Ethereum Foundation: Building on Ethereum | Austin Griffith at ETHOnline 2026

I'm Austin Griffith. I think a lot about how to help developers create games based on Ethereum. And I've been doing this for seven or eight years. And this work has changed a lot. I used to teach developers about Solidity syntax and how to, well, think about your code, think about your applications, about your react, about centering divs, etc.

But everything has changed a lot. We don't...We don't center divs. You use the Frontier model to center divs, and it's always a one-time centering. This isn't just centering divs once, it's centering entire applications once.

This is a one-time centering of smart centering of smart contracts. This is a one-time centering, well, of all this as it goes along . This is my share, I'm having a hard time understanding whether it says there or not that I'm sharing this for free. OK. Yes, we are on the screen in Ethereum.

We are fine. We are fine. Don't worry. OK. I did something weird where I pinned you, so your photo is looking at me photo is looking at me , but as long as everything is okay and working, then it's okay.

I'll just keep chatting. OK. So, um, the gist of my talk boils down to about 5 minutes. And I can explain the basics of what I want to cover here in just a few minutes. And if you manage to get something, here are those first few minutes of conversation.

And that's basically what happens when you build something on Ethereum, you're usually building some kind of marketplace where there are multiple participants using it. You're usually building something that allows people to coordinate their actions in a new way, or allows more people to coordinate their actions in a certain way, or allows for more decentralization around coordination. And, uh, one way to look at it is Uniswap. Everyone knows Uniswap, but before we had Uniswap, it was essentially an order book. And you would have this order book, and it would be a centralized order book.

Even with smart contracts, where it was possible, you could create a signature that said signature that said you wanted to buy something and someone else wanted to, you know, there's a creator- creator- buyer system. Eventually it is justified by a smart contract, but initially it was still on some centralized machine. And problems with centralization are starting to appear there. If someone wants to censor you, they can just press a button on this centralized machine and you are no longer allowed to use it. If someone wants to tamper with your stuff...

Now we use cryptographic signatures in this case, so tampering will just break it, but it's not like they can, you know, trick you into buying something. They could trick you into signing something, but that's a whole other thing. But, anyway, this idea of ​​decentralization really boils down to this censorship-resistant, open-source privacy security. This idea of ​​the harvest, this idea that you can create something and you can walk away from it can walk away from it , and it will work forever. And this is a test of the rejection of implementation, and it's kind of like the power of decentralization.

Um, I had Ross on my podcast recently, and he was talking about how he even builds frontends into his smart builds frontends into his smart contracts, so everything is completely decentralized. And he kind of referred to it as laziness, which is really funny, because once you create something that's completely decentralized, you actually move away from the process and it just works, right? If you can't stop it, which is a really great feature. So, usually the things you're going to create on the blockchain need this property. They need this property that allows many people to coordinate in such a way that they only have to trust the code and not each other, meaning at some point they probably become somewhat hostile towards each other.

And it just happens on a grand scale, does n't it? You can't trust a random person on the internet, but if you can both use a smart contract to, um, interact with each other in a way that you don't have to trust each other, you only have to trust the code inside the smart inside the smart contract, then that allows you to do other interesting things. And this is what you need to think about when you are going to build on Ethereum. And this page that I shared here is called speedrunethereum is called speedrunethereum . It was a training program that we sent people through, and you went through it, and you created thing after thing.

In fact, to go through the service and understand everything, you would have to create a bunch of things. And that's still the case, but what you can do is just open speedrunethereum, give it to your agent. If you take on any of these challenges, there will be a thing you can give to your agent after connecting your wallet. Uh, you can give it to your agent, and you can start with a slash, and you have a tutor who understands all of these things and a bunch of other knowledge about Ethereum, and they can tell you about different things, and you can ask any stupid question you want, and they can answer it. So, it's just a much better way to learn, a much faster way to learn, and a lot of people don't write in Solidity anymore.

This could be a hot deal. I don't even look at the code anymore, not even Solidity. But don't do what I say. Like, understand and understand what you're building, and make sure you audit your code. And if you're going to have millions of dollars going through your protocol, obviously you need to go line by line and understand your Solidity.

You need to do audits and all that. But AI is pretty good at writing smart contracts, and it will get better and better. And AI is quite good at auditing smart contracts. And AI is pretty good at creating, you know, disposable front ends. So, a lot of what you need to learn as a person is just conceptual, what you need to do.

And for that, you can go to Speed ​​Run Ethereum. So, head over to Speed Run Ethereum and essentially take a speed run. Connect it to your AI, go through each of them. They are built in a way that gives these moments of awakening about how we coordinate, what we create, and how that helps scale. And so at some point it had to be like you go through one after the other and study all these moments of awakening, and it takes you 6 months or 2 months or however long it takes.

You can do it in about a week or less, right? You could get a lot of information in one day just by playing. But you as a person need to understand the concepts. Decentralization concepts, why decentralization why decentralization , test for abandonment, types of things you build on Ethereum. Uh build on Ethereum.

Uh , one really interesting one, if we want to dig a little deeper, is if we look at overcollateralized lending. Overcollateralized lending works in such a way that there is a liquidation function. This means that you, as a platform, say, create a protocol and can borrow tokens. But you can, you want, you never want the tokens you borrow to be worth more than the collateral someone put up. Excessive collateral.

So how do you make sure this happens? Essentially happens? Essentially , you have a liquidation function where when the value of someone's borrowed stuff starts to approach the value of the stuff they put up as collateral, you can press a button and it sells all of their collateral to cover what was borrowed, and your protocol remains intact. At first you might think, "Okay, we'll call the server and we'll ask it to be liquidated when the time comes." But this doesn't always work.

Sometimes your server is down. And if you don't do that, if you fail to liquidate someone's loan that goes bankrupt, your protocol is going to lose a bunch of money, and you just can't afford that. You cannot allow such a scenario to happen. So such a scenario to happen. So , how do you guarantee that the button will be clicked?

Guarantee that the liquidation will take place. The epiphany here is that you need to provide incentives. And incentive design is very important for building Ethereum. You need to write the rules correctly and say that if this loan disappears, loan disappears, anyone in the world can press this button, and that will eliminate the collateral to cover the loan. But the key thing here is that whoever presses the button also receives a certain amount of collateral.

And this is an incentive. This is an important part. This is the key element that makes everything work, no matter what. Even if all your servers are down and there's a huge DDoS going on and who knows, who knows what could have caused half the world to collapse right now, but, you know, a shark would eat some cable in the ocean, right? But somewhere, someone will have access to the blockchain, and someone can push that button, and someone can prevent that credit from disappearing, and that will preserve your entire record.

And this is the kind of case where you can just write the rules correctly, provide the right incentives, and walk away, and the thing will always do what it needs to do. So, these are moments of enlightenment, of concepts. There are a lot of concepts. You need to go to speedrunethereum and view them all. I will just give a few examples.

Prediction market, stablecoin, ZK voting, overcollateralized lending, oracles, DEX, dice game. There are a whole bunch of things here that you need to learn. Dice is fun because randomness is really hard if you're trying to make games. But all the concepts are here. Go to speedrunethereum, spend a day with your AI, dig around, and just understand all the concepts.

Even have your AI create a little HTML tutorial for you that says, “Oh, we just learned something new. Add this to the textbook." Oh, we just learned something new. And then go back to this tutorial a few times. You will understand the concepts pretty quickly.

Then, I think the advice I get from a lot of people I talk to is that you need to play. You should use these things. You don't... Let's see, Adam from Fake World Assets said something like, "You're not going to write a good novel if you don't read books, right?" And you won't create a game that everyone will like if you don't play games.

You need to play games. You need to use apps. Connect your wallet, exchange, borrow, buy poopcoin and try it all to understand. Of course, poopcoin does not need decentralization. Let's, you know, put that aside for a second.

We're going to play around with some things that aren't entirely aligned with yield. I mean, you just have to play around with everything play around with everything to understand what things are currently on the chain. And if you get used to playing with things, you will learn to create them better, understand what needs to be created, and what the pitfalls are. So pitfalls are. So , I said 5 minutes, but it takes longer .

Conceptually, Ethereum is a speedrun. As a person who is learning Ethereum, you will learn a lot of things. Then, when you're ready to start building, when you 're ready to let your agent do the work, hopefully you won't be creating with your fingers and typing code anymore. If you do, good luck. I think you might have a problem.

Maybe you know something I don't. But hopefully you're using something like Claude and Codex. And the next step you want to take is to let your agent copy and create things for you, and create a bunch of things. So, the concepts are the Ethereum speedrun, and the tools, the ability to quickly create things, are available at ethskills.com.

So, ethskills ethskills.com. So, ethskills is a library of skills that you can provide to your agent, and you can just start discussing the idea you want to implement. Many people I talk to do this in two stages. They are looking for Claude or Codex and working on a plan.

First, they come up with an idea, think about it, maybe even think about what kind of experience I would like to have. And then they work backwards to some kind of prototype, a blueprint for the prototype, and then they ask the agent to implement it. And you basically just talk to the agent for a while, give them these skills, come up with a good plan, hmm, I think a lot of people, and I like to do this too, I like when you use Claude, Codex or even some local stuff, you want them to argue a little bit. You do this, let's say we made a plan with Fable 5.1, and we're quite happy with it.

Go to Codex and say, "Codex, this is a plan we're planning to create, what do you think of it?" And maybe even another document will be proposed that, um, maybe will be critical of the first document. And then you can go back to Claude and say, " to Claude and say, " Look at this other document and let's think about how we could, you know, revise our plan." Until it seems like everyone is happy. And then you start, um, creating and giving your agent eight skills.

You can even just copy this one task, paste it, and say, " and say, " Create me a Perp Dex." Insert this. Forward". Yes? And then you start.

There will be a lot of things that it does quite well, but for now it can create a disposable app that you can connect your wallet to and use, but you should be very careful with that. It can host it on the base, on the mainnet , Robinhood or Arbitrum , Robinhood or Arbitrum , hmm, wherever, depending on what you're building. Um, but you should probably use it use it . You should use it often use it often before someone else does. You should audit the smart contract before a lot of money flows into it.

Um, here's an audit for $1. If you want a quick and easy audit, you can use the $1 audit and pay one USDC and my agent pipeline will conduct it for you. Of course, this is no substitute for a real audit, but 2 years ago it would have cost $10,000, and now it costs $1. And that's a pretty crazy leap in technology. We live in a pretty interesting, uh, time zone where around November 2025, things became, uh, basically, LLMs good enough that they could start creating these things for creating these things for real, and there was a lot less, you're absolutely right, the need to keep it in perfect context and so on.

So, the launch was around November 2025. It's really started to turn into what I call the summer of the idea guys, where the hardcore developer is almost a bit behind because the guy with the ideas can go to Fable or Codex, give them ETH skills, ask for a product, and get a prototype. And this is also, by the way, something of a revolutionary approach to venture capital, isn't it? For example, if you don't need to raise a bunch of funds to hire a team of five developers to build something, you can use ETH and Codex skills and you can build a prototype and you can get initial support without even raising any funds. You need about $200 for this, right?

So at the moment there are things that are much more accessible. Here is basically my story. I'll come in and we'll talk. Well, let's talk about some interesting things about some interesting things when we're done, but the main thing I want you to take away from this is learn the concepts, get up to speed on Ethereum, walk through each one of them, whether you're actually building them or not, pretend you're building them, think through the difficulties, be sure to critically evaluate what you're building. Go through each one and go all the way through , and on the other hand, whether it takes you a day or a week, you'll have a lot more knowledge about Ethereum and what kind of things you build on Ethereum and how those things work.

And then use a bunch of things, try a bunch of things on Ethereum if you haven't already . And then when you're ready to let your agent play around, use EthSkills, whether it's Claude or Codex, give him EthSkills and create a bunch of stuff. And this cycle where you learn a bunch of concepts, play with a bunch of things, and then start creating a bunch of things, you just keep creating until something sticks. And you will see that this cycle is very successful. Good.

So, uh, let's talk about something interesting. So, I have, uh, it's strange that it's gray. I wonder what I did there. It says " did there. It says " open episode" instead of "watch episode".

Maybe it's still on the air. I probably need to make a transaction. So, my podcast on Ethereum. I actually need to make a transaction in the mainnet contract to go live. And then all these podcasts are published, uh, to IPFS and then indexed on Ethereum.

The entire website looks like slop.computer, but it could be slopcomputer.eth. So I do it very much like on- it very much like on- chain, kind of like Cypherpunk, like doing podcasts. And then, uh, the environment itself looks like this, if you see me.

I am here, in this environment, and also on Slop's computer. But I have people here, and I have this really interesting Slop environment where we have multi-signature, a shared browser, and all sorts of games like poker and putt-patt, but we also have cool Cypherpunk stuff like Interfold voting and secure transactions, and, your own wallet that you can communicate with. This is really great. There is something like communicating with your wallet. I can just go into the chat and say, "Hey, trade this for that," and it'll set everything up for me set everything up for me .

So why am I showing all this, I interviewed 50 people over the last 80 days or so days or so . And I would like to think of these people as Slop operators. These are people who create with the help of artificial intelligence. So now I want to go over some concepts. Let's see if I can find some lessons.

Yes, Slop lessons. Okay, cool. So, I just said to my AI, " said to my AI, " Hey, I'm going to give a talk soon." I want you to review the transcripts of all 50 episodes." And highlighted some of the lessons we learned along the way .

So, uh, I'd like to share some of them with you. So, I think we'll go back to the most interesting things first, to the 14 things, without the story arc. I think that's how we'll end this. We'll continue for about 5 minutes, maybe another 8 minutes? Everything is fine.

Everything is fine. Everything is fine. Do n't worry. Are you resting? We are resting.

We are resting. Okay, honey. So, uh, the first thing that's quite interesting is the different types of psychosis. So I would ask you in the crowd, do you think you have AI psychosis? And most people would answer, “No, I don’t have AI psychosis I don’t have AI psychosis .

” For example, "I know someone who has AI-related psychosis and they're like best friends with GPT-4.0, but not me." And we'll look at that later, because I think there are a lot of different forms of psychosis. But the first quote, and this is But the first quote, and this is probably the most quoted on Slop Computer, Slop Operators, and the people who run Slop Operations, is that there has never been a worse time in history to be a junior developer junior developer . But there has never been a better time in history to be a solo founder.

Going back to what I was talking about getting support, anyone with a great idea and 200 bucks can now build a prototype and get initial support. Well, that doesn't quite meet the product market requirements, and we learned that in the south, computer support doesn't equal product market fit. Special thanks to Jesse for this. But the initial idea is that you need to stop thinking about adding to your resume. You need to stop thinking about hackathon rewards, instead collecting multiple hackathon rewards to pay for your flight.

It's online, so things don't work quite like that. You should consider going to a hackathon to create a product that people use. Create something people love and instead of worrying about getting your flight, you'll be thinking about choosing the private jet you want. There is a huge difference here. And that's about the way of thinking, right?

As for the difference between a junior developer and a solo developer and a solo founder now, there isn't much difference between the two. This is your way of thinking. You need to think less about impressing someone or showing your resume to someone who likes you and wants to hire you. I wasted probably ten years of my life on this, and I shouldn't have done it . I should have thought about how do I create something how do I create something that people like?

If I create something that people use, if I create something that gets initial popularity, all these other things will fall into place, and of course it's a whole new battle, but it's a good battle. So, I think the most important thing is to create things that people like and stop getting into psychoses like how do I create things that make people love me and how do I get hired, right? Okay, let's look at the six types of psychosis. There are a few good ones. few good ones.

First of all, he thinks you're a genius, right? And I think that's generally the simplest explanation for AI psychosis: it tells you that you're super smart, and you think it's your friend, and you get into this cycle where you think, " cycle where you think, " Hi, good morning Hi, good morning , buddy." What are we doing today? We do it well. Everything we do is great."

And you're not very critical of things, are you? So I think to add the critical element you need, try to get them to argue. We've already talked about this, but think hostilely. Think critically. Don't believe it when people tell you that you are a genius.

Like, take this thing from Claude and the code from Claude, give it to Codex and tell Codex to tear it apart. Tell Code to be absolutely brutal and find everything it can find wrong. And then take all of this and go back to Claude and tell Claude to be absolutely brutal with everything and find everywhere it could go wrong it could go wrong . And you will finally get some kind of decent answer. I think it was Jameson Lopp who said, “Give a man a law degree and he will know everything.

" Give a man two master's degrees in law and he will never be sure of anything ." I think that's a great quote, and I think it's really true here, don't you? We often see this when two law masters argue with each other. And by the way, I think we're debating this question right now, does AI make you smarter or dumber, eh? And I think that, uh, it again comes down to a mindset.

You can go to a law professor and ask them anything. This is so cool. For example, I build equipment here. I have Linux installed. I never did it for ten years because it was so annoying.

Now it's so accessible, you can just ask and he'll tell you. So whether AI makes you stupid or not is up to you. If you are curious every day , trying to learn something new, you will become much smarter thanks to this new AI. But I think another great quote that rings true is that you can outsource your thinking, but you can't outsource your understanding. And leave it for a while.

Your insight is extremely valuable. And going back to the Ethereum rapid launch, you can ask the agent to look at all these things and tell it, let's build something, but without your real understanding of everything, it will find something like a local maximum local maximum . He won't find the next sweatshirt that will do something incredible. And to find the next sweatshirt, I mean, some of the things that will do something incredible are extremely stupid, but in general, to find the next sweatshirt, you need to understand something. You need to have a deep understanding and let him do all his thinking do all his thinking , which you can let him do , but you need to understand.

Okay, so other types of psychosis, it's just fun because we're kind of filling in the finite space here. Stock psychosis, right ? When you know your subscription will reset in about 5 or 15 hours, that's why you try to spend as much as possible . You know your agent is just sitting there waiting for you and you need to go back and tell him something and you probably have mobile harnesses now and you can talk to him and so on but that's another one and I think the cure is—not thinking about using, right? An idle agent is actually worth nothing.

It is you, it is your mind, it is your understanding. You must be thinking, right ? So go ahead, touch the grass. I think a lot of the people who were at Slop Computer who are slop operators are kicking out their agents and making them do things. They outsource their thinking, but they don't outsource their understanding.

They go outside, sit in the sun, hold their children's hands, walk on the grass. But they also think about what I could build, what would break it, what would people like, right? And in that moment when you're there where your thinking is outsourced , the sun is shining in your eyes, you have your son in your arms, and you can think about things freely. You can play with your child in the sand your child in the sand . There are moments when these thoughts, like showers, really come out.

And I think that's a good point to make, it's something like, "Go find these things." Uh, another really "Oh, did I get circumcised there?" Is everything okay circumcised there?" Is everything okay ? Another really interesting thing is the local equipment.

I think, uh equipment. I think, uh , 0xsero said this, uh, that basically, you know, one DGX Spark will fix me. And I had this happen with my Mac Mini in February. I have about four or five of them. I actually use all my Mac Minis and they do good things.

But I think you'll find that if you become addicted to buying equipment, you'll have a lot of equipment on hand. Or your equipment will be doing things that may not be very important. Uh, 0x0, on the other hand, uses its own hardware to train and build models, making open source models more accessible to people with less RAM. So its trick is always in cyclic execution and assembly. So, you know, there are definitely these extreme cases where a machine that runs all day is actually valuable.

I think there's this really good quote from I: "You probably don't hear this, do you?" No, it doesn't know, it can't . You probably don't hear this, but Billy talks about something special happening right now, whether it's because of cheap subscriptions or because we're, you know, in a transition period between escaping the underclass, being permanently underclass, etc., but there 's something really magical about this moment, and his idea was that we're going to look back on this moment as a really good time for builders. And something will happen that could make things worse, like models becoming more expensive.

I do n't know exactly what n't know exactly what , but his instinct tells me that we will remember now as something incredible. So please enjoy this time. Enjoy these models, enjoy these seat belts, build cool things, think of cool things to build. Try not to let psychosis take over you. Another form of psychosis is looking at everything I've done.

I just spent, you know, 5 hours working on Fable 5.1 and generated 30 GB of MD generated 30 GB of MD files. Look at all my progress, right? And that's not true, is it? As if if no one actually uses it, if it doesn't actually serve a useful purpose, then it's not really anything useful.

As if burning energy doesn't count, right? So you can get into this psychosis by generating a bunch of data or creating a huge HTML interface. But if no one uses it, it doesn't count. And I think that's another form of psychosis. One hint–it's fun.

For example, people spend a ton of time optimizing their hardware, thinking that one more tweak might make this thing a little better, might unlock all these other things. I think the key here is that you should spend 80% of your time building useful things that people can use, and 20% of use, and 20% of your time fiddling with your hardware, trying something with LLM, trying new things. But if it's the other way around, you're probably not getting many results. And it may seem like you have a ton of results, but if people aren't using it, it doesn't count. A tank to kill a fly is fun, but I don't even know.

For example, I'm using Fable 5.1 to center a div. Uh to center a div. Uh , I don't know. I don't know about that.

Uh, let's see. What else do we have? Uh, patient zero, we have Claude. I use my agent for everything. Okay, let's do this quick "bang-bang-bang" session.

And by the way, all of this is available. I...I'll paste it into the chat here. Okay, let me just...Yes, I copied and pasted all the links you showed earlier.

Perfectly. Everything is fine. Do I still have time to go through one more quick little " quick little " bang-bang" thing? Yes, let's do 2 minutes. 2 minutes.

Okay, let's begin. 14 things, each of which is separate. Okay, so the smartest people I know use vanilla settings. Don't get too deep into harness maxing. Don't worry about it.

Vanilla cloud code, vanilla codex. This is what the smartest people use. Uh , traction is the new entry- level round, right? We talked about this, like, stop trying to impress some venture capitalist and just make things that people like and everything else will fall into place. You will have venture capitalists knocking on your door.

Uh, the best engineers have it the hardest. If you're the engineer who says, "I don't need AI. I still want to write code by hand,” you’re probably falling behind and need to hurry up. Er need to hurry up. Er , alpha lives in the training data.

Whatever you're building, for example, right now I'm building this AI audit pipeline. The data that passes through it is extremely important and can be used to learn all sorts of different things. So the training data is where the alpha is. When you build a system, understand how people use it. Understand where it fails and improve it, right?

Uh, verification has become a process, right? You need to check everything right now, right? You are letting the agent do your thinking for you, so you need to make sure it is correct. Immutability is the lazy option. We've already talked about this.

For example, a failure test—that's great. If you've created something that you If you've created something that you can abandon, that's great. Uh, yes. The models didn't get smarter, or the models got smarter, not faster, but everyone got patient. There is something like an interesting psychology.

Agents do the work, people say " the work, people say " no", right? You shouldn't argue with him. You should tell him to do a thing and then decide "yes" or "no decide "yes" or "no " whether it's good. Uh, do a pen and paper economy trick. Find someone who has a physical business.

Talk to your father-in- law, who has a physical business, and do a magic trick for him . Show him artificial intelligence. We have this tiny bubble of scumbag operators where we do all sorts of cool things, and the rest of the billions and billions of people in the world know nothing about and hate AI. You can do a trick for them right now and change their business. Stop reading the code.

I agree with that . A pretty good bar. It has always been that way. I don't know what this is about always been that way. I don't know what this is about .

You can even walk through all these points yourself. Hmm, it's easier to crack people than code. Imperfect systems that win. You can view all of these options. Let me see if I can find any others.

I think that's enough. Thank you for inviting me. I'm Austin Griffith on Twitter and Telegram. My private messages are open. Don't hesitate to contact us hesitate to contact us .

All my contact information is there contact information is there . Hearts, hearts, hearts. Thank you for inviting me.