Shifting to an AI-Native Ethereum Developer Stack

For years, we've been building our dev tools and learning materials for humans: docs explained what our toolkit did, we created videos, we taught developers with guided and hands-on experiences.

But the reality now is that an AI agent is almost always sitting in the loop across the whole development and learning cycle. So we stopped asking "how do we add AI to our tools", and started asking: what would our stack look like if we treated the AI as a primary user alongside the developer?

And if the AI truly understands the stack, a really interesting door opens: what if the person directing the agent isn't a developer? What if they're someone with a vision and the ability to steer an agent, but they don't have a technical background?

These questions pushed us to rethink some of our core primitives. Thinking about builders and not just developers. We've been doing exactly that across our dev toolkit (Scaffold-ETH 2) and our Solidity curriculum (Speedrun Ethereum).

Making Scaffold-ETH Agent-Ready

Here are the three biggest changes we've made so far.

Agent-first docs

We noticed in early 2025 that the main consumer of our docs was already an agent loading context before writing code. So we shipped llms-full.txt: the entire SE-2 documentation as a single flat file. Same information, just formatted for how an AI actually reads it.

We also added Cursor rules to Scaffold-ETH 2 as an early way to give agents project context. We quickly hit the limitations of that approach though, so we replaced them with AGENTS.md, which is picked up by Claude Code, Cursor, Windsurf, and basically any other agent harness. One file, every tool. Each conversation starts with all the context it needs.

Skills over scripts

We've started stripping out custom scripts wherever we find that a model just naturally handles the task better.

/add-extension Adding a Scaffold-ETH extension used to mean resolving package.json conflicts through hundreds of lines of template processing code. We replaced it with /add-extension, a simple agent skill built to work across different harnesses. Node.js handles the deterministic operations (fetching, copying), while the AI handles the judgment calls (merging).

/pr-create Writing a good PR is key when you are working asynchronously with a team. This file tells an agent how to inspect the diff, create a good PR body (summary, how to test, things to look out for, etc.), and open it via gh.

The same pattern extends to Ethereum protocol knowledge. SE-2 now has a growing library of domain skills: erc-20, erc-721, eip-5792, eip-712, siwe, ponder, solidity-security, defi-protocol-templates. Each one is a focused context file the agent loads before implementing the functionality.

AI-assisted code review

Code review is a perfect fit for "AI as a judge".

Take Grumpy Carlos: a subagent with a personality inferred from scraped BuidlGuidl PR history. Ask for a review, and it responds exactly the way Carlos would: thorough, strict, with actionable feedback, and yes, occasionally grumpy.

Human review is still critical for us, but AI review is a great first pass.

Rethinking How We Teach

Speedrun Ethereum is how most developers get started building on Ethereum. With AI in the loop, we're making that experience accessible to all levels, not just developers (i.e. Builders).

Per-challenge context files. Each challenge gets an AGENTS.md detailing the challenge overview, gotchas, smart contract structure, and frontend architecture. When a learner opens a challenge in their IDE, the agent already knows the environment.

AI Teacher Mode. Type /start in the directory and an agent walks you through it. It asks questions, checks your understanding, guides you without giving the answer away, and reviews your code at your pace.

We're also shipping /build-prompts: curated build ideas that you copy into your agent, get a working demo, then tweak it and make it yours.

Going from idea to working dapp has never been easier.

The Research Layer

The AI space is moving fast and there's a lot of noise. We are focusing on understanding the primitives of AI engineering, and adapt to whatever comes next.

raked is a minimal TypeScript agent built for exactly this: the agent loop, sessions, memory, tools, skills. It's under 100 lines for the core. Built to be read, understood, and extended to solve your specific use cases.

Experimental RAG pipeline on Arbitrum DAO governance data. It lets an agent pull relevant context before answering. We built it mainly to understand how retrieval and evaluation actually work under the hood.

Give it a try

We're iterating on all of this in the open. Check the code, the PRs, and try it yourself:

• —Create your own agent with raked: Read the agent loop. Understand how tools, skills and sessions work. Then, build your own agent on top of it.
• —Build your first onchain app with Scaffold-ETH 2: Check out all the available skills. Install it, open it with your favourite agent, and start building.
• —Try Speedrun Ethereum challenges with AI assistance: Available on Speedrun Ethereum as they ship.