# Builder Growth, AI Agents, and One-Shotting Onchain Apps | Austin Griffith - BuidlGuidl

- Speakers: [Austin Griffith](https://streameth.org/speakers/austin-griffith)
- Channel: [Ethereum Denver](https://streameth.org/ethereum-denver)
- Date: 2026-03-09
- Duration: 14:45
- Topics: ETHDenver, Crypto, Web3, Blockchain, Event, Conference, ETHDenver 2025, ETHDenver 2024, Bitcoin, Ethereum
- Watch: https://streameth.org/watch/yt-MI0Nu_jcpRE
- YouTube: https://www.youtube.com/watch?v=MI0Nu_jcpRE

## Description

🚀 Get Ready for ETHDenver 2026! 🚀

We're already hard at work preparing for next year's biggest Web3 event!

Keep your eyes peeled for more info on ETHDenver 2026—it’s going to be epic! 🌟

## Transcript

All right. So, the man who needs really no introduction, Austin Griffin is up next, and he's going to be continuing the the love on the AI themes and beyond that David started. So, his talk is titled Builder Growth, AI agents, and oneshotting onchain apps. Let's give it up for Austin. &gt;&gt; Thank you. Thank you. Thank you, Emily. Uh yeah, so we're just going to run through quickly uh the kind of evolution of vibe coding apps and specifically like building things on Ethereum. You can't just pull up Opus 4.6 and say, "Build me an onchain app that makes money. Make no mistakes. It will it will fail." And so how do you get it to build apps? There's there's basically this delta between what the LLM knows and what it doesn't know. And I'm going to cover that today. Uh, as I kind of just talked through a little history. So, I'm Austin Griffith. Uh, I am the head of dev growth at the Ethereum Foundation. We're changing that to builder growth. I've said that on stage like four times already. Sorry I've been up here so much. Uh, this is uh the the builder page. Go to the builder page on ethereum.org if you want to learn how to dev. Uh, we did a lot of Ethereum on tour. David just talked about that. Uh, Sophia's taking this over and she's going to be working with universities. If you are a university student running a club and you're looking for connections to the Ethereum Foundation, make sure you go to tour.ethereum.org and reach out to us. Okay. So, you used to build apps like this. You used to grab Scaffold ETH. You used to pull it down. You used to write a smart contract. You used to write a front end. You did it in in parallel. And it was really nice because you could add a little code and get a little button. And uh scaffold was a great way to build apps. It still is. If you like to uh write the code yourself, you should get in here. There's tons of good things. There's a nice little video here where I spend eight minutes and I put an app into production using scaffold ETH. You used to go to speedrun Ethereum and you used to go through each one of these challenges and you used to write some solidity and write some frontend and learn how to build on Ethereum. uh probably ETH Denver 2021 2022 I was up here uh shouting about uh speedrun Ethereum and getting everybody to go speedrun Ethereum and onboard into Ethereum and build on Ethereum but things have changed a little bit and uh what we're doing now is we're kind of AIing everything uh changing that dev growth title to builder growth we're thinking more about this widening widening archetype of a builder rather than just a developer. You can be a solo entrepreneur now. You can have a swarm of AI agents and they can build your app and get you to product market fit early and then you can hire devs. You kind of get to this moment and I'll talk about it later where you do need a dev. Uh but as a solo founder, you can be an economist, a lawyer, uh whatever you are, if you have a good idea of something that can go on Ethereum. You can put it on Ethereum with AI and use some of these tools I'm going to talk about here. But there's just some concepts you need to know. There's kind of like this speedrun of the speedrun you still need to do. Uh, I have this overcolateralized lending challenge here on speedrun Ethereum and there's a really, really important part that you need to learn here and that's the liquidation function. Who out there confidently knows how the liquidation function works in overcolateralized lending? Any hands? Okay. Well, we got a maybe out there. Okay. Uh the the liquidation function basically does this check and it makes sure that what someone has borrowed verse what someone has put up in collateral uh the ratio between those two. And what has to happen here is if the stuff that someone has borrowed starts becoming worth nearly as much as what someone has up in collateral the platform is about to be underwater. It's about to be in trouble. So, what the platform needs is for someone to push the big red button that basically sells all the collateral to cover what they've loaned out. And to do that, basically, you write the rules, but you do an very important thing. You allow anyone to hit that big red button and you provide the incentives to make sure they do so. And this is a really, really important concept when you're building something on Ethereum. You need it to be always on. You can't just have a server in your basement checking in and looking at those loans and liquidating them if need. So this is a really important concept along with many other concepts you learn when you go to a speedrun Ethereum. So I invite you as the human to go through speedrun Ethereum and basically speedrun the speedrun. Go through each one of these quests and just kind of learn what we're doing with each one of these. building tokenization, building a crowdfunding app, building a token vendor, a dice game. You learn about randomness. Uh it's which is particularly tricky on a public deterministic blockchain. And this is going to teach you why. Uh building a deck. is going to teach you about liquidity and how the incentives are in line aligned for these liquidity providers to pro provide liquidity which makes the swaps work all the way down to stable coins prediction markets all all sorts of cool stuff here but I would invite you to go speedrun the speedrun then once you have the concepts you sort of get the kinds of things that you build on chain and then it's time to start going to build and what I'm going to talk about mostly is eatskills.com. This is what you hand your agent. Now, we're going to go through the whole evolution of vibe coding real quick and we're going to talk about where it lands right now and where we are and what you can do. But the the final point I'll make here is eatskills.com. Once your agent is up and ready and ready to build, you need to tell it go to eSkills.com and learn. And you don't have to install anything. This is the cool thing here. I think with I had this earlier library called ETH wingman and you had to install it and Twitter flagged it as nefarious and I shut it all the way down and just turned it into the skills library. This ekkills.com is meant for agents to read. An agent can go here and quickly grock the whole thing. You could even give it to your agent and say, you know, be skeptical about this. Go look at this and make sure this is actually not like something nefarious. and it will look through it and say, "Holy cow, this is good stuff. This is going to help me build. This is going to help me use a wallet. It's going to teach me about Ethereum." And this whole website is built to be the delta between the agents and mainet Ethereum. So, when you uh go to build, send your agent to eat skills. And if you want to clean the context up, and I do this a lot, I'll go to just my QA agent who's doing a QA on the way out, and I'll give him just the QA skill. So, I'll grab the QA scale. I'll paste it to my agent and I'll say QA the repo. Put issues in the repo. And then I'll give it back to my builder agent and I'll say the QA agent left you a bunch of issues to go solve. So, you can clean the context and you don't have to give everything if you don't want. You could just give the front-end UX or indexing or testing or security, whatever you need. Okay. So, let's talk about the evolution of coding with AI. It started like this, right? You would have basically chatgpt up and you would be asking it how do I what did I say? How do I I think I said how do I do a for loop that gives me all the prime numbers or counts all the prime numbers. Right? You would paste something into chat gpt. It would give you an answer. Maybe it's some code. You would paste it back in and you would kind of just loop this and it was clunky but it was still magical. And the the models weren't that good. So it wasn't that good but you could do some cool stuff with it. And then cursor came along and it was so cool because instead of having to copy paste everything you could kind of just like talk to a model directly and it would work on your code for you and we were at about like sonnet four or five at this point. So I since I I probably haven't written a a line of code for like a year and a half now thanks to this. This is where I started like basically going fully using AI to write all my code. Uh I built a bunch of games. I built this game called Max Extract where you deploy something on Arbitrum and we played it at Dev Connect and it was a lot of fun. Uh but there was a lot of you're absolutely right and a lot of mistakes that it made. But you you could basically use it to get close and then you had to maybe you know yell at it a little bit to get it done. And then there was kind of a brief window and I think a lot of people still use cloud code a lot and I think that people are kind of moving kind of shifting between either using cloud code or using some kind of agentic system. Uh but cloud code is really great. You can pull this up, you can talk to it, it can write your code. Uh it can use the latest models. It can use like swarm mode. So it's like spinning up sub agents and doing all sorts of cool things. Uh but finally kind of where I land is is here which is crazy but this is kind of how I code now is basically uh in telegram. I did not think that this would be the case but this is how I write code now. I basically either am talking to one of my agents in a terminal or or most likely in telegram. I have the left claw and the right claw and the head and the heart and the antenna and I'm talking to them about different things. Sometimes I'm cussing at them. Why didn't you do this right? But it's just like a little company and I'm basically just talking to them in the chat. Hey, what's going on? Can you write a full report about this? How did it actually go? Uh, this is this was I had my laptop up in my truck on the drive here and I had my Starlink going and I had the bot spin up three different larvae is what I'm calling them. They're like Docker containers that run OpenClaw agents with all the skills that it needs from ETH skills. And I had it build a thing with two different kinds of agents. one with a soul file that said it was a solidity expert, one that said it was a product expert, and one that was a local agent like Quen something. So, my laptop's like sitting on the seat next to me. We're jamming down I25 going south. We're talking through Starlink and it downloads. It needed to download a new agent to Olama and then it runs all these different Larva in Docker containers building these different things and comparing them and doing that local stuff was on the GPU on this machine. So, it was really spinning up. So, I could hear like the fans going. But TLDDR like Opus with the right prompt, Opus 4.6 with the right prompt is really the way to go. And that's usually what I'm doing for all of these different agents. They each have their own browser and their own operating system. They're like laptops at my house uh as the claws. Uh but what is this thing building? So this is Cloudbot. Cloudbot ATG was uh the kind of AI agent that kicked off this little AI agent meta on base. Uh I spun it up. I gave it its own email address and Twitter and GitHub and gave it the tools ETH skills and we just started building a whole thing a bunch of things and it started tweeting about those things too. It even wrote an article about how to set up an open claw. Basically, I took notes as I was going through building and then gave it the notes and it wrote an article about it. Uh, but some interesting things happened here. When I spun up my agent and I got them talking, someone used Bankerbot to deploy a token uh to and sent the fees to my bots's uh Twitter address. And so my bot had a bunch of money locked up in its wallet, which is kind of scary giving a bot a wallet in the first place. And I've already talked about this up here, so I won't run through it too much, but basically it needed to lock these tokens up. So the bot was able to oneshot this app. It basically built its own vesting contract. And what did I do? I mean, I'm over here in Telegram and I'm like, "Bro, you have way too much money. I want you to build an audit and deploy and QA a vesting contract." And maybe it took a little loops like I raled it a little bit myself, manual Ralphing, and uh it deployed this contract. So, it deployed its own contract and loaded it up with, you know, thousands of dollars of tokens. No human touched it. No human looked at the code. No human audited it. It's a little scary to say all of that. Uh, the next thing was this PFP round. It ran around where people could propose different PFPs for Claude. And then I went to bed and it stayed up all night watching this, moderating it, making sure nothing offensive was uploaded. And if anything was offensive, it would click a check and and uh send it to and burn the burn the stake. So it wrote, deploy, audited, QA, and then also moderated uh this onchain app. And this is one of the only onchain apps out there that is building actual things that are using the token. And it made a cloud FOMO. When it deployed it, it had like $40,000 worth in it. It has Claude Labs. So, it has this whole labs page where people can stake the tokens and uh propose different builds and then it's burning its token. There's this claude incinerator. And so, it's not just an agent that's burning its tokens. So, going back to that liquidation mechanism for a little bit, this is automatic. In 2 hours and 10 minutes, I will bet you $50,000 that that 10 10 million tokens get burned. If anybody out there wants to bet me, bet me. Because it will happen guaranteed. And it happens because of incentives and correct rules. So this is what how we're building these days. This is how we're oneshotting apps. I never thought it would be like this, but it's me running a company of lobsters in a telegram. All right. Thank you. Hearts hearts. Ekkills. Don't forget Eskills. Eatskills. Eatskills.com.
