Gemini 3 Full Review & Mind-Blowing Use Cases

Gemini 3.

I'm just kidding. Gemini 3 is a fantastic model. Like every other model

fantastic model. Like every other model provider comes out and says as the best model in the world as of right now for the next 2 or 3 months, Gemini 3 is the best model in the world according to all the benchmarks. But I'm not going to

the benchmarks. But I'm not going to bore you with the benchmarks. What we

are going to do as usual is we're going to take some use cases that I believe were hard to sort of pull off on previous models. So, I'm going to

previous models. So, I'm going to compare this to GPT5, which I've done a full deep dive video the day it went live, which was several months ago. And

I'm going to show you all the criticisms that people lobbed at GPT5 that Gemini 3 can sort of solve. Now, remember, we're getting model updates and progress every 3 or 4 months, which is a good reason for you to hit subscribe on the channel

because here we actually deep dive.

We'll try a lot of prompts. I'll try to keep it simple and we'll test it in multiple different ways so that you walk away with some use cases, some ideas of what the model does. But the only way to truly learn is to open Gemini, make an

account, and start using the 3.0 model.

Okay. So, I'm going to be doing most of this in something called AI Studio, which is you can use the Gemini subscription and do it from the Gemini official gemini.google.com or whatever

official gemini.google.com or whatever the link is, or you can do it in a studio.google.com. I like using AI

studio.google.com. I like using AI Studio. Google.com because it's got this

Studio. Google.com because it's got this section called build. So, if you remember, the main criticism against GPD5 was that the websites it generates all looked kind of the same. They all

had that same purple crypto website vibes. I think Gemini saw that, but I

vibes. I think Gemini saw that, but I think it's best proven with examples.

So, I'm going to do examples in two or three ways. The first way is I'm

three ways. The first way is I'm actually going to just prompt and see what it comes up with. And then I'm going to take images from other platforms, other people's websites, and ask for a replication, but with some changes because we don't want to, you

know, control CRV that website. I just

want to see what it comes out with. I'm

not saying that you should use this or this is the format you should do it with. Let's start our series of tests

with. Let's start our series of tests with websites. We'll do a little bit of

with websites. We'll do a little bit of front end. Of course, there's going to

front end. Of course, there's going to be no back end attached, but we'll do websites with a little bit of let's see what apps look like as well. So, I'm

going to start with a prompt. I'm in AI Studio right now. I have Gemini 3 Pro preview selected. I'm going to prompt

preview selected. I'm going to prompt make a website for a credit card company

called Bread Bread Pre with nice skuomorphic design animated

header. Sorry, I spelled that wrong.

header. Sorry, I spelled that wrong.

Morphic design. animated header text that says not everybody

gets it.

That's their tagline. and a lot of hero icons and testimonials that are, you know, sort of

marqueeing from left to right. A lot of hero icons with orange with with

Let's do black and gold with gold accents. Okay,

accents. Okay, the general website theme is black and gold with the black seeming like a UI

recreation of the credit card texture.

Actually, instead of saying like a UI recreation, let's say seeming like with the background seeming with the background seeming like a

recreation of the credit card texture.

Okay.

Make it look cool, modern, and animated. Let's go see what it

and animated. Let's go see what it makes.

So, one thing I like about Gemini is it gives you suggestions. For example, it's saying add interactive card application form. It's saying implement dynamic card

form. It's saying implement dynamic card feature showcase. Uh, it's saying

feature showcase. Uh, it's saying enhance testimonial marquee with user avatars. I really like those. Although,

avatars. I really like those. Although,

I don't think this is the way we're going to be making websites. I think

what's going to happen is you're going to give it a reference. You're going to go to you're going to go find somebody else's website. You're going to be like,

else's website. You're going to be like, make something like this. But when

Gemini creates it, it's not going to create exactly the same thing. It's

going to create it with slight difference. And that difference is going

difference. And that difference is going to be enough to give your website uniqueness because it turns out in the real world because I've hired so many website designers. I've made websites

website designers. I've made websites when I was much much much younger. First

in HTML CSS raw with JS and then many years later with web flow. You'll see a tutorial that I've done seven or eight or nine years ago with web flow and then of course with now these modern tools that exist right I've tried making

websites in all different eras. In fact

I've done websites in an era. We used to use HTML CSS but we used to use a paradigm called tables where you actually had to make the website in Photoshop and then slice it up. So I've

seen all these different iterations right and every time a new iteration comes out the old people say oh the new way sucks but it's always been very reference based. We always look at

reference based. We always look at somebody else's reference and then say I want to make a version of this which is why you have this huge mimemetic drive of when somebody in like when there's a new style of websites that comes out

everyone starts copying it. linear for

example for all SAS companies when they start making a certain type of websites everyone start copying it it's a very common thing so I believe this putting a reference and then saying make something like this is going to be very common therefore there's so much human value in

uniqueness if you come out with an actual new unique idea of a website style that's how you stay ahead which is not easy to do right it's not easy to come up with unique ideas all the time

all right and our website is ready as you can see it's it's not great. Like it

still looks cheaply done, right? Like if

you look at this, not everybody, the bevel here is like ugly. You can tell it's using a bunch of different things and you can tell there's some animation.

You can hover over. You can see some of this nice, you know, let's just let's just expand this a little bit. You can

see a little bit of nice, you know, uh, vibe. You can see these moving around.

vibe. You can see these moving around.

You can see one nice pause, that double dash here, uh, the apostrophe here. It's

okay. But this is not how we're going to make websites in the future. I'll tell

you how uh you'd have much higher success. You would go to and I'm

success. You would go to and I'm actually going to do this in Gemini directly. I'm not going to do this in AI

directly. I'm not going to do this in AI Studio because it's a little bit faster in Gemini directly. We interrupt this segment to talk about today's sponsor, Higsfield. Now, we've been generating

Higsfield. Now, we've been generating websites with AI for a while now, and honestly, the layouts are incredible.

The code is clean, the design is sophisticated, but there is always one thing that keeps breaking the illusion.

the assets every time. Either you get basic placeholders, or it pulls these generic stock images that make everything look templatized. And I've

tested this across a range of models.

They're all geniuses at structure, but terrible at visual identity. It's like

having a puzzle that's 95% complete. But

that one missing piece makes it obvious something is off. But logically, that shouldn't be a problem anymore because diffusion models have gotten insane. We

actually have the tools to fix this. You

just need to pair the right art engine with your coding engine. That's where

Higsfield fits in. They have everything in one spot like Nano Banana Pro for illustrations, 2D assets, custom graphics, basically the exact tools you need to fill in those generic placeholders and it just connects

everything perfectly. You generate your

everything perfectly. You generate your structure in Gemini, but you create your specific assets in Higsfield. Suddenly,

you have something that doesn't look like a prototype. It looks like a shipped product. And for developers or

shipped product. And for developers or anyone vibe coding apps, this is something you'd actually ship. Asset

quality is what separates amateur work from professional results. We actually

partnered with Higsfield on this Black Friday sale. It's 65% off for 3 days and

Friday sale. It's 65% off for 3 days and that includes unlimited 4K image generations, so you can really dial in the look without worrying about limits.

Now, back to the video. So, let's

actually go to Perplexity's website.

Let's go to Comet's website. So,

perplexity.ai/t.

Let's see how well Gemini does. I'm just

going to copy this. I might have to send it a screen recording, but for now, I'm just going to copy the link. I'm going

say make this website in canvas.

Exactly. That's it. Stupid prompt. Make

this website in canvas. Exactly. When

you give references, it might be useful, but I haven't actually tried just giving a website link and asking it to look at it and come back. Uh, it's probably better to give it a big screenshot or a,

you know, sort of video recording. So,

let me do that.

Select a recording area. So, I'm going to record this area. Okay.

I'm going to start and I'm going to record this website.

It is a pretty website.

Okay. And I'm going to stop recording.

Okay. Now, what I'm going to do is I'm going to save this on my desktop as perplexity recording.

So I'm going to select thinking.

I'm going to build. I've turned on canvas. So as you can see the canvas

canvas. So as you can see the canvas icon is here. And I'm going to upload the video I just recorded. I'm going to say

make this website for a browser called Comet. Exactly. So now we've uploaded

Comet. Exactly. So now we've uploaded the video. We've asked it to make a

the video. We've asked it to make a website for a browser called Comet.

Exactly.

So, as you can see, it's saying, I'll create a pixel perfect recreation of the Comet browser landing page based on the video provided. I'll use React and

video provided. I'll use React and Tailwind CSS to replicate the layout, typography, which is serif headings, animations, and custom UI elements like the 3D spheres, and the bento grid

feature cards. Okay, cool. I'm waiting.

feature cards. Okay, cool. I'm waiting.

Ladies and gentlemen, vibe waiting.

Look at how much work I'm doing. I have

an idea. Every time, every single time the thing is generating something, I'm going to speak in Hindi. Okay, so my Hindi is terrible. I studied it for 14

years. Oh, thank God. Saved by the

years. Oh, thank God. Saved by the output.

Okay, so here's the output. Let's take a look. Let's share this. I'm going to

look. Let's share this. I'm going to copy this. Let's see this in a big

copy this. Let's see this in a big screen.

Wow, this is actually pretty cool. Icons seem

hoverable, clickable. I can't seem to click inside this. I don't know if Perplexity allows that, but it wouldn't know from the screenshot, from the

recording. These look very nice. They're

recording. These look very nice. They're

not exactly the same color, but they look very, very, very nice. Like all

these placeholders, everything looks very nice. These are very simplistic 3D

very nice. These are very simplistic 3D balls. I assume you need to generate

balls. I assume you need to generate these images with nano banana or whatever which I think they will let you do at some point. The FAQs and you can see a little bit of this cutff as well.

I think that's by design. It feels like it's by design because it fits here very well. And then download. This is not

well. And then download. This is not perfect and you can see those lines, those wavy lines in the background on perplexity's website are missing. Uh but

it's still very very very good. Like if

I went to a website that looked like this, I potentially would make a decision, right? Like if I had to make a

decision, right? Like if I had to make a buying decision or a download decision, this website wouldn't turn me off. It

still requires a little bit of work.

That's where the human comes in. The

last 5 10% of work, but it is a reasonably good output. It's a much better output than what we got with bread. And that's what I'm telling you,

bread. And that's what I'm telling you, right? I think we're entering a world

right? I think we're entering a world where let's say you don't want to use somebody else's work. You'll probably go to an image generator or nano banana, generate a design, you'll like the design, you'll just you'll generate a

screenshot of the design. You'll be

like, "This is a very very good design."

And then you'll dump that into um Gemini and say, "Hey, now can you make this website for me?" I think that's going to be the path. Unless it's already a very well-known website or very well-known

app. I'll give you an example. Okay. And

app. I'll give you an example. Okay. And

this comes this brings me to my second example. Let's say on Gemini I want to

example. Let's say on Gemini I want to create I'm just going to create in canvas make Facebook app front end Facebook app front end submit now you

can see I'm putting in four four five word prompts uh and you might think why don't if you prompt this a little bit better little better the outputs will be better maybe but I'll tell you what I prefer these days okay especially with

things where where Gemini already has context Gemini already knows what Facebook looks like so I prefer an output and then I prefer to manually go in and ask for changes And you can do that very well. Earlier when you asked for changes, it would rewrite the entire

code base. It still does that, but the

code base. It still does that, but the entire output would change like every part of the output would change. But now

you can go micro and be like just change only this type of this specific button.

And it's somewhat good at keeping context and saying I will only change this button which is very different from you know some of the earlier you know GPD4 outputs or even the earlier Gemini outputs right where it would lose track

of the entire thing if you asked her to change one small thing. So anyway, it's making Facebook clone.

I think we are reaching an era where if even if you want to communicate with AI, you need to give a lot of context. So

either you write like five sheet prompts or you just give it an image. And I

think most of us will choose the approach of first generating an image and then sending it to AI. So here we are. We've got Facebook,

are. We've got Facebook, which is nice. It looks exactly like Facebook. Like there's nothing changed.

Facebook. Like there's nothing changed.

Uh, as you can see in the front end, this got updated. This is not storing it anywhere. Uh, Alex Johnson, hello. Yeah,

anywhere. Uh, Alex Johnson, hello. Yeah,

this is an exact copy of Facebook, right? None of these buttons work. And I

right? None of these buttons work. And I

could probably prompt it to make those buttons work because this can all happen at the front end level. Uh, but this is a very, very, very good Facebook output with none of the buttons working. Now,

you integrating it, hooking this up with your back end should be straightforward.

Now there are tools like lovable and emergent and bunch of others that have come out even replet now that allows you to take the front end and connect it with the back end that they give you. So

they have natively they have things like superbase which you can just set up on the fly which I think is useful but I think I don't know if anyone's going to directly use this or even a back end

that any of these tools generate and just put it on online there. Maybe you

can do it for simple apps. So, let's say you want to build a blogging website, you would use WordPress and then you'll use AI to regenerate the front end to make you look different from everybody else. Same if you're building a commerce

else. Same if you're building a commerce store, you'd use Shopify and then you would use AI to take that existing front end and change it up enough. I think

that's the route people will use this in because I feel like backends specifically and frankly the more complicated it gets, the more complicated your back end gets and the more you want stability, the more you're at scale, uh you're not going to oneshot

with these. In fact, the way to build

with these. In fact, the way to build backend with AI is probably to use a cursor or an anti-gravity, right? You

would probably use a cursor and say, "Hey, uh, I'm writing code." And this is good autocomp completion. It's code

completion. So, you still need a human guiding the process of what the back end is, knowing the entire context and using AI to kind of write it for you. But I

think front ends, you'll generate whatever your UIUX or your product manager has given you and you'll hook it in. But I think that's going to be

in. But I think that's going to be exposed to end customers and I think end customers will just directly use this, right? So I think that's pretty cool.

right? So I think that's pretty cool.

Although I do believe that a human is required in the loop a little bit, right? Like you saw with the perplexity

right? Like you saw with the perplexity website, somebody has to swap out all those images. It's not impossible for

those images. It's not impossible for the AI to do. Okay. And I think AI tool use which is can you go to nanobana directly generate those planets and bring it back in my perplexity website

in this location. That I think will take a little more time. It's not very very far away. And I'm confident that at some

far away. And I'm confident that at some point that'll be solved as well. But I

think as a software engineer now your job is more the synthesis role. You're

not no longer writing code. You're

playing one step up and deciding what to build end to end. And I think these are tools that help you go 10x faster. So I

do believe the top software engineers are going to do 10x what other people can do or 100x what other people can do which unfortunately comes at the price of other software engineers. Remember

I've said this infinite times right?

India has overproduced software engineers. Uh we have uh real estate

engineers. Uh we have uh real estate companies that are pretending to be colleges selling seats but they're creatively charging you rent. So it's a rent company instead of uh you know like a college like teaching you stuff and so

many of those people who graduate from college don't even know how to use git.

So the minute you have that combined with this really if there's somebody who understands software very well and also has the distribution to get enough clients I think they'll win like your competition is is is not very good. So I

do believe that you know the 10x software engineer will do 100x the jobs now which we are entering the world of.

The other problem with this is I think that also dips the value of software. So

when you have these sort of phenomenon where it's easier and easier to build something the value of that drops.

Supply demand is a curve that applies to all jobs. If tomorrow we found a you

all jobs. If tomorrow we found a you know a huge gold mine that had infinite gold that you can mine very easily the value of gold would dip. Similar thing

has happened to software already. We've

already seen some of the layoffs that are happening at TCS and other companies. But remember, none of these

companies. But remember, none of these macrolevel things matter if you're a great software engineer. This is your time. This is the time for you go for

time. This is the time for you go for you to go 100x of everybody else. But

there is a lot of competition. And the

problem with extremely competitive spaces like this is it's very hard to take breaks. It's very hard to like

take breaks. It's very hard to like spend time with your family this that because it's a it's a rat race, right?

You're competing with hundreds thousands of people. So, I really think you need

of people. So, I really think you need to make your choices around this wisely.

Anyway, I want to show you two or three more websites and I'm just going to put it up here which we've already made with Gemini. So, you can see the website, you

Gemini. So, you can see the website, you can see the prompt, you can copy it if you want to. Anyway, now let's ask for harder stuff. So, I'm going to ask it to

harder stuff. So, I'm going to ask it to make a simple video editor in the cloud.

Uh, I can upload images and music and stitch them together in a timeline. the

basic features of Premiere Pro in the cloud.

Okay, cool. Make sure it works.

And I hit enter. While this loads, right, I want to tell you the crazy thing about progress. I feel like we just become so normalized to progress.

There's a person the other day on Twitter who was just like, "Oh, you know, only design has improved. That's

all. Wow, I expected more." But think about it, right? Like the difference between GPD5 and Gemini 3, it's just been a few months. And the entire AI thing started, I remember we built God in a Box in November 2022. We launched

it in November 2022, which means we started building it around October, let's say 2022, 2022, 2023, 2024, 2025.

We're in 2025, end almost. So let's say 3 to four years. In three or four years, we've had so much progress with AI. Do

you remember the early images? Do you

remember the early LLM outputs? Do you

remember GPD3? I put GP3.5 in WhatsApp.

So I remember how terrible it was to see it from there and we used to use things like automatic agents. We built

something called autocode pro. To see it go from there to here has been phenomenal. And it's so easy to lose

phenomenal. And it's so easy to lose track and be like oh progress 3 months 4 months. Like what are you expecting it

months. Like what are you expecting it to do? Like are you expecting it to come

to do? Like are you expecting it to come in and just like replace everything and then be like oh we have AGI. Are you

waiting for it to come replace everything and then wake up? Anyway, so

we've got Premiere Pro. It's called

Cloud Edit. Let me just let me just open it. Let's see how much of this works and

it. Let's see how much of this works and how much of this doesn't because mostly in the front end.

Okay. So, I'm going to add an image. Let

me just download an image. I go to Google and type image.

Okay. Download an image. Let me just work with images for now. Uploaded the

image. Okay. Now, let's

let's add one more image.

Okay. I'm adding another image.

It's added two images next to each other. Let's see if I can rearrange

other. Let's see if I can rearrange them. No, it doesn't seem like I can

them. No, it doesn't seem like I can rearrange them. Can I move the timeline?

rearrange them. Can I move the timeline?

No, it doesn't seem like like I can move the timeline otherwise. Or maybe I'm missing keys. Okay. Oh, yeah. I can add

missing keys. Okay. Oh, yeah. I can add duration. So, let's say I add 10

duration. So, let's say I add 10 seconds. Okay, at least it's able to

seconds. Okay, at least it's able to stretch out the images. I'm pretty sure I can ask it to move the timeline. It'll

do it. We'll try that. Let's see if I can export video.

Oh, it's moving the timeline.

It's rendering. Not bad. Very simple.

Are you going to give me an output, my friend? I'll be very surprised if you

friend? I'll be very surprised if you give me an output. Although, this should be an easy operation.

Okay, I stuck. So, yeah. I mean, by the way, I don't think this is spectacularly any different from GBD5 because I've run some of these experiments on GBD5. It

kind of does the same thing. Sometimes

the timeline works, sometimes the timeline doesn't work. Exporting never

works. I'm fairly sure that look apps like this need somewhere to process right especially if you're using Python to stitch you know clips together and stuff it needs some space to go out and process it needs dedicated compute which

doesn't exist here so it's obvious that this is failing that's why I said right now the tools are fragmented so you need a human to come and say I'll generate an UI with this I'll generate this with this I'll generate this with this and put it together and here I will write a

little bit of manual glue but compared to writing software 20 years ago it's much much much faster and I think that Whoever starts binding all these things together and moving quickly can end up

building hundreds of pieces of software.

But again when that can happen when supply demand gets kind of you know muddled the price of software goes down.

So if you take the specific field of software as a service, what will end up happening is that let's say all these you know sales forces of the world will continue to become big right they already are big they're going to keep their do domain Zoho will continue to be

big they'll keep their domain uh but you know the new entrance coming in will find that they now have to compete with hundreds of other micro entrance who are all using AI to build their apps much much quicker and one of the people will

wake up and and be like I will sell my app at zero or I'll sell my app at 0.1 so there's a price raise to the bottom which means nobody makes money which is not a good world to live in. So there's

the enterprise segment where there are big companies who are willing to pay the sales forces and the Zohos of the world and then there's the lower segment where you know everyone's racing to zero. I

don't know how much money is in there and I feel like because I've run a SAS company in the past I know you have to choose one of the two domains mass markets low ticket SAS or very big enterprise SAS the midmarket graveyard.

uh and there's a very beautiful article about this which I'll link in the description which is actually about and I I sort of read this when I was building a SAS tool right for for companies to run their communities many

many years ago which is that when you're selling to these companies you have to choose this either you're like selling at like a 50 lakhs a cr a year but you know big solution services along with the solution human beings to help them

set it up etc or you're going very low ticket very low price but huge like thousand hundreds of thousands of c millions of customers the middle area is graveyard and Nobody makes money there.

Where you're like $200 or $500 and it's not clear who the decision maker is. Is

that $500 or $1,000 a month? Is it like a junior person in the company who's going to use their card? No, they're not going to spend that much a month. A

senior person with a big check writing capability is not going to care that much about $500,000 a month. It's a very weird spot. Um, so mid-market is not a

weird spot. Um, so mid-market is not a very good place to be in. Actually, it's

more about the kind of customers you target. But this is something to to uh

target. But this is something to to uh to think about. Now, let me show you some other more fun stuff that we've already done with Gemini. I don't want to waste time writing the prompt. I'll

just tell you what the prompt is. I

don't want to waste time writing the prompt because it just takes some time on the video. So, we generated a nice little iOS weather app. Again, we're

running all these on code pen. Um, but

yeah, it's just a very simple, you know, weather app. Uh, next, and I thought

weather app. Uh, next, and I thought this was interesting because it was one of my first projects when I was learning Ruby on Rails 12, 13 years ago. I want

to make a PC builder. And the thing about making a PC builder, which is what motherboard, what RAM, what CPU to add in, is that some CPUs are only compatible with some motherboards, right? So there's a this compatibility

right? So there's a this compatibility chain that's very very important. And

also you actually have to go out and figure out the price of all of this on Amazon. So I actually asked Gemini to do

Amazon. So I actually asked Gemini to do that. I went to Gemini 3 and I asked it

that. I went to Gemini 3 and I asked it for, hey, I want to make a PC builder in India. go to Amazon, find products of

India. go to Amazon, find products of all these ranges, figure out compatibility, and give it a nice graphical UI where I can add vis in a wizard step by step each element, and

then when I add an element, it looks like a PC is being completed. I actually

typed this prompt in, got this output.

So, I'm going to show you. It's called

gaming PC builder. So, I'll let's say I'm adding the ASUS ROG. And as you can see, it made this nice little illustration of a uh motherboard. And

you can see the slots that are available. So, then I'll go to CPU. And

available. So, then I'll go to CPU. And

as you can see, lots of these are grayed out because the Ryzen CPUs don't work with this motherboard. Let's say I'm taking the Intel i9 i3900 K. I drop it

here. Okay, I do that.

here. Okay, I do that.

Okay, and as you can see the the i9 i3900 K just dropped in and you can see it here. I can press X. Beautiful UI.

it here. I can press X. Beautiful UI.

Thank you so much, Gemini. Uh, let me add in some RAM. GK skill.

Okay, sorry. I have to drop it in the right location. So, let me add two RAM

right location. So, let me add two RAM sticks. I added two RAM sticks. They

sticks. I added two RAM sticks. They

turned yellow.

Very pretty. Let me add a 1TB NVME. And

it's very nice. Let me add a 4070Ti here. And as you can see, it generated

here. And as you can see, it generated it with the fans. It's so beautiful.

Okay. And finally, of course, let me add in a PSU, which is a power supply unit.

Uh, let me add the deep cool 750 watt.

As you can see, there's an electricity bolt there. It's 2 lakhs, 19,800.

bolt there. It's 2 lakhs, 19,800.

I should have asked Gemini to add another button saying link to all the products with an affiliate code and then they can people can buy it, which I didn't do here. But as you can see, I actually spent months learning how to do

something like this when I was 17 or 18.

The the compatibility chain actually is a little bit hard and I can't believe we've oneshoted it. It's crazy that that we're able to do this today, right?

Right. And Gemini 3 can do it out of the box. Yeah. The only problem is now

box. Yeah. The only problem is now everybody can do it. Right? So it

depends on what ideas you have and how much you know how many people can you get to this website. You make let's say you make this PC builder website. How

many people can you get to this website?

That's all that matters. And I hope this explains why 3 years ago we started focusing as a company as a so much on distribution cuz we kind of saw this coming right. It's like even if we

coming right. It's like even if we disagree on timelines, even if you say okay von it's going to take two more years for coding to get better. And I

and let's say 3 years ago we said I had a timeline. Let's say you had a

a timeline. Let's say you had a timeline. Even if the difference was one

timeline. Even if the difference was one or two years, a career doesn't last just four or five years, right? It lasts a very long period of time and you'll do multiple things during that career.

Doesn't it make sense to do first what is most valuable if things do get automated? That's what we focused on. So

automated? That's what we focused on. So

I do believe that now that everyone can build this, the value of building it is lesser. The value of now this synthesis

lesser. The value of now this synthesis this plus can I get users on this plus can I negotiate with all the people providing me parts all the Corsair, Deepcool, etc. MSI, Nvidia and all and

tell them, "Hey bros, can you give it to me at slightly more discount?" So, it's a little bit of B2B sales with them to give you a little more discount, a little bit of, you know, getting users to your platform, advertising, content

creation, however you want to do it. And

then, of course, using AI to help build the products and of course, if you want to change it a little bit, you can. But

I think let's say even if you improve the UI on this, like 10 20%. My question

to you, does it matter? Does performance

on this website matter? Like, it's

loading like I'm in the corner of my office, right? I'm not even on land.

office, right? I'm not even on land.

It's loading reasonably well. I see a lot of conversation on Twitter about, oh, but AI can't do like the code won't be performant or those images will load slowly. If you look at this website,

slowly. If you look at this website, they're not even images. Like all of this is like just like even the fans moving here are not real images. So I

just I do believe that AI can write even more performant code. Those are not the things to focus on. Those are not the things to keep improving because you're like you're competing with this thing that's moving really fast. Instead,

you're like, how can I do synthesis? How

can I do the parts that AI doesn't do?

How can I do that 10 20% adding masala on this getting the users negotiating for discounts those matter those matter more and actually if you look at India's IPOs recently look at lens look at even

zomatos of the world even if you look at grows of the world even if you look at physics fallas of the world they are not pure tech right if you want to start a stock broking platform tomorrow do you know how many licenses you need to get do you know how closely you need to work

with you know all the government bodies that's a business mo right that's a that's a relationship mo if you wanted to build zomato do you know How much of it is product? How much of it is logistics? How much of it is people

logistics? How much of it is people management? Like these are all synthesis

management? Like these are all synthesis businesses, right? And it's now easier

businesses, right? And it's now easier to do synthesis. It's actually easier for you to go from being an engineer to uh running a company. It's cheaper to do it. Money is no longer, you know, as

it. Money is no longer, you know, as much of an excuse. Anyway, next then we made some, you know, we we we did this thing for uh to check configuration. I'm

just going to show you what this looks like. So, as you can see, you can see

like. So, as you can see, you can see this spindle we created. It's a helmet builder. So I can choose between two

builder. So I can choose between two different types of helmets. So let's say I'm choosing highspire and I can change the nasal position where the nose goes.

Uh yeah just some experimentation with 3D. We keep doing this whenever a new

3D. We keep doing this whenever a new model comes out. We have a bunch of things we create to see how it ends up doing. The models are now oneshotting

doing. The models are now oneshotting most of these. So we need new tests. Um

these are this is a solar system we interactive solar system we created which is nice. It's very useful for education, right? Because you can get to

education, right? Because you can get to visually see.

But I think videos have replaced that in some way. Like I would rather watch a

some way. Like I would rather watch a video of the solar system than play around with one of these. But there are games there are there were simulators when I was young which allowed you to see physics simulations of, you know, planets colliding and stuff. I think

it's still there, right? Universe

sandbox or something like that. There's

a game that allows you to do it. We

generated, you know, a simulation of of an OS where you can click on stuff but and and it's pretty nice but none of these work and the close button also doesn't work but it's not far away that

all of these will work from a front end level but now somebody has to build the back end to bind all of this. I do

believe that AI at some point like if they want to solve AGI they have to solve everything like you type in make an OS and it does everything front end back end. That's the goal for Silicon

back end. That's the goal for Silicon Valley. That's why trillions of dollars

Valley. That's why trillions of dollars are being spent. So that's why I don't think that's a good domain at all. But I

still feel like in the next 3 four years, people who do this com combination of everything that this provides with your backand will end up doing a good job. We generated like a fight like a Mortal Kombat type fight

between emojis. So Kraken and let's say

between emojis. So Kraken and let's say Harambe.

And by the way, these are all like very very simple prompts. We created a Lego builder. So as you can see, I can put up

builder. So as you can see, I can put up some of this stuff.

These are all very very simple prompts and you'll be surprised by how simple the prompts are, how much of the context it fills in for you. But remember that now it's about seeking these ideas, right? Like where do you find new ideas

right? Like where do you find new ideas that other people haven't already put up? This Lego idea is actually an idea I

up? This Lego idea is actually an idea I saw on Twitter, right? Whereas some of these ideas we came up with ourselves, some of these ideas we saw on Twitter.

So it's like thinking of new unique ideas means you need domain expertise, right? So let's say you work in like I

right? So let's say you work in like I don't know like a lights company or something like that, right? And then you see these tools and they're like wait I had a problem statement I'm dealing with which this tool can help me. So I think problem first approach is very very

useful because then all these tools that are coming in you can use it in the problem like 3 years ago our problem was how can we use all of AI in video right our goal was video and video games when we started and it's part of our we had this I had this I wrote this one

document about what we're going to do and everything in in AI we have used towards that everything so if you look at AI avatars if you look at you know the some of the software we built to manage our videos at scale we have our

own internal GitHub for video where you can see multiple versions if you see the video editing school that we built if you which is has no AI in it right we teach a little bit of AI but it's a it's a manual school that's being run if you

look at VO3 if you look at nano banana for thumbnails if you look at genfill in Photoshop these have all helped the main problem statement of videos so any new model that we see with AI we're like how can we use it to solve our main problem of video which I think has been crazy so

if you just pick a problem statement then everything in AI suddenly you'll have like 10 good ideas if you run a computer shop you'll be like this is a good idea I should do this it'll get me more scale it'll get it'll make it easier for clients to come in with

already spec Therefore, I just have to focus on optimizing how many views this website gets. It's much better than

website gets. It's much better than saying, "I'll pick up the phone and talk to hundreds of clients." There's other stuff with Gemini, which is how it writes and how it reasons. And to be very honest, I don't think most people

can notice the difference. U especially

with writing, I have not noticed too much of a difference between Gemini 2.5 and Gemini 3, which is why I'm not actually adding it in this video. I feel

it also reasons pretty well, so it'll solve most problems you throw at it, but so did Gemini 2.5. And none of us are sitting here solving very difficult math. The last piece of math we used AI

math. The last piece of math we used AI to solve was in our game we have a chakra, right? The fragmented chakra. So

chakra, right? The fragmented chakra. So

AI helped us a little bit with the physics on the return of the chakra.

Well, that was GPT. But it doesn't matter like you can use any AI model.

They're already at that, you know, level where they will solve 90% of your problems unless you're sitting and doing physics or math or something like that.

So I'm not actually covering that in this video. That's it for me. I hope you

this video. That's it for me. I hope you like this video. I hope you subscribed.

I tried covering the differences. We'll

keep doing this for every model henceforth. Make sure you subscribe.

henceforth. Make sure you subscribe.

Namaskar. And thank you for your time.