DAPPCON 2025: AI-Enhanced DevRel - Francesco Andreaoli

DappCon 2025 · June 2025 · 25:05

Francesco Andreaoli presents AI-enhanced developer relations at DappCon 2025, looking at how AI tooling can help teams support developers building on Ethereum. The talk situates the practice within the broader AI for development and security landscape reshaping how projects engage their builder communities.

Speakers

Topics

From DappCon 2025, hosted by Dappcon Berlin.

Transcript

[Applause] Okay. Okay. Okay. Can you guys hear me? So, first uh big shout out to you guys. They stayed until the last talk. It's a big responsibility and uh you know like we're missing uh we're missing couple of I think 20 minutes and then we're ready for dinner. Okay, let's just start uh to speak a little bit. So, this talk is a bit different for anybody that uh um I don't know never hear me speak. Um I will speak about devil specifically but I also like trying to uh leverage all the uh interviews that I did with bunch of like agents builder in the space and how to optimize aantic economies workflow for specific food de let's just start with a quick um kind of like resume yeah for the people that doesn't

know me I'm Franchesco this is an old picture that I take in in Denver a couple of years ago uh in a buffy corn it's quite fun but uh yeah I'm leading uh devil a company called Consensus the mother company of MetaMask Infura Lineia and I've been around developer communities for quite a long time. I'm excited to speak about DevX and on boarding developers and really building sustainable uh and out of sustainable communities and I think now that we're going into kind of this agentic era we're speaking more about how we can automize uh workflow is super important that we're speaking the same language.

Let me start with my frustration. My frustration was when u um couple of years ago I was in different de meetups and everybody was speaking about different type of activation, different type of type of kind of like ais different type of like oh this is working this is not working. We're really not speaking the same language as I don't know how many people are actually uh developers here in the audience. I'm not really okay sounds good. How many de do we have in the audience? Okay. So, shout out to you guys that stayed. Um, so it felt kind of like randomly. I think like the space is a little bit kind of fragmented because you have bunch of like different type of organization building developer activation. Everybody's calculating the

same. That's why kind of like I thought about let's build kind of like universal framework for actually like debating and speaking uh the same language and I came with like different things like you will see the dev pyramid the builder funnel. This is something I've been working for the past couple of years but also like been working on a third framework. It's basically like uh how we can actually bring all the developer main work streams to more AI agent kind of like specific things. So let's get started. I want to give quicker context right. How many how many of you guys heard about metamas naps? Okay hopefully some someone okay sounds good. So basically like with naps you can extend the functionality of what is metamas today.

So you can basically do add-ons. There are a bunch of head-ons from kind of like nonvm head-ons from like snaps that are uh nonvm right you can actually like do things around uh cosmos you can do things around BTC you can do a lot of different things for gamers there is a lot of like snaps around uh decentralized identity there is also like different things around um multiple uh signing transaction there is a lot of like interesting things but the long story short is that I've been around building and drawing like snaps from the start and one things that I really understood was permissionless innovation is key. We need to build text stack that is permissionless and also is making uh builders around the ecosystem uh being

able to build on top and actually like further extend the functionality of what the extension is today. Uh but uh I just want to also state I think building cool product is great but also like building cool technology is not enough. What you want to do is also like uh building a driving developer ecosystem meaning also you want to kind of align and having a similar frameworks when you speak with different stakeholder. I mean when you are in your organization you will meet like people randomly coming from engineering you will meet people actually driving marketing product uh marketing people are driving product marketing managers right so a lot of kind of like different activation different TG you want to make sure you

kind of like speak uh the same language so basically the first framework that I think a bunch of people saw it already I've been working uh on this for quite a long time but basically this is all about defining what is needed in the ecosystem and I basically took this um Maslo pyramid I don't know many of you guys saw this before but basically I trying to build a Maslo pyramid for builder needs so I came out with something called hierarchy of builder needs and we actually went to different levels from psychological so being and some of the flagship hackathons having access to resources having basic education through your docs tutorial through the academy and then you are

basically going one level up So every time you're speaking with company one things that you want to make sure is that advocating for this leveling of activation without having the background and the second and the third you cannot really go up we cannot speak about building safe space for developer if you cannot on board developers on your dock if you don't have actually like credible CLI that people can just fork and build from it right so I think that was something that I really learned that was really uh interesting so just to moving on I also like build a paper if you want to see it I can also like uh send it over but basically then I'm asking myself what's going on around how we can actually like engage uh with people what were the lessons learned right I think you know attention curiosity think about

about uh think thinking about also like building safe spaces so no shield um uh meetups kind of like things that are able to make a developer being feeling like safe and I think like this we achieved since we're building like something called builder nights when we actually speak about specific specific topics, specific kind of like uh um uh design architecture. We started with 437. We went into layer 2. We went to around scalability, wallet ecosystem. We spoke about uh stuff like smart account specific. There were a lot of like different things. But also one things I learned was we need to understand in which stage we are and what kind of like devs we are targeting. So I actually thought about myself why we're not

actually building a framework that is encompasses encompassing this uh funneling. So, and then I basically came out with kind of like developer uh final specific and uh this was kind of like a lesson learned after after those years when I actually learned that building a safe space first for developers was actually more important than shielding a specific like product on the face uh to people and basically that's what what I did so was all about kind of like valuedriven on boarding first and then product specific conversion later. So the whole idea was having uh community calls that are based on the ecosystem having a very specific live coding session that are let's build an app instead actually let's use this SDK

first and when you actually have this kind of like safe space what you want to do is involve your uh um involve your kind of like product at a second stage but it's always very important that you're building your community first and you're driving this through values so I think that was one of the um uh kind of framework work that that that I think like is useful for people also like they're learning around how to build successful developer communities but also they got into the space they want to actually like drive uh strategy further to could be higher up could be like people stakeholder inside your organization um yeah I mean these are kind of some of bullet points like I think I cannot uh you know um specif specify enough you

know docs are super important they're evolving in different things and Um um and I think also like strategic activation from like ambassador uh program uh and also like automatization is uh is very important but also fostering supportive inclusive environment is something that I learned a lot around building I think uh 14 ambassador program so far including in all all my career and I think uh that was a big lesson learned that scaling is all about actually like driving and amplifying passion. So why AI is changing the game right? I think that's all the kind of like the fresh talk that I want to give today. Uh what is actually all about? What is this kind of like third framework that I'm trying to work on implying like uh agency andic

economy will um uh will drive innovation. We also like shorten up our timing in delivering a specific activation. So content I think everybody kind of like know already you guys throw like different content around LLMs optimizing for Twitter threads. There are a lot of like different things around content distribution. But I think one of the most important things that I kind of like learned soon as possible was you need to build docs that are like uh LM readable. You need to actually start think about some API playground for agents. So we need to actually serve both but also like agents uh in in the future. And uh and basically like an important part is uh Devil is actually in the middle right devil need to understand that there will be agents

that will actually like uh consume our technology, our API, our kind of like framework. So we need to think about also having agents consumable docs. So we need to serve both. It's a lot of responsibility right? There is a lot of things that probably like need to be done but also most important is a is a long-term game. uh even if we saw already you know like specific kind of like content being scraped you see already where is the source of through for agent you see stuff like Reddit coming back kind of like a a top of the list so there are a lot of different things so just to define a little bit like what what everyone want here so obviously we want to run uh things for humans there are like tutorials we want to have actually sustainable support

from discord discuss uh and all this kind of like uh um supporting system and then obviously we want to scale through actually uh human and relationship like based programs from ambassador program to like academy. So things are actually maintaining this safe space from uh the framework before but also like they're enabling uh getting like feedback from uh developers in the space. They're building their DAPs from kind of like zero to euro. And on the other side, we actually what what agents want they want a standardization. They want to have structured APIs. They want to have predictable schemas and a lot of like um API playground that we see today in companies that are definitely not uh predictable and also like JSON RPCs.

They're actually consumable from agents that have maybe also different type of like direction. So, and then I asked myself, okay, what is happening here around the uh the pyramid? what are the things that can be actually um uh how we say like speed it up around um around this hierarchy right I directly thought about okay the access to resources the basic education and trading uh training what we saw right now around for example metam mask or linear docs is that people are starting to actually like AI search much more around like 50% of like uh people are going to AI searches what does it mean they're also like expecting having like LLMs directly involved in the I searches. So like I think like we really cool like devil here in the audience we

were basically like checking uh different kind of like third party solution building our our own solution that has enabled people to directly search not through keywords but actually through specific uh um to specific kind of like LLMs and I think uh you know shout out to the to the guys like from cookbook kava there is a lot of like kappa there is a lot of like companies I think enabling that and then stuff like ambassadors commit engagement And then things like virtual hackathon. We are running every month something called metam mask and linearia developer cookoff. And these are kind of like evergreen uh hackathon. But the cool thing is that a lot of like prizes are delivered directly by agents. A lot of coordination a lot of first-time submission to check fresh code to check

actually what people are building. They're actually like um um due diligence by agents. So there is a lot of categorization. And then I asked myself what categories I came up with these six categories that can be optimized around um around several specific frameworks start from developer education community ops something that I just mentioned hackathon activation so around kind of like could be a first round judging to actually prize coordination distribution um for example on the hackathon side we've been running uh something called metam mask and linear commiticals so something that is happening after the hackathon and is happening monthly and the idea is to give a space to winners to developers building cool stuff in the community to actually like uh show what they build in

front of like 10 15,000 people in those calls and I think has been very valuable and I think the coordination can be always uh optimized and then stuff like feedback loops content distribution and then um that I see I have seven minutes here I'm just trying to kind of like speed it up here but basically like I try to bring what are those uh categories and what are those warstream example I think I kind of like mentioned some that are really like kind of like passionate about is stuff like for example translation. We are also like translating doc specific and also having version directly uh printed in in the LM uh version control access. I think that was also super interesting. We're running also like monthly release notes to the SDK and some of the docs are out

updated in different languages. So they're actually quite good. Content distribution is something I'm very passionate about because I think the source of through is having your I don't know 30 minutes 45 minutes interview with a founder and then you distribute the content in different kind of like uh u um different kind of like versions right we're using like for example NN8 to have some of those workflow enabled they're actually like interacting with different NCPs is actually very very powerful obviously this is not like kind of optimal right we're going in this direction because it's kind of like needed Because people obviously want to have uh content that is a better suited for them but also they want to have personalized on boarding personalized journey right and you see that already

with uh with doc specific for example we have for meta mask SDK we have something called a CLI that you can actually build and uh build on top and we saw different kind of like feedback people actually like to use the CLI in different ways and I think um developer analytics is also something that on the lamb side can be helped a lot. So uh why this kind of like their framework? I think like uh you know I mentioned already like uh I think um we will speak about agent we are starting to speak a lot about agent in this different field. I think there was no kind of like proper breakdown of de workflow in the ecosystem. That's why I thought about okay what can we

actually optimize and have standardized around our kind of like developer relation professionals in the in the ecosystem and this is not just about web three right everybody can take this this out and basically I I came with this boom so this is kind of like a human progression what you can see here from the left is kind of like humanled going until kind of like the agent native right so you have like a column that is specific humanled AI assistant and agent native so we are going in a direction where we want to optimize in terms of like uh um having agency actually running a specific task and we actually saw some tests already I think like last year in December we had a company called Gaia that was that is doing the

centralized AI I don't know if you saw it but they run an 100% agentled hackathon judging hackathon distribution were actually super interesting and then you see also other um other tests for example Alo capital from uh from Kevin from uh from Gitcoin they're starting actually to build retroactive funding and evergreen funding totally optimized and uh and driven by agent so I think that's super interesting what is happening around also on the funding side but I just want to kind of like uh dive into the specific uh of like what is happening around those war streams so just to give an example I think you can take a picture I can explain a bit more uh deeper around for example the the

content and ops, right? Obviously, like on a human ledled basis, you will actually like build your videos, tweets, blogs. Uh but what you can actually do uh is actually much more. You can do auto uh summaries. You can actually run uh voice voice over to make it like consumable for other stakeholder in the company. You can actually have a snippet of videos. You can have actually have different kind of like distribution of content. and then stuff like from from Aaton and activation specifically you can uh you can have uh frameworks that are going more and more on the automation side. So I think this is helping a lot of like giving a breakdown of what is happening around the the ecosystem in terms of like what can we optimize and I think it's also helping

stakeholder understanding how to optimize your organization and how to actually like build a functional uh developer relation team and advocates around the ecosystem and uh as I said you know there is a lot of like things that are that are happening but uh one things that I learned is that we are going in that direction things are not optimal I think you want to test it out build some experimentation h but I think you also need to drive home three things right having frameworks is also helping you your organization you as devil if you're in the audience pitching this to your stakeholder but also making sure you have clear mind where your execution strategy you have clear idea when you're going to anaton what you will do after

the hackathon how you can optimize uh you have clear idea if you're running like uh like stuff like a builder night or if you're running like a community call, what you can do after how you can build this kind of like evergreen cycle where every developer in your ecosystem is able to um be feel kind of like safe but also like supported. I think you know having your door open uh but also like having this support always open is is super important. So what we learn today I think like you know you saw a pyramid uh that uh is basically like what we we are doing what devil is doing in the ecosystem you see this level you can actually learn understand why is important and how to proceed from level to level don't start from the top trying

to make understand you have your foundation uh rock solid then we saw a funnel the funnel explaining you when you want to interact which type of developer and what type of activation you want to do first and I think this is super important trying trying to understand build a safe spa safe space first and then uh convert your developer don't just go to workshop to random developer and shield your SDK I think this is the most wrong things that you can actually do today and then we saw a third things grid that actually tell you what can be be automatizable what can actually be uh native and more assisted so and I think uh I want to kind of close up uh before everybody's leaving and have some question just want to

close up with one uh uh one question, right? Are there some risk in this? Will AI take our job? And I want to close up with uh this uh this picture here. Uh they don't appear to want to take over. They just want to dance. It's a bit rhetoric, but thank you so much everyone for your time. Thanks so much for for inviting me here, and I hope you learned something today. [Applause] All right. Do we have some questions? right over here.

Hi Franchesco, thank you for the great talk. Have you implemented yourself some of these automations? Yeah. Yeah, this is a good question. So basically what we're doing for cost and content distribution where we we're running this workflow around the NN8 uh for example for hackathon due diligence for the developer cookoff for example we're running a first round of due diligence uh run by basically specific agent so what you can do is uh uh we're running against kind of like fresh code and also some of the use case they're are done against uh the the bounty verticals. Um so those are kind of like the two main things that are on top of mind. Uh but I think the most important for devil

specific is uh documentation documenting things and also like building this distribution of this content. Maybe next time I will go to dance you know the the agents will uh will start running things. Uh thank you for great talk. Uh I have a question. What did you see? what didn't work with AI what you kind of you know predicted yeah yeah I mean this is a good question what so I've been running those um interviews uh building uh agents with this uh basic frameworks from the start um I think what didn't work was uh we started to see we are already seeing like bunch of

forks of forks of agents I don't know I'm I'm just literally like reseing defy summer uh from uh from couple of years ago, right? There are around 400 Eliza forks the building defi agents. I think it's really um like a space evolving fast, right? Uh but my most important kind of like concern is that we need to make sure that we defend the right for kind of agent consumers to understand which agent is actually providing true value. It's also I think it's also when you're speaking about Defi or Social Fi, we're super early. I mean we saw Giza one month and a half on production and he's the only kind of like agent that I see uh that really provide value. One other kind of like lesson learned that I

see is that we need to think about also agents that actually provide value to protocol specific instead actually like B2 kind of like C um that was actually one uh uh one lesson learned there but I think the first part is what scare me the most that people will start seeing and getting kind of like ragged by forks of fork of forks. That's why before even speaking about uh let's build more agent uh and this is a bit different than this talk I think like um contributing to for example I'm contributing to a a dark model that is basically trying to run um uh trying to run uh standardization and attestation frameworks for agent meaning uh uh they will be able to build uh a

basically like a unified repo where you will see which agent is kind of like the the OG agent that run on the the first timer code and I think that's is super important because uh if we're going this direction we will have a marketplace of like 10,000 agents doing the same things and will be difficult to actually understand which one is providing true value and is one which one is not dragging us right oh there there you So quality question is what make this audience awesome?

What are what are some of your like most important or like interesting use cases for AI agents in your opinion? I think I think optimization community ops I think what we s we always started right now is uh I mean we're running like uh this uh this ambassador program a lot of work has been done around social coordination uh I think uh community ops from the ambassador program side but also from the uh virtual hackathon side we're spending so much time involving different stakeholder in a first time due diligence uh and also like in running more distribution of the the hackathon.

So I think like uh this specific use case are are tangible ones. That's why I try to also break down devil specific work workflows that you know from from next kind of like week you can actually start automatizing this and that's why in the framework I put kind of like the difference between where we want to where we are right now on humanled to actually like assistantled there is a big difference I mean AI agents has been already been around for like web two stuff does they have PMF you know it's another story I think like what we want to do is looking at specific niche optimize specific specific niche and uh at the end of the day I think uh uh unique content the source of through content will be always uh people that

actually build uh uh how do you say like relationship based uh uh video tutorials it's always like has been always about people can underestimate can overestimate always a kind of like networking based things but I think devil specific now with all the aantic economies coming the relationship part of devil will play a more important role. Maybe this is an old take. Nice.