New Ethereum talks, every Monday. The week's conference uploads by event, in your inbox.

Loading player…

Shipping Onchain Apps With Agents | Shiv Bhonde | ETHMumbai 2026

ETHMumbaiTue, Jun 9, 2026, 12:00 AM

This talk covers how Scaffold-ETH 2 became the go-to stack for agentic onchain development. We'll break down what makes a codebase agent-ready, AGENTS.md templates that give AI agents full project context, built-in Skills which SE-2 ships. We'll see the development workflow of pointing an AI agent at SE-2 and watching it build the full stack app. Follow Shiv on X: https://x.com/ShivBhonde Follow ETHMumbai on X: https://x.com/ethmumbai ETHMumbai 2026 is the BEST Ethereum Conference & Hackathon hosted on 12th – 15th March 2026 in Mumbai.

Transcript

We have been also shipping a lot of skills as well in the repo itself. So mind that you want an indexer in your repo. So like we just ship a ponder skill. Uh we also have a lot of different skills as well. So that you can just integrate all the stuff in the scaffolded repo as you need.

Yeah. So this is the prompt which I did yesterday. So like you just point it to a doc link. So docs.scaffolded.

io/kill.md. You just paste this uh link in your prompt and then you just give it the idea like your cursor agent or the cloud code agent you just give your further prompt and your harness the cloud code agent or the cursor is smart enough to get the skills uh scaffold the whole repo for your building. So this is what I gave it the prompt for and it's just a dab which I wanted to create for the ETH Mumbai so that everyone can play it and can we go to the national so these are the things which happens basically uh when you give the whole prompt to the uh to the agent it just loads agent.mmd in agent.

mmd we have listed a domain skills as well for the agent to understand what skills he need to use okay it needs to use uh indexing. So it will pull the ponder skill. It needs to use ERC20. So it will use the ERC20 skill. But yeah and it writes the contact and generates the front end for it.

And if we go so this is the the prompt which I gave in the earlier slide and this is the UI which it created. So like it just oneshotted the whole app uh in I think it took like 5 minutes or 4 minutes for it to create the contract and the generate the UI. But yeah and it's completely uh working like you could also do mobile resizing and everything. So that's the best part about scaffold is like it gives uh I don't know if I'm able to show the start but it also has these kind of components which AI understand and uh basically these are the components which are built specifically for web 3 and AI pretty nails it down and create this nice UI. If you give it a prompt like you are sure that AI will create something good out of it, you just need to make sure that your idea is pretty good and is structured a bit and if you give uh the docs link to the repo and yeah like this is one of the success stories.

So if you know Austin Griffith uh he has been using our stack uh and he recently developed this claude bot ATG uh which is an AI agent with wallet uh which is kind of a open claw agent and it has been shipping like crazy like it's just a bot which uses scaffoldly uh for developing the apps and it has also an wallet so it just started shipping app and now it has 99 public and 12 shipped apps around 85 G follow it also launched his own token since he's a bot and yeah like it's doing pretty well and it's under the hood it's using scaffoldly to create all these app or shipping these apps so like Austin is sleeping while his bot is shipping the apps so I think you guys can use this as well for the hackathon uh cool yeah like so yeah I mean you guys could start building like even if you are listening to some of the workshop you could just give a idea to your cloud code agent like just add the skill.md link in the prompt and just append your build idea and your agent should do the rest of the work. And if you have any doubts like I'll be here today so feel free to ping me or reach out to me on telegram. Thank you.

Automatic transcript — names and jargon may be misspelled.