Your startup idea is their weekend holiday

Yeah. Okay. It's a fair question. Is

software development utterly and completely beeped? Because think about

completely beeped? Because think about it. In a world where AI models can

it. In a world where AI models can create code, should founders still create SUS startups? Does it even make sense? And you might have noticed how

sense? And you might have noticed how right now is the worst time ever to fundra for a SUS company. But it's

paradoxically also the best time ever to build software. In this video, we go

build software. In this video, we go quickly through what changed. Why is it so hard for me right now as an investor to actually invest in SUS companies? And

I have like over 100 investments beforehand, right? and also where I

beforehand, right? and also where I think that there is still a lot of opportunities for founders that if tackled the correct way, in my opinion, a billion dollar use cases. Let's jump

into this. If you don't know me, my name is Andreas. I used to be an engineer. I

is Andreas. I used to be an engineer. I

I still think of myself as an engineer, but the rest of my team actually tells me I'm not. This is me at my previous workplace. Uh this is Ryan on the right

workplace. Uh this is Ryan on the right hand side uh electrocuting me for not shipping fast enough. True story. Don't

ask Ryan. It's completely true. Uh this

is pure workplace torture. The old word was about SUS versus my Gmail, my spreadsheet, my laziness to ever adopt it. And in almost all cases, my Gmail

it. And in almost all cases, my Gmail and spreadsheet hack was good enough.

And this is the problem why it was hard for SAS to establish yourself. Nowadays,

you have additionally CHDP. And quite

frankly, I most likely can do the core functionality good enough with my CHDP.

If not, I can use codec or cloud code to actually ad hoc build what I need. And

this is now the world where you're competing as a as a founder. This is a a crazy space to be in. Two of my favorite examples. This is uh from Toby of

examples. This is uh from Toby of Shopify who got an MRI scan from his doctor who asked him to install a Windows software and he couldn't be bothered so he just quickly asked Claude to build him his own software. This is

another example. Christopher Jans he's a very famous SUS investor. 10 years ago he would have run after startups building this kind of stuff to invest in them. Now he builds that stuff on the

them. Now he builds that stuff on the weekend for himself just for the fun of it. This is Project Genie by Google.

it. This is Project Genie by Google.

It's an effort to make an whole interactive word with a prompt. Um think

of it like as games for example.

Ultimately these will be word models being able to understand physics to understand complex reasoning and so on and so on. What I'm trying to say is like we get more and more into a world where oneshotting of software is absolutely possible which means if you

think this through uh we ending up in a world where personalized software is a thing like it used to be you know like use Salesforce or build a custom thing and now it's kind of like ask claw to

make it and it's 80% there right soon we will have software that exists for 30 seconds and just for this one moment and then it's gone. This is a quote by a friend of mine who's a very famous software developer. He asked me not to

software developer. He asked me not to add his name. Basically, his argument is like he thinks there's five good years of software development as we know it and then it's gone. And whenever I have this discussion, there is two clear

paths that people argue. One is the law path. Basically saying look, it's 90%

path. Basically saying look, it's 90% there, but 90% by the power of 10 is horrible, unusable, complex crap essentially. And quite frankly, they

essentially. And quite frankly, they have ingested all the public data that exists. So what is the ceiling of the

exists. So what is the ceiling of the capabilities? Like is it already there?

capabilities? Like is it already there?

Will it get like better like TBD right?

Nope. It will get better. We will use reinforcement learning for example cloud code which is basically just a hack to get a very clear yes or no decision acceptance criterias. We will use

acceptance criterias. We will use multimodal models uh to understand complex reasoning and be able to make way better decisions in code. Maybe

we'll use uh in latent space reasoning which basically means instead of like having the output of a model then become the prompt you just like stay in the latent space and are 100 times 1,000 times 1 million times faster. What does

this mean for you as a founder? The way

I think of this is like you need to make a decision. Do you want to build a

a decision. Do you want to build a product? Do you want to build a company

product? Do you want to build a company that quickly makes money? Or do you want to build a unicorn case? And for all three, it's a different strategy. It's

also a different goal. For a product, your goal is fun. You want to learn something. And that's very important for

something. And that's very important for a company. It's like how quickly can

a company. It's like how quickly can this make money? And for a VC case, at least in the beginning, there's a reasonable question like can this even get funding?

And this is obviously very important.

Let's start with the first one. Fun.

Meet Peter. He's a very good friend of mine from Vienna and his project Open Club went completely bonkers viral over the last few weeks. If you really don't know it, please look it up. It's one of the coolest thing happening right now in

AI. It's basically him using a side

AI. It's basically him using a side project where he just vip coded everything into existence that he wanted to be the perfect agent. And he's also in my opinion the perfect archetype of a

person who just like ships. He got into coding after being essentially retired basically just for the fun of it. He he

he discovered agents and that like sparked again this joy of like creating stuff and he created a ton. Another

example kit he built like a crazy person. All of his apps that he builds

person. All of his apps that he builds are fully monetized.

>> This is the kind of indie hacker that just like builds over a weekend a thing like uh sto which is essentially super whisper uh like a tool where you can speak and then have text and all this kind of stuff. He combined to AI models,

can do all kind of cool stuff. And he

sells it for oneoff uh payment. $50 and

you're done. The reason is because he wants to get rid of all subscription rappers. And he also created a whole SAS

rappers. And he also created a whole SAS site project community. Uh they call it a tinkerer club with like thousand plus people already. The whole goal of this

people already. The whole goal of this club is get people together who just like want to automate their home, want to build new apps, want to build like whatever. Basically, their goal is to

whatever. Basically, their goal is to like rebuild the world around them, the software world around them. And in a weird way, it's kind of like you versus them. Now, it's like your great idea for

them. Now, it's like your great idea for a little small SAS app is their weekend project. Let's look at the other case.

project. Let's look at the other case.

Let's assume you're a founder who wants to actually be able to fundra for the for the startup and like right now this is really really hard, right? If

software is no longer worth anything in extreme case like per hyper bowl speaking domain knowledge and sales marketing is the main thing that you are actually pushing now or differently put

why are companies still buying SUS or different like actually why are enterprises still buying SUS it's a domain knowledge that you productized it's a ton of decisions that you did

tradeoffs where maybe there's not a correct solution there's just like one solution trade-offs trade-offs trade-offs decisions decisions decisions you all put them And then you use a ton of marketing and sales and branding to establish a trust that they actually

want to use you. While researching for this video, I actually found out this was a real ad. I always thought this was more like a saying. This was an actual real ad that IBM at some point like ran.

Keep in mind like yes, the typical big company now can slop code everything over the weekend. Yes, that's true. But

making those kind of decisions is really really tedious and nobody wants to do that if they don't have to. So there's

still a market for people who want to buy that. So what I'm trying to say is

buy that. So what I'm trying to say is like if you want to do SAS build for the enterprise and what I would do personally is I would pick a vertical go deep at get a ton of domain knowledge

add all the workflows add all the trust and compound compound compound basically build a mega app. Uh this is a good example this is Legora. They basically

started as like chip for lawyers and nowadays they're like AI for lawyers and soon they're maybe like everything for lawyers. They're like one of the fastest

lawyers. They're like one of the fastest growing startups right now. And maybe a one take away from this video is like build lagora for whoever you know like doctors, dentists, realtors, clowns or anybody else, right? And keep in mind

like why are they buying it? Because of

the domain knowledge that you added to it and the trust that you built. But

what if you're like me and you don't want to spend your life doing enterprise sales? You just want to build awesome

sales? You just want to build awesome software, right? You have these three

software, right? You have these three paths. You can like build products for

paths. You can like build products for the love of it. You go indie hacking, which I personally highly recommend as an investor myself. if you can go this path or if you want to go VC case like what is the actual frontier right now

and think of the rest of the presentation as like a request for startups where I don't know the real solution but I think there is something there in this space number one GitHub is obsolete right now let's start simple

you know pull request code reviews like the pull requests were always like kind of like bad in GitHub but now it's like horrible because all of that code isn't even the main thing anymore like there's the prompt before there's all these kinds of other decisions before the

whole review process of GitHub doesn't really make sense anymore for the world we're in. Maybe the future of um reviews

we're in. Maybe the future of um reviews is closer to something like linear and like figuring that properly out. Or

maybe it's like closer to an actual editor or cloud code. Maybe the f the next GitHub is actually a combination of cloud code or like a text editor like set or something like this and the review process and this whole thing

maybe in real time and collaboration all this kind of stuff. And because of that, in an extreme way to say like open source is almost dead that think about it like you as an open source company right now like you get dozens if not

hundreds of pull requests every day by people who haven't read your documentation, doesn't don't understand your code and most likely use some random AI agent to just push the thing.

Maybe it was even like completely automatic and no human ever looked at it and now you pay a good 20% 10% 30% of your team just to look at these PRs to be able to actually decide if you want

to integrate it or not. You get dozens of security vulnerabilities a week now, if not like potentially soon hundreds.

It used to be that open source is more secure. Now it's like actually better if

secure. Now it's like actually better if you closed source because quite frankly if you open source there's a high chance that you're already hacked. The other

thing is business model. It used to be for example there's an open source core and then there's hosting as the business model. It's actually quite easy now to

model. It's actually quite easy now to oneshot the hosting or it used to be in the case of Tailwind there's an open source core and then there's all these like modules and other stuff that were like the business model which you can

now with cloud code and every other autocomplete just like build on your own. This is a tough spot to be in right

own. This is a tough spot to be in right now. So I'm very like if you find a

now. So I'm very like if you find a solution for if you have an solution for that like it can be monetization it can be reviewing it can be security it can be hosting it can be anything else. This

is a big market because I sense that this will be running into a problem very soon. Another one. Mexican standoffs.

soon. Another one. Mexican standoffs.

>> You ready?

>> I'm ready.

>> One, two, >> three, THREE.

[screaming] >> OKAY. SO, BASICALLY, you have now in

>> OKAY. SO, BASICALLY, you have now in every company you have an Mexican standoff between all the different roles. Designers think they can replace

roles. Designers think they can replace engineers with code. Engineers think

they can replace with designers with code. And product managers think they

code. And product managers think they can replace everybody with code. But you

have this in every industry right now.

uh think movie making uh the script editor the director the VFX artist is this the the storyboarder like all of them think that the other roles can be like automated to some extent this is happening in every industry right now or

will over the next years and genuinely I think if you rethink collaboration decision-m if you rethink these roles the processes and all this kind of stuff there's a big opportunity in almost every industry another one we are right now in the age of adoption basically

every company has onethird of their people that 10x themselves through AI tools another third that could be and another third who never will, but you don't know which one is which. Helping

them figure this out is a big opportunity as cynical as this sounds.

What are you going to use as a metric because it's definitely not like lines of code or how much they ship, you know, because that can be now done basically for free. And if you're a software

for free. And if you're a software developer, like one takeaway I want you to have for this video is you're no longer getting paid for the code. You're

getting paid for trade-off decisions. In

a way, you always were, especially as a senior developer, but now this is absolutely true. Knowledge work is no

absolutely true. Knowledge work is no longer I know and I'm getting paid for what I know. It's like I get paid for what I decide hopefully based on my knowledge and then I automatically build. This is now your job. In the next

build. This is now your job. In the next years there will also be build versus buy will be like one of the most horrible discussions in every company.

You will have Bruce who says I can't just build this over the weekend. Then

there's somebody else who will argue for like just pay play please pay the monthly license fee for this SUS app so that we can continue working. And then

there's the boss who thinks like hey we can't just acquire this one startup I saw on like uh online. we can't just acquire them and have this functionality. The next use will be

functionality. The next use will be absolutely horrible in this. And a weird side effect of this is actually something we already saw in the past.

This is Facebook camera. If you have never heard about Facebook camera, no wonder. It was built before they started

wonder. It was built before they started the acquisition of Instagram and it was actually finished building and launching 3 months after they needed just needed too long. But now with the AI tools, you

too long. But now with the AI tools, you can actually way quicker build this. So

if you're currently in a discussion about an M&A with a company, expect that this company's R&D team is meanwhile testing if they can just build your functionality directly and in a worst case using your DD documents. Get an

NDA, get everything else. Another one is fighting over the pie. Over the last years, it was much more about growing the pie, extending a use cases, finding use case and all this kind of stuff. Now

it's much more fighting for the pie and as Pen Thompson says in his latest article, model makers will become the arms dealer for this. If you are something like Legora, you don't want to build AI for lawyers or trad for

lawyers. You want to build the platform

lawyers. You want to build the platform for lawyers that does everything like source of truth of documents, any collaboration, any communication ultimately maybe everything else as well. You want to be their salesforce.

well. You want to be their salesforce.

Do I think this will just like scale into infinity like uh Microsoft now builds everything and everything? No,

because of course not because of complexity. You cannot just like add

complexity. You cannot just like add more and more trade-offs, more and more features, more and more stuff to it.

Sooner or later you will have stuff that just contradicts and this is just like pure normal complexity and will always happen. Will this stop big companies

happen. Will this stop big companies trying and just like building a crapon?

Absolutely not. And this is like be ready for this in the next few years.

Another one if you're currently working on a new startup. What you want to do is like think about where models will be in 18 months. Especially if there's certain

18 months. Especially if there's certain things that are just not good enough yet, just assume that they will be. Do

not build for the current times and don't build feature sets for the current times. Build for something like it's

times. Build for something like it's available in 18 months and reality in 18 months. One problem I currently see is

months. One problem I currently see is that prompt injection will be bigger than SQL injections. And yes, the model makers will be better in avoiding them, but still this will be a massive issue in next years. And all these VIP coded

apps that we talked about like that the marketing team quickly built, they will become a whole industry. Securing them,

standardizing them, making sure they use the same infrastructure, they use the same standards and everything that they're actually maintainable. This will

be a whole industry. We also entering an arms race almost with token usage like lovable for example and we are entering a world where maybe the whole editor will be for free and they only will can charge for hosting because they have to

spend as much tokens in their building.

Think about surface area. This will be the new mode uh cloudbot here for example. It doesn't matter which model

example. It doesn't matter which model runs in the background. It's all of a sudden more important for matter or for opi or for anthropic to own ideally

clawbot than the other way around. Like

clawbot couldn't care which model they use. This is all of a sudden power. Like

use. This is all of a sudden power. Like

if you're thinking about hey I want to build something small or maybe like a new little smaller thing. Think about

what surface area you can actually own.

And most importantly don't forget have fun. Do it because you like it because

fun. Do it because you like it because genuinely you cannot spreadsheet your success. Obsession is completely

success. Obsession is completely underrated. You need to figure something

underrated. You need to figure something that you actually are passionate about and focus on that and get obsessed with it and so on and so on. Also, bias to action is completely underrated. Open

claw is a few weeks old and is completely revolutionizing the whole AI industry right now and it's like the talk of TOWN INSTANTLY.

>> JUST DO IT >> is basically what I'm trying to say.

Also, like don't listen to random YouTubers and don't listen to all this ramble. Like do your own thing. Be

ramble. Like do your own thing. Be

successful. If you're not sure about building SAS uh and you are anyway like hey I have to learn so much maybe get into robotics. I have a whole video here

into robotics. I have a whole video here why when I want to convince you why you should actually start a robotics company. Uh check it out afterwards.

company. Uh check it out afterwards.

Thanks so much for watching and see you in the next