Recently we were posed with an open challenge to build a solution for simulacrum that was able to parse the intent shared below. I have open sourced a repository that provides LLM based parsing for user commands, and has an example of the output of its execution. This took me about an hour to implement. The reason I built it was to make something clear. We are not a team trying to build an intent parser for status updates on X. The example relies upon a few external APIs, notably Moralis and Chainstack, as well as the twitter API. We used our LLM library for building the agent and our EVM library for the onchain requirements. The prompt could be improved in a multitude of ways, what I put together is a general framework but is in no way complete. If we were going to put something in production we would probably use multi-shot prompting as outlined in this blog post, written by a community member using emp-agents: I might also recommend creating a classifier for different types of requests from users before setting the prompt dynamically. I just wanted to show how that workflow could be built. Our focus is on decentralized intent execution and trustless execution environments for AI and Oracles, in order to improve the Web3 user experience. I have gotten pushback from a variety of builders on the importance of decentralization in workflows and my feeling is... what is the point of building in web3 if you're just going to build like it's web2 with extra steps? Most recently, we have been collaborating with the @oasis team on deploying agents that run inside of TEEs. This approach can be generalized, as we would like to enable more builders to develop with trustlessness and decentralization as first principles, and to be able to do this without having to sacrifice user experience. This is the most efficient way to create executors (agents, oracles, etc.) that are able to view the outside world, while still maintaining the flexibility of more standard development workflows. I would also love to see more ZK based solutions, but I'll save that for another post. The reason this is so important to me are pretty standard but I'll share a couple: 1. If i die, the project should not die with me. 2. I shouldn't be able to "accidently check my private key into github" and have the users funds be at risk 3. A rogue engineer shouldn't be able to cause damage to the protocol or rob some amount of user funds The goal isn't to protect the project from itself, it's to build with this concern as a goal. And it's totally fine if goals haven't been met yet, I'm not saying we've achieved all of this. It's just an important part of our process. I remember when I was at Meta, an engineer killed all of Facebook for like 8 hours. They definitely screwed up, but the point was to figure out, "How do we make it so no one can ever make this mistake again" and not to just fire that engineer and move on. We see projects make the same mistakes constantly. The goal should be to protect the project from singular failure points, not to just get better at personal OPSEC. I get that people think that teams that are doing something kind of similar on the surface might be mistaken for competition. To me, we're all working on completely different goals and outcomes, and that confusion can lead to people misunderstanding how different things begin to look as you compare the long term roadmap. We're largely an infrastructure company, we don't consider SocialFi a main focus for us at this point in time. We also hope to see dozens of teams running intent executors; we want to make it so anybody can create their own parser and support their individualized use cases. I really want all teams to thrive and I'm not here to encourage anyone to hate on anyone. To be real... we're tiny teams in a tiny ecosystem in the early stages of Web3 adoption. The competition is Centralized Banking and Cloud Behemoths, not each other. If one user decides to use web3 slightly more frequently, that is a win for everybody, regardless of who or what or why (as long as the user has a positive experience). The attitude that we're competing for the existing web3 users is extremely bearish for the longevity of this industry. If we're just playing hot potato with a few hundred million dollars, what even is the point of all this? It's a dog-eat-dog world and I'm wearing Milk-Bone underpants. Let's all try and work together.