.

Dalton (00:36:33.620)

And that's one way to

Dalton (00:36:34.460)

prevent. There's many

Dalton (00:36:35.380)

other instances. And just

Dalton (00:36:37.700)

to make sure that this

Dalton (00:36:38.540)

is clear as these

Dalton (00:36:39.380)

verification gates, these

Dalton (00:36:40.540)

are custom commands that

Dalton (00:36:42.260)

you can run. They won't

Dalton (00:36:44.060)

auto fill, but if you

Dalton (00:36:45.140)

type them in, they will

Dalton (00:36:46.460)

pop up. But yeah. Oh, the

Dalton (00:36:49.890)

token efficiency right

Dalton (00:36:50.650)

here. So agents spawn on

Dalton (00:36:52.130)

demand by default is off.

Dalton (00:36:54.410)

And then max three

Dalton (00:36:55.930)

concurrent directors and

Dalton (00:36:57.130)

two concurrent

Dalton (00:36:58.050)

specialists, sub-agents

Dalton (00:36:59.850)

auto kill on task

Dalton (00:37:00.810)

completion, and then

Dalton (00:37:01.810)

parallel writes use

Dalton (00:37:02.890)

branch isolation, and then

Dalton (00:37:05.050)

tasks, completable and

Dalton (00:37:06.890)

under 500 lines of

Dalton (00:37:08.090)

output, always in line

Dalton (00:37:09.610)

never sub-agent. So

Dalton (00:37:12.960)

those are the rules to

Dalton (00:37:14.200)

prevent this ballooning

Dalton (00:37:17.040)

of token usage, because

Dalton (00:37:18.560)

you'll see stuff online.

Dalton (00:37:19.720)

They're like, I'm running

Dalton (00:37:21.000)

50 agents, and I'm getting

Dalton (00:37:23.000)

so much work done. But

Dalton (00:37:24.040)

really what they're doing

Dalton (00:37:25.080)

is super inefficient

Dalton (00:37:26.280)

because that agent needs

Dalton (00:37:27.520)

to communicate with the

Dalton (00:37:28.200)

other agent and that

Dalton (00:37:29.160)

agent needs to

Dalton (00:37:29.680)

communicate back and that

Dalton (00:37:30.760)

agent communicates

Dalton (00:37:31.440)

another agent. And then

Dalton (00:37:32.680)

before you know it, it's

Dalton (00:37:33.640)

the classic game of

Dalton (00:37:34.920)

telephone and information

Dalton (00:37:37.040)

might change, but also

Dalton (00:37:38.840)

it's not efficient. So you

Dalton (00:37:41.680)

want to prevent that when

Dalton (00:37:42.760)

you can, for the most

Dalton (00:37:44.900)

part, that was

Dalton (00:37:46.620)

anti-gravity and anti-gravity

Dalton (00:37:49.880)

command line and anti-gravity

Dalton (00:37:52.200)

2.0. I've had such a great

Dalton (00:37:54.480)

time building with these

Dalton (00:37:55.760)

tools. I've been having

Dalton (00:37:56.760)

such a great time with

Dalton (00:37:57.960)

anti-gravity. I love it. I

Dalton (00:38:00.440)

love anti-gravity 2.0. I

Dalton (00:38:02.800)

love how generous their

Dalton (00:38:03.920)

credits are if you don't

Dalton (00:38:05.480)

have one of those crazy

Dalton (00:38:07.320)

subscriptions. And I love

Dalton (00:38:09.320)

how easy it is to build

Dalton (00:38:10.480)

stuff. It's great. I am

Dalton (00:38:12.620)

a huge fan of Google and

Dalton (00:38:14.780)

their products. And this

Dalton (00:38:15.980)

is another one that I'm

Dalton (00:38:17.380)

fond of. Hopefully it

Dalton (00:38:18.780)

doesn't go on the Google

Dalton (00:38:19.660)

kill list. Who knows?

Dalton (00:38:21.220)

We'll see. But I doubt

Dalton (00:38:22.620)

it. This is really

Dalton (00:38:23.780)

important to their future

Dalton (00:38:24.860)

initiatives. And it's

Dalton (00:38:26.780)

very clear that they

Dalton (00:38:27.620)

think so as well. Of

Dalton (00:38:29.620)

course, wherever you are

Dalton (00:38:31.180)

in this world, good

Dalton (00:38:32.100)

afternoon, good evening.

Dalton (00:38:34.260)

Good morning. Thanks

Dalton (00:38:35.500)

for listening and

Dalton (00:38:36.820)

listen in next week.