Top repo explained: builds product + lands customers

0:00This guy [snorts] won the Anthropic hackathon with an AI-powered customer discovery [music] platform. Then, he open-sourced his setup so you can use it, too. It's called ECC and it fleshes out your idea, it builds the first version of your product, it creates marketing videos for you, and it even sells your creation to people you know. The repo has a massive quarter million stars on GitHub. He even used it to build a company that he just sold, and in this video, he's going to walk you through how to use it. I've got a link

0:30below to the repo. Let's get to it. Presented by Zapier, the AI automation company. Affan, what are we looking at here? So, right now we're in our orchestrator pane, and we're going to give it a task and build a product. So, it's going to use ECC skills, uh spend the first 10 minutes building using various skills such as plan, test-driven development workflow, um and we have it specked out, and ECC actually built the spec out as well. Then, what we're going to do is we're going to actually try to market this thing. So, we'll use some of the market research and also video generation tools to actually create a

1:05video, um and then use that video to then market uh on X. So, it'll find people in my ICP on X uh using more ECC skills, um and then it'll send them DMs with the video or message them asking them if they're interested to try this product out. And we're going to try to get this all done

1:22um within the 20 minutes that we have. All right. Let's kick it off. What's the product? So, it looks like the product that it chose is um a a analyzer or a repo analyzer. So, it roasts your pitch essentially. So, whatever it is, so if you're trying to kind of might be good for a VC to use something like this, or it might be good as a founder before you pitch to a VC to use something like this, and we'll see if like other founders are interested in

1:49using it. When you say it picked it, where did it pick it? And I see you've just kicked off the the prompt. Where did it pick it? How did did you get it to come up with this idea? Um it that was stochastic. That was completely decided on its own. Um I simply asked it to come up with an idea that would be applicable to me. And there is some level of memory and context that is um set up. So, it has some idea that I'm a founder, that you know, I have, you know, I've been building things for some time. And so, it decided that this would

2:19be and maybe I because I was raising rather recently, not that long ago, you know, those things can go into it. And then it was that in the prompt that you asked it to come up with the idea? So, that was um previous. This was like an idea I already had sitting. So, this was

2:35previous to that. Got it. So, you took some idea that you already had come up with before and now it's taking the idea, it's running with it, it's building it, it's helping you market it. Got it. What Why do I keep seeing these uh one password alerts come up? So, what it's doing is it's utilizing uh one password to pull environment credentials. It's a way to have uh environment credentials will be pulled

2:55securely. Um I use a CLI to do this. So, to keep everything secure so I don't have local environment variables laying around laying around everywhere. Um and it's just one way to have um it all working. So, what it's doing is like it was testing the different things. It didn't have to pull anything

3:10into the shell. Mhm. Um it was able to do that all itself. It utilized one password to pull in these credentials and it went ahead and it started it created this four box here. Now, this is in a tmux pane. Um and it's going to start um driving these panes, this orchestrator over here, which we are now going to minimize a

3:31little bit. It's going to start driving these panes and it's going to start building. And it's not as important what's going on over here cuz it's going to move kind of fast. Mhm. keep focus on the orchestrator. But as you can see, it'll tell us what's going on um over here. Okay. So, it's running those panes um on the right. We don't need to pay attention to it. We're just You're just using it to

3:54keep an eye on what it's up to. Right. So, I've kind of Otherwise, you know, there's going to be a lot of context switching that goes around. And this is the easiest way for me because it's like hard for me to keep track of like 20, 30 panes and it especially for people who claim that they're running 20, 30 plus agents in parallel. How do you keep track of all your work? Well, normally you might have some sort of streaming or orchestration service that's explaining to what's happening in every pane. And that's how I normally kind of [snorts] view things at a high level. I might

4:20also take a look with my eyes sometimes and make sure things are going correctly. And uh but I'll mainly uh take a look at the orchestrator and make sure that everything's going on. So, uh over here it's monitoring these different panes. It's claiming that the show's like pane one's cloud code has written and committed seven uh failing tests. So, they're not doing well right now. Uh to kick off the test-driven development. So, now it's going to actually have to beat these tests um and fix its own code so that it passes uh through the test. Um and there's more

4:52rigorous ways to do this as well. And so, you can see it's kind of building some code. Uh over here it's hard coding a few like general uh it looks like it's uh you know, um it's creating a some general market and some general search terms. I think this might have some This one might be also it has like a category for web three, had category for B2B and

5:12SaaS. Um and uh enterprise sales. Um so, if it's like a enterprise, you know, B2B SaaS type of product, and then it's kind of setting this like creating this tuple here um right now. So, of course, you could do a lot better uh with more data ingestion if we had were to give it a lot of time to build a product that, you know, um you know, learns. So, you know, iteratively learns over time. But this is like getting the right um everything down. And sometimes I do look at the

5:38panes to see what direction it's headed. But most of the time I am very hands-off now um with the way I code. And uh looking over diffs might just be done in real time as it's going to make sure it's on track. And if it looks good, then I just let it go on track. At the very end or when Sorry, go ahead, and then I've got a

5:56question. Yeah, at the very end I might go and review through some things to make sure that it's, you know, everything's good, but a lot of the time it's relatively hands-off now. This is using ECC, the repo that got you and I connected. It now has over 271,000 stars on GitHub, which is huge. It keeps coming up as a top weekly trending repo,

6:18which is why I keep covering it. How did you put ECC together? How do you describe what it is for people who are trying to get to understand what you're doing here? Right. So, I think the best way I want to skate it explain it is that it removes the startup costs or the time to time to product from an

6:42idea. Mhm. Because it saves all the repeatable, the the the workflows that not just me, but engineers over the past decade, two decades of software engineering have created. We've had We've created systems that are very well trained on to create tests, make those tests pop a pop, and make those tests pass so that your product pops. And you know, you're doing very you know, you you you're doing a very test-driven development. I just want to pull this up here cuz it's

7:12already built out. yeah. So, it already built out. This is a little like quick and it's very simple, you know, easy UI, but I just wanted to say it's a very quick to get to your product, right? And so, now it's even going to start doing tests. So, I want to like kind of look at those tests because that's one of the main skills. One of the main reason I actually built ECC was that I needed some level of deterministic workflows to apply that were rooted in engineering practices to my otherwise very free-flow style of prompting that I had gotten. And the models have gotten better over time at

7:47this. Um so, now a lot of these things they do naturally. So, what did we do? Instead of leaving ACC as it is, we started getting more and more niche. So, there was a lot of niche work, Kotlin, C++, low-level programming, finance, fintech, quantitative finance. There's a lot of other things that are not very well trained on. So, we had skills to actually, you know, make sure it follows particular practices that those industries and also um those

8:11programmers, researchers, MLEs, etc. follow. Um and this goes beyond that. Uh we made it so that you can now chain skills together cuz I don't want to leave my orchestrator to go use some other SaaS tool to do my marketing. I want to build the product. I want it to make a video out of the product using skills. I want it to post that video on my X or send it in a DM to a bunch of people that are in my network that it literally pulls the network uh that I have and finds the best people that would actually buy this and get my first few customers. And this

8:40is all doable within an hour now. I've already built this. It's already kind of building this You know, this is a really bad pitch. So, it's going to roast it. Uber for dog walkers but on the blockchain. It's ridiculous, right? Um but maybe maybe there's someone in my network. Uh maybe that should be my pitch. So, it kind of pulled that up. It's localhost, of course. And so, it's kind of creating this like rendered promo video here. Um It's But the the tool that it the tool that it's making that it's now trying to promote or soon will try to promote to your friends and followers is

9:09the roast your idea tool where someone can come up with an idea and the tool itself will roast it for them. And what you want to do now is That's true. And then what you want to do is send it to people who you know on X and say, "Hey, would you use this?" And see if they

9:23would use it. Right. So, I'm not sure how Claude is going to interpret this, if it's going to find my network that it thinks has bad founders and then send this to them as a way for them to be like, "Hey, like this is I'm I'm implying something like you should probably use this to improve your product. Or if it's sending to founders that got to raise and they got to think from the perspective of a VC, what is a VC going to say or someone else going to say and improve the product that way so just so that they're prepared. But Cuz you know This is um because this is another

9:50feature of what you've built. You want founders and creators to run their ideas by real people and you're forcing that in, right? Tell me more of Tell me more of what you've baked in because that is part of your ethos for building. Right, so um I think you know it's so amazing that you know with all this technology we can build incredibly fast. So the question then becomes after that is like you can build incredibly fast but can you actually create a product and then from a product to a startup and a startup to scale? And things kind of fall apart in between then and you know this is as someone

10:22who's you know started with projects and then tried to scale them and build actually like full-fledged products. What you need to do and you need to do this as fast as possible is iterate. And what you need to do is get a lot of outside counsel and opinion, not just your own self and and you know what what the agents are saying cuz they might start guiding you as opposed to you guiding the agent. So you need to keep yourself in check and you need to like understand who you're building for, the market that you're building for, the ICP. And you're also going to need to

10:45understand even if you have a certain idea in mind, if you're raising you might want to present the product a certain way because you know that this particular VC is specifically keen on this on this exact vertical and he's specifically keen on looking for one angle you present on a how you're solving the problem. Even though your vision might not completely align with that, you're going to sell it to that VC

11:06in that way. Mhm. And then once you get the check then you get the check you can build it out as you want, right? You can like get the rest of your vision going and that's just how the game works frankly. As someone who's like raised a few times and and also exited, that it's just that's the way you want to play so you have a different market that you need to play for. Um this is something we know that this approach works because you had just launched a company, you got community feedback and from what I understand from friends of people who I

11:34trust, you sold it. You exited. Um yes, yes. So, so we have we have joined a new new organization. Uh we're building out what we're working on before now at scale with the with the proper resources and and ability of this organization we have. Um you can take a quick look at here. It already created some DMs and I have maybe these guys get

11:52free shout outs too. you're intentionally pushing me off of talking about how you sold the company because it is early. I'm going to use that as my phrase. Let's take a look at what we've got here. So, what's this on the left? It said, "If finished" and now it has five people who you probably know. Do you know Youssef on number

12:07five? Yes, so I do know these people and it's interesting. So, it chose a bunch of YC founders. Interesting. sense, right? So, it found five uh YC founders. I know some of these founders. Um and and it and it wrote in my in my writing style and it created these like uh intros, which I think are actually not

12:28that bad. Okay. So, Let me see what it said. What what it would send to them. Yeah. So, over here [snorts] it drafted um and it said for Lance and maybe he had a tweet where you know, um YC partners had pushed them back on something, but it gives them this and it says ask whether these scores match what YC partners actually push back on when

12:48he was doing his demo day potentially. Um which who was the CEO of Traverse? It was in my network. So, he had another uh YC founder. Um It uses something from my DM history where I think I actually did once I was supposed to and I think I forgot about this. I was supposed to give him an intro to a Shenzhen supplier when I when I visited Shenzhen for building his I I think it's like eyewear. Um and and then asked how it would roast a hardware pitch, which is very very targeted. This is highly targeted. This isn't your when you open your email and you see the hey

13:21Afon, I know that you've been working on blank and you know they just pulled it from some like public, you know, like random data and it might not be accurate. This is very targeted. direct How How does it know? How does it have access to this? To your ex? So, as you saw earlier when it pulled up, Combinator is using one password. I have a bunch of things set up in my vault, and it's very good at this. And

13:39so, this is an incredibly good one, too. See, co-founder and CTO of Radius opens with Eto office offer in NYC still stands because now I recall I did say ask him to visit my office. I think I might have sent the comment. It might It might not have been DM. It might have been a comment. And ask whether the engineer judge is too nice cuz what they're building is something for

13:57engineers. Wow. Yes. Anya, I've spoken to her before, works on computer markets, which is what my previous company was in as well. And then it just opens with fellow computer markets person here, and then a comma Eto lowercase. Um and then you know, it choose some other uh ones as well. Um And then it allows me to just simply

14:15send at to do it. Um And you can also commit uh the the actual uh what's going on over here. Um and I think we should I'll ask it to before sending um play the video for us and see if that um Oh, there it is. So, you can see. [laughter] It made this like It just really, you know, last minute kind of like bang bang type of video where Yeah, Marcus I could already happened. That the, you know, cursor for blank is good. Four out of

14:5310. Drop this. Just only sell this part. Roast your pitch. Pitch roast. Five judges, one verdict, one change. And now we could like iterate on this. This is a one-shot it, right, guys? I one-shot it this. So, we could iterate this like But this is like boom, your pitch is getting roasted. These are what the what they

15:08said. This is like a way to entice them. They sent this We sent this over DM. Um let's see. Maybe I should send it to only one of these person that I then can explain that all this was for a video. So, like so that they don't think you're running this business. But yeah, let's

15:20an appler. Okay. Is it also Does ECC Do you still call it Everything Cloud Code or does it just go by ECC now? [snorts] Uh it just goes by ECC since it's applicable to any kind of system. Um but uh If I if I have an idea, I bring it to ECC, it will help me think

15:40through the idea at first, right? Tell me about that thinking through process which we hadn't seen, and then it will go through building that first model, the first uh prototype, finding people who we could go and pitch it to, get feedback. What happens before what we saw, which is like helping you think through the idea, and then what happens with the feedback that you get from

15:59people after? Yeah, so um it works best when you have like a good amount of a knowledge base or a context uh of your history. And what works is that um often times there's a few scales, there's also like a plan scale, and when the planning process of using the different ECC plan scales will help

16:17you uh do more of an iterative process. So, you can like say you have a baseline idea what you want to do, um and then um what it'll do is that it'll like work through you, did you think about this or did you think about that? Maybe be like, "Oh, does this exist or this this doesn't exist already?" So, it'll combine a scale of deep research, and deep research what the actual market looks like. So, now you're having an actual conversation instead of getting right to building. Once you go back and forth on the conversation, you keep looking at the new spec sheet and the new PRD, um and at times I've done

16:48honestly I've done I've ran a bunch of sub agents using this exact method, and I've built a 72-page almost uh like manual of the entire like vision of something before I even start building it, and then I can go reference it um back and forth and make edits, and then build it in small small chunks. And I think that component of it, which is the beginning component, we didn't get to see much of it here cuz it would take too much time. That's actually the most time intensive process, where you actually go back and forth and iterate

17:14um with the idea. Um and so uh that that process super important. And then, once you have that in place, just throw it in there, change some ECC skills, ask Claude to like recommend ECC skills to use for this, and you'll get this entire process that you just saw happen in the video thing. It's giving us a transaction ID an ID

17:35here saying that it sent a DM to Mike. So, what I'll do is I'm going to on my screen over here, I'll I'll go check my XDM's. Let's take a look. And I'll see if I sent Mike anything here. Um, if that went through. It might be loading. [snorts] So, this is Let's see it. Let's see it. Bonnie and then shout out Mike. So, I guess Mike gets a free shout out here

18:01for what he's building. went out to him. Hey Mike. Yeah. So, [laughter] this this is now so, hey Mike, you tell office office offering NYC still stands. Built a thing live on a podcast today in less than 20 minutes with ECC base of startup pitch. Five just arose in give you one task makes fundable. Video attached would love your honest take from the CTO side, especially

18:24whether the engineer judges too nice. Roast the roaster if it deserves it. Wait, they didn't include a link to it? No, it didn't because this isn't public. This is on local host. public yet. It's just showing this idea and I'll get the idea. We could have also made it public. Basically, if somebody has an idea already that they're playing with that's their own tool maybe, something that they're proud of but they never took outside of their own computer and their own idea, they can start applying ECC to it, develop the idea a little bit further, send it out to people, get some feedback, iterate and take it from

18:50there. All three, what do you get for doing this? It's incredibly popular repo for a reason because it has the end-to-end end-to-end solution for everything for building a business essentially. What do you get for giving it out? I had like this passion of writing and like teaching people about what I've learned and also talking about what I've learned over, you know, like year building with these all these new tools

19:09that have been coming out. Didn't expect it to go as as popular as it did, so I decided to release it. I'm like, "Okay, sure. Everyone can have my tool set. I'm not gatekeeping it for any any any given reason." And then people started contributing to it. And so I benefit from that, too. I think we all

19:21benefit if we can keep the open source. Um, there's almost 400 contributors now. So, everyone has contributed maybe something they've learned from whatever profession or skill set they're in. All right. We'll have a link, of course, to the repo below. And the way that we discovered Afun was that his repo just kept showing up in our top 10 list of trending repos for the week. If you haven't seen one of those shows, I've got one for you right here that you really need to see. And I'll see you in

19:43that one.