← Marketing Against The Grain
Loading summary
Transcript91 lines
- [00:03]
Kip
Hey everyone. We're coming to you with the most valuable episode that we've made this year. It is more knowledge per minute than any episode of Marketing Instagram. We're joined by Horatio Lopez. He runs revenue ops, Biz ops over at Replit. Replit's this amazing tool that helps anyone code and bring their software to life. He's going to show us the core technical terms and everything that you need to know. He's built you a custom resource to check out and he's going to walk through how you use an AI coding agent to build real working software. This episode is jam packed. Let's get to today's show. Hey guys, real quick. You know we love building custom GPTs on the show and we love sharing it with all of you. Well, we wanted to kick that up a notch. We just developed this free guide that teaches you how to build your own custom GPT on chatgpt. We've taken the guesswork out of it. We've got templates, we've got a step by step guide to design and implement custom models. So you can focus on the part that's actually fun, the part we love actually building it. And if you want it, you can grab a link in the description below and go check it out now. Now back to today's.
- [01:28]
Kieran
Welcome to the show, Horacio. We are super excited to dig into all things Replit, to dig into all things AI and coding. And so one of the things that Kip and I have said on this show is we believe marketing and go to market in general is going to become much more engineer led because of how you need to interweave AI across all of these disciplines. You and I got talking at Inbound. You were building some really cool stuff. You work in rev ops and sales ops over in Replit. And also I've kind of managed to go through some of the things you've been building, like the AI sdr. So we thought, why not come in the show, give us an example of how to put together all of these different code and disciplines and tactics to build cool AI stuff and kind of showcase it to our audience.
- [02:12]
Horatio Lopez
Yeah, great to be here. I've been longtime listener of the show and watcher of the show on YouTube. And yeah, I mean, knowing the content, this is probably going to be the most technical episode. But, but like you said, hopefully in here this will be sort of like a compressed speedrun of the technical concepts that are still worth learning in this AI first coding world, especially if you're in sales and marketing. You know it's not necessarily worth your time to go get a CS degree, but hopefully in this episode we're going to give you the 20% that gives you 80% of the leverage to build any idea that you have. Yeah, we're going to be building an AI SDR and cover all of those like technical concepts along the way. You can rewatch and rerun if any of the terms are too technical, but we'll make sure that everyone gets some value out.
- [02:59]
Kip
As a non CS person, I will play the role of clarification, questions or simplification if there's anything that folks might not understand. But I'm excited. This is going to be fun.
- [03:09]
Horatio Lopez
So, quick overview on Replit. Replit is basically an all in one coding environment. But with AI it's the simplest way, even if you don't know how to code, to create, run, write and deploy code. So we're most known recently for our Replit agent, which will do all those steps end to end. You can give it a prompt and just through natural language. It's going to use the same concepts that any developer is going to use, like coding files, the same coding language to go ahead and write all the code in a coding environment and then go ahead, compile it and deploy it for you. But if you want to, obviously you don't have to do the fully autonomous mode same way how we're building our own sales agent. You don't have to use the development agent here. In Replit, you can actually go ahead and edit files directly, which is what we're doing here. We're going to go ahead and edit the code directly, still using AI to help us, but in a little more step by step way. So Replit is sort of the bundled experience where you can go from any level of autonomy that you want to just build any piece of software that you want. And that also means you can integrate any tools. So back to like, when is it worth switching off from workflow tools to code? The way that you recreate workflows that you would do in Zapier or Make is again using APIs and then API keys. So right now we're using Slack because that's the workspace that we use for communication. But if you had teams instead of connecting Slack, you can also switch it out. Let's say if we were using teams instead, this would be Teams Token and Teams API App Token to connect our Teams bot as well.
- [04:45]
Kip
Got it. What you're basically saying is like, hey, this like virtual SDR bot could run many different places Your perspective is like you want it where the teams are working and communicating and doing work, which is basically either in something like Slack or teams or like in a CRM basically. Right. And like depending on what's easier, you're going to kind of run it in that environment.
- [05:06]
Horatio Lopez
Yeah, absolutely. I mean, the way to look at it is code gives you the leverage to connect in any instance that you want. It's just all about giving the coding AI all the context to the specific tool that you're using. Great. Well, why don't we jump in by first demoing what the end product is going to look like? I have the end product set up and running in repl.it so we have this AISDR Slack bot and the way it's triggered is you can basically call it directly in your Slack workspace. I have it in ours right now and I'm going to ask it to create a campaign targeting five marketing leaders at tech companies based in the U.S. i'm going to go ahead and hit run. And what it's doing is kicking off a series of API calls in the background to other services. So the first thing that it did is it grabbed a data provider that, you know, I don't have the most premium subscription to. So you can see there's no email addresses in this list of the five marketers. I'm going to use another service to enrich those to get those emails. It's going to use clay.com, but basically all of the steps that it's going to do here in the tools that this AI SDR is using cover the same thing. APIs, webhooks and JSON. And that's sort of going to lead into the first types of technical concepts that are still worth learning. I'm guessing most people who watch the show are familiar with workflow tools like make.com, zapier. Basically it's the same concepts that are being used in those tools that are being used here in code. Even though it looks a lot scarier because it's all written in like this weird text, it's still the same concepts where it's connecting to other services, sending a request that the computer understands, and then getting the data back in the way the computer understands.
- [06:49]
Kieran
I think that's a super interesting point in that when no code tools came out, it helped marketers and go to market folks do a lot of technical work and automate a bunch of their work. And actually AI is starting to make no code not irrelevant. But like you can actually do A lot of that automation through code. So the thing that you're creating could have been this whole huge no code workflow, but you actually have built it all from code to begin with and you are a coder. I have a coding background. I have a computer science degree, but I'm a terrible coder. But even I have followed along. You're teaching this and I could probably build the first version. So do you in replit see a bunch more beginners and people who have not coded, started to use replit to kind of do stuff like this?
- [07:34]
Horatio Lopez
Yeah, absolutely. I think the build versus buy equation is like completely changed because of AI coding. Before, it absolutely would not have been worth taking a workflow that we have set up in a common workflow tool. And for me to go and code it, because everything would be manual two and a half years ago, but now I can just basically write out the description of exactly what that workflow is doing and it'll just print it out in Python. And the savings, you know, for our scale is like 10 to 100x. Because some of those workflow tools can be pretty pricey, especially when you're working with like inbound flows or communication flows around email, SMS and like messaging. The volume just makes it so much worth it to just write the Python script and deploy it.
- [08:20]
Kieran
Very cool.
- [08:21]
Horatio Lopez
So I'll just jump over to the first API and webhook setup that it's using. You can see it's going through a enrichment workflow where basically it's calling clay and this is the webhook. I'm not going to show it because it's like a secret URL, but basically the agent has a URL that it's going to call and a path that it's going to call to use this enrichment workflow to send in the leads without the email address, and then kick off a series of steps in clay to find the email address, if it's valid, and then return it back to the agent. And then I also want to shout out like a pretty good resource. Colin Matthews has this course called Technical Concepts for Product Marketers. And a lot of the concepts translate to what you would use in sales and marketing as well, because it's going to cover the same concepts like APIs and webhooks, everything that we're going to build, basically all the tools for the AISDR all follow the same format, where we're either gonna send, get requests or post requests. And all that means is it's just URLs with a secret little endpoint or a path telling the computer where to point and where to get the data it's searching for. So for our enrichment endpoint, we gave it the clay base URL. And then there's an endpoint that's called enrich and it's gonna call that and send in the data that it wants enriched. What's going on behind the scenes for all of your workflow tools is the same exact concept. There's going to be a base URL, then the specific endpoint that you want that tells it, you know, where to find that specific function, and then the body of the code which includes all the data that you want manipulated and used within that function.
- [10:06]
Kieran
And what's that course? I see it's on Lenny's newsletter. Was that a post or a free course?
- [10:10]
Horatio Lopez
Yeah. So this is the written format. He did a free guest post on Lenny's newsletter. So if you search for Colin Matthews, Lenny's newsletter, this article will pop up. And then I believe he also has a maven course where he covers it and it's titled for product marketers. But the same concepts will apply for if you're in revenue operations and sales and marketing. You know, he didn't tell me to do the shout out. I just thought like this article alone, like covers like a huge chunk of technical concepts. That will just give you so much more leverage as a professional trying to learn to code.
- [10:42]
Kip
We'll link up the article in the YouTube comments below. So if you're watching this on YouTube, check down there and you can go check out Colin's article because I agree with you, Horatio, like this type of simplicity of kind of the basic technical concepts is really important to kind of understand what an agent's going to be helping you with, right?
- [10:59]
Horatio Lopez
Yeah, absolutely. So it finally finished the enrichment flow. So it did take a while. What we're going to do now is like have it use another set of API calls to prepare an email sequence. So it's going to use some Slack tools to just give us these little modals and then for here we're just going to give it some information. I've saved here about replit and basically it's just a little coverage and I'll show you what this gets used for in a second. But it's going to go ahead and generate emails. It's using OpenAI to create this email draft. I'm not going to worry too much about looking at it. It's a little verbose, but we're just going to go with this email copy. Right now. It's saving it. And now what it's going to do is use another set of API calls. Again, we'll show how it's doing it, but it's saving it to our HubSpot CRM. And then lastly, it's going to go ahead and actually create the campaign in our cold email outbound tool called Smart Lead, which, you know, is our tool for sending out outbound emails where we don't want to be using our primary domain. So all those steps, all those tools that the agent is using is just again, another set of API calls. But switching from. Instead of calling Clay, you're switching over to the HubSpot URL. Instead of the HubSpot URL. For this part, you're using the Smart Lead URL. In this episode, we're just going to like, build out these last two steps to sort of show those concepts of, like, how you actually go and build these tools yourself.
- [12:22]
Kieran
So, like, the agent is basically taking a list. It is enriching that list from Clay. It is pulling back that and starting to draft emails, Store those contacts in HubSpot, and then automate that sequence through your email tool. And within that flow, it's not fully autonomous because I guess the human is going to validate each step.
- [12:44]
Horatio Lopez
Yeah, I mean, I customized it and programmed it to just go step by step. But there is an autonomous step. The one thing that will never be automated or at least I don't recommend. You could automate it, the sending of the emails. I have it right now where it just drafts it. Right. And it'll just sit in there and then we do a final check on the copy and then we hit send on the campaign.
- [13:06]
Kieran
Yeah, I think that's the better way to do it. I think even if you're integrating these tools into your sales team, for the most part, there's some nuance here, but it's better to, like, craft the things and have the sales rep still have the ability to, like, edit it and then post it versus automating all of it. Like automating all of your outreach. I do like having the human in the loop to validate parts of that sequence.
- [13:29]
Horatio Lopez
Yeah, absolutely. And so, again, this is the running version that we just did. This is like the completed, you know, AI SDR right now. And you can see it's like, returning successes. Again, it looks like a wall of text, but hopefully by the end and once you read through this article, it'll make a whole much more sense about, like, what's actually coming through right here. What does it mean? When it says 200, side note, that means it's a good thing. 200 is like sort of the code for the computer understood and successfully received your request. So that's basically what that is. But what we're going to do is again build out the tools that add the emails to Smart Lead and add the tools to the HubSpot CRM as well. So let me go ahead and just turn off the one that's running. And so now our goal is that when we hit run on this version of the code like this one that we're starting from scratch, is that it'll have the same outcome that we just ran through with the end result. And if we don't, you know, we have a backup code just quickly before.
- [14:29]
Kieran
We start building the AI SDR from scratch to show us some of the core concepts you should learn if you're a non technical or non coder to start doing some of these things. Like replit has its own coding agent, right? Like it has an autonomous agent and I've done things with it. So you can actually ask it, hey, like build me this app and suggest other features that I should add to this app. And it's like really good. Like it will start to build you the app, but then it will actually start to tell you like, what are other cool features you should actually think about having within that app. So if you asked it to go build an AI sdr, would it get close to the work that you have done, like the coding that you have done or what's your experience like there?
- [15:07]
Horatio Lopez
Yeah, so what any AI coding assistant is going to do is get you to the spot that we're in right now where it's created 80, 90% of like the workflow. But the thing that it's not going to be able to do is be able to authenticate and connect to your actual workspace and tools. That's why we use the keys. It's not yet ready to fetch these on your own. We do have some in the replit agent specifically where if it understands your tool is going to use OpenAI, for example, it will ask you within the prompt it'll have a little pop up that says, hey, can you give me your API key so that I can actually connect to your OpenAI account. But in the near future we'll be able to have that for many, many other tools. And we're working on a few that will be out soon. But for now the safeguard is you sort of have to keep in mind that you are going to have to Go out and get those API keys. You can see the list here of all the tools that we're using. You're going to have to have those at the ready or know how to get the instructions to properly connect to those external services that you own.
- [16:12]
Kieran
Yeah.
- [16:12]
Horatio Lopez
So authentication and then last thing is API accuracy is something that AI assistants won't always get right. So what I'm going to do in a second here to create these functions so that I make sure that I'm making the right formatted API calls, is actually use Perplexity to index and see what's the most up to date documentation for this particular API, like Smart Lead or HubSpot. Like how do I properly call the API endpoints? Like how do I properly call? Because these paths and the format that you need to send a request to can change and even the smallest changes can cause your tool to completely fail if you don't have it typed out. Exactly correct.
- [16:54]
Kieran
Yeah, Just coming back to the other screen where you actually do have a bunch of code ready to use. So you basically have a bunch of code. You said like 80, 90%. Did you write that code or did that come from the replit agent? I think we missed that step. Where did this come from? Did you use the replit agent to create this first draft?
- [17:10]
Horatio Lopez
Yeah. So the replit agent, because I built this tool like a couple months ago, I did not use the agent because it was always going to try to build a web app, but now it would be able to build this. What I did was actually use this other AI assistant to the side, which is the Replit AI assistant. It's sort of like ChatGPT, but it's integrated to your code base so it has full knowledge of all of your code. And for me, like knowing a little bit, you know, the technical concepts that we covered, like the API calls, it's a more controlled version. So like I can give it specific context on the tool that I want to use and then have it create like the features that I want and take a little more granular and controlled approach than the agent would give me. But nowadays, yeah, I would probably start with the agent and then after a couple of steps switch off to what we're about to build here.
- [17:57]
Kieran
Okay.
- [17:58]
Horatio Lopez
Or like this mode using the AI assistant rather than the agent.
- [18:01]
Kieran
Okay. So the agent is fully autonomous, can kind of build you a first version, then you can switch in and start to use the coding assistant to actually help you finish the last parts.
- [18:10]
Horatio Lopez
Exactly. It's sort of like the same concept for any AI Tools nowadays agents are sort of like your senior engineers that can take very general instructions and then be able to break it down and build that app end to end. But it's going to be more expensive. The assistant is more of like an intern where you have to be more specific about what exact requests you're looking to do. But it's much faster and it's much less expensive in terms of like token usage. Okay. In the background. Yeah, cool. But yeah, so like I said, the first thing that we're going to do is build the Smart Lead campaign tool. So that basically when we get to this step on preparing the email sequence, once it writes the emails, it's going to go ahead and call the Smart Lead API to go ahead and load the emails and the leads into our Smart Lead account. So the way that we do that is I always go again to Perplexity, say using a Smart Lead API, how do I create a campaign? And I use Perplexity Pro and I do recommend it if you're doing coding tasks regularly, especially like these types of queries, because it sort of takes two steps. It's like going to go to find, you know, the proper request and it actually pulled out straight from the documentation. You know, this is the proper format for the API. And then it's necessary to do multi step where it grabs a source and then formats the source into the proper coding format that you need. And Perplexity Pro lets you have like that sort of step by step logic built in.
- [19:41]
Kieran
Do you use Google at all anymore for search or do you find yourself.
- [19:44]
Horatio Lopez
No, actually I do not. Very rarely, I think I only use Google for like local. Yeah, you know, local search. Yeah, local like landmark search and stuff.
- [19:52]
Kip
What's interesting about this, right, is we often talk about AI coding, Claude and OpenAI and Replit, but what you're actually saying is like, oh, if I am building like some custom code, whether I'm using any of those tools, in this case, Replit, right. I need to understand how the actual tools I'm trying to integrate work because I'm trying to basically build a custom integration. And for something like that, AI search is actually the best opportunity because you don't have to then go and find all the documentation, put them in Claude, ask Claude to find it. It's like Perplexity is doing all of that for you and saving you all that time. And so like that's actually a pretty interesting part of the workflow that Perplexity fits in that like, I think Kieran, we don't really spend much time talking about. Honestly?
- [20:40]
Kieran
Yeah, yeah.
- [20:40]
Horatio Lopez
I mean, that's a great point. The alternative, you know, before perplexity. But I would have to essentially take this whole page here and copy and paste it into my AI assistant and then ask it to just focus on the create the campaign part to create that function that I want for my agent to have. But yeah, essentially this cuts out like a couple of copying and pasting steps and gets you right into what you're looking for. Exactly.
- [21:06]
Kieran
I do wonder if you went and ran the exact same search in OpenAI's new ChatGPT search. Does it look pretty similar? I think that's the challenge for Perplexity. I think perplexity is awesome. I just. I just don't know how differentiated they'll be when OpenAI continue to build out that. But anyway, that's a different point.
- [21:23]
Horatio Lopez
Yeah, we'll see. Time will tell. I haven't tried out the new GPT search. I do think focus is a superpower though. So we'll see.
- [21:30]
Kieran
Yeah, I agree.
- [21:38]
Kip
We'll be right back to the show, but first I want to tell you about another great podcast called Entrepreneurs on Fire. Hosted by John Lee Dumas. Available now on the HubSpot Podcast Network, entrepreneurs on Fire is fast paced and packed with valuable stories that'll help you be a more successful entrepreneur. And recently they did a great episode with ex Hollywood actor Ben Kurland, who is a blockchain expert specializing in Web3 adoption. It's fascinating stuff, so listen to Entrepreneurs on Fire wherever you get your podcasts. Hey, it's Kip. If you listen to this podcast, you know how much I love keeping up with the latest and greatest in technology, but very few podcasts actually give you a dose of the future. The A16Z podcast is the exception. It's hosted by our friend and frequent guest Steph Smith. The chart topping show brings on movers with a track record of being both early and right. Like Apple co founder Steve Wozniak, or the CISOs of OpenAI, Anthropic and DeepMind. Even the very first CTO of the CIA. From the science and the supply of GLP1s to drone delivery to the economics of deepfakes. Go ahead and eavesdrop on the future. Check out the a16z podcast wherever you get your podcasts.
- [23:00]
Horatio Lopez
So now that I've gotten this initial code example, I can give our other AI assistant proper guidance. I don't have to call it explicitly because again, it has full code based knowledge and it should know what I'm talking About. But I'm just going to be a little bit more explicit. I'm going to say finish the Create campaign Smart Lead function using the Smart Lead API. And then I'm going to paste in what Perplexity gave us so that it has concepts of how to properly call the Smart Lead API. And then really quick, this is again back to the core concept. Like, the same pattern plays out here as well, where we have the base URL and the endpoint that we're trying to use. What Perplexity gave us is the same thing. This is the base URL, server, SmartLead, AI and API v1. And then this is the endpoint. So the same concepts that Colin outlines here play out right here in the Smart Lead API. It's the same thing, base URL, the path that we want, which is Campaigns Create. And then we need this extra format for, you know how Smart Lead requires you is basically appending the API key and then within it we have a post request with like the name of the campaign that we're trying to create.
- [24:12]
Kieran
So you basically went and told the agent to write a function and you could provide it with the API documentation or the relevant part of the API documentation to go write that function.
- [24:24]
Horatio Lopez
Exactly.
- [24:24]
Kieran
It's pretty cool.
- [24:25]
Horatio Lopez
So, and then the last thing I want to say is like, name it whatever you want. We have no client id.
- [24:33]
Kieran
Like, when you think about the work that you're doing here to do this AI SDR and how you're using these different AI tools, do you think the kind of level of code knowledge you need, you can do this stuff with a lot less knowledge of code, or you need the same knowledge of code. You can just do way more. Right. You know what I mean? Can someone who has relatively basic understanding of code, because you still have to understand a bunch of stuff here. You have to understand API documentation, you have to understand how to write functions, but you're just able to direct an AI assistant that can be a coworker to do those things. So what do you think and what have you seen for people who are maybe more in the beginner camp of.
- [25:13]
Horatio Lopez
Coded, I would say more of like the latter point that you made. I think the basics give you infinite more leverage. Amjad, our CEO, was on a recent, I think, interview with Y Combinator and he said, I fully agree with this point where it's the value of learning to code doubles every six months in this AI first world.
- [25:33]
Kieran
Yeah.
- [25:34]
Horatio Lopez
So even the basics give you way more leverage than they did three years ago. Like, if all I knew three years ago was API structure and webhooks. Then, like, I would only be able to use workflow tools really proficiently. But now if I learn those same concepts, If I learn APIs or webhooks and like the data structure stuff that we just covered, to be able to go through this, like, now I can build full web apps with a little bit more direction. That's what I've been seeing too. Like, some z zapier experts have started to come to repla and be able to build full AI web apps using code because they've had the foundational structure that doesn't give them the level of a CS graduate, but it was enough to give them all the superpowers they need to basically build 90% of ideas they have.
- [26:16]
Kieran
Yeah, I can imagine Revops teams being very AI centric, building a lot of these tools for teams.
- [26:22]
Horatio Lopez
Exactly.
- [26:22]
Kieran
You probably see this in replit, right?
- [26:24]
Horatio Lopez
Yeah, yeah. Because in sales and revenue marketing, we have so many SaaS tools out there. And I think because most of them, like have APIs, that's where we're seeing like a lot of like this type of workflow play out, where it's like people learn the basics of Python and now they're able to run and make something that really works for their team. I'm bearish on integrations teams for the future.
- [26:46]
Kieran
Are you bearish on no code for the future? No code platforms? You don't have to say it if you don't want to, but like, say it.
- [26:54]
Horatio Lopez
I think I will say, from what we are seeing, there is definitely some cannibalization there. The new AI coding is now prompting an AI assistant to generate code for you. And so, yeah, it is starting to play out. But the thing that will always be the bottleneck is like the ability to authenticate and connect to your specific services. But learning these concepts that we have in the article and in other places will give you all the leverage that you need. So I actually have this website, this very simple web app that basically has all the concepts. It's just like a static webpage that people will be able to go to and people can come and visit this. And this is basically the list of everything you need to know in terms of more technical concepts. As a sales and marketing professional.
- [27:40]
Kieran
Yeah, we can put this in the show notes and people can go through this. These are the concepts. And maybe what we would want to continue to focus on here is if you go through your AI SCR and just go through those different parts. So like the part where you get the list, the part where you ingest the data and enrich the data. I think all of those parts, like, if you kind of learn how to do those concepts, everyone can kind of do them at the same level because they're using the same sort of tools.
- [28:01]
Horatio Lopez
Yeah.
- [28:01]
Kieran
Just from you building some of these tools, when you get into the how do I craft the email and then how do I get the people to actually engage with that email? Is there anything that you think about that you could do different in there in terms of, like, the way you use the data or the way, like, what step in there do you think that there's some leverage in if you can actually approach it creatively when you're building these AI agents For the SDR part? Because the AI SDR is like that support is probably like one of the more common use cases that people are standing up and this is a really good way for anyone to stand up. A little use case that you can put in Slack and start to use with your sales team, even if you're a relatively small company. But then if it becomes ubiquitous to how everyone does it, then everyone's kind of doing the same stuff. And I'm just wondering, like, do you see any way that you can do some stuff in coding or creatively approach one of those steps?
- [28:47]
Horatio Lopez
Yeah, I mean, in terms of email copy, you can get pretty explicit. I'm actually going to use AI so that other people who come to the code base can see where to find it. It's like, where can I find the prompt for that writes my email copy? I mean, the number one thing is you probably want to be more explicit and templatized on the email copy. You've probably heard this many times and just teams have learned this from experience that having AI write your full email campaign is not the smartest. What's smarter is being able to take and ingest specific data points and then bringing those into your prompt, which should be pretty direct and to the point of, like, how the AI should use it. Like, you don't want the AI have all the creativity. You just want the AI to, like, flesh out the very specific details. And that's what I have here in this prompt is I give it, like a very specific guidance on what variables it's able to use and what it has to, you know, include in the email copy. Most of this prompt focuses on formatting it properly for smart lead so that it appears properly, like bolding and signature and brackets for spintax. I think that as many data points as you can ingest into your API call, then that's going to give you the most leverage and like the best outcome for your email copy.
- [30:05]
Kieran
Can you just go through the prompt and how you tell it to use the data that it's grabbed from Clay? Because what you've done here is again, just to recap for people following along. You've gone and asked for a list, you've enriched it via Clay, and then you are crafting these emails to those people within that list and Clay has enriched it with a bunch of data. So how have you asked within the prompt for it to use that data in the outreach?
- [30:29]
Horatio Lopez
Yeah, I didn't use too much data just for the initial AI SDR because I knew people would have different requirements. The only thing that I used was the email address. What I told it specifically is to generate a single email. So it does it one at a time. And then I gave it like very specific instructions on the structure of how it should output its response. I do have other AI workflows and prompts and I'd say the number one pieces of data that I include all come from LinkedIn, which Clay has like a very good LinkedIn data source for grabbing things like last posts, you know, last place of work, current place of work. And honestly, those main things are the only things that I ingest to like create any personalization in the email copy. When we do use it for outbound.
- [31:19]
Kieran
Do you do anything with the build on the data? So if you have like last place of work, do you then get any kind of context about that last place of work from anywhere? Or if you get last job, you kind of overlay on context about that last job, you know what I mean? Like you take the data from LinkedIn and then you enrich that data as well. Or you kind of just use the data that LinkedIn gives you to kind of craft those emails.
- [31:42]
Horatio Lopez
The main thing that I used last place of work to do is to verify that the email provided, because sometimes we get emails from different places, but I verify that like the email domain matches with like the last place of work or any of their last places of work. That's the main thing honestly is like the job title, does it match like our icp? And then second is if we got the email from outside of Clay, does it match with like their current workplace domain?
- [32:08]
Kieran
Right. This one here is kind of the core business development rep get list, enrich list, send that email. I can imagine, like if anyone is building cool go to market agents and AI flows replit art. Because you know, you have the tool that is building a lot of these. What are other things that you've found useful? Because I think the folks that I talk to that are using AI wouldn't go to market. They're still the most obvious things that people are using it for, like customer support, prospect and outreach. Have you seen anything different? Like, have you found any other uses within the go to market, within replit that you're finding AI is really good for?
- [32:46]
Horatio Lopez
I would say we're mainly using it to leverage our inbound. We haven't tested it too thoroughly, but one project that I'm currently working on is we're a relatively small team and we're trying to handle and manage the inbound deal flow. And some deals, you know, with inbound, like, you have to quickly decipher, like, what deals are worth taking on fully yourself and what deals, you know, could essentially be automated by an agent. We talked to an AI SDR company, like an official one, not this tool. And they only have AI SDRs for outbound, which is what this tool does as well. But what I'm currently trying to build is a way for us to service inbound requests via email. That's still TBD to see how it works. But again, it would probably be this asynchronous flow where it might generate a draft. Let's say a request comes in, it gets all the enrichment data from clay to learn where that person came from, like, what's their title, et cetera. And then that will feed into the assistant that will draft an email and then also, like, get context into our products, what we're offering, and then just draft like the first one or two lines. And I mean, that's actually like a huge help. Other than that, like, haven't seen too much.
- [34:01]
Kieran
You mentioned something really important there, which is so, like, you can't come in and you want to understand whether it should go to, I assume, the sales rep or be handled by an agent. Is that what you meant by, like, should it be handled by an agent? Is potential ARR. And if it's a big deal, go to a rep. If it's a smaller deal, be managed by an agent.
- [34:18]
Horatio Lopez
Yeah.
- [34:19]
Kieran
How do you do the routine?
- [34:20]
Horatio Lopez
The number one thing that weighs heavily on that is the domain. So it checks the domain gets enriched on Clay. If it's a company of a certain size.
- [34:28]
Kieran
Right.
- [34:28]
Horatio Lopez
Then it automatically, you know, the email that they get is going to include like a meeting link to book with us. Yeah. And then after that, the next thing that gets weighted in is the actual message copy that they sent in. So in their submission, like read it, does this look like a support request or does it that route it to support if it's a support request or does it include like any mentions of a specific thing like pricing questions or questions around functionality and scalability of a certain product that we have? Those are the two topics that I think are kind of the deciding factor in the actual analysis that the AI does, which is what you know, LLMs are really good at is looking at content and being able to categorize and summarize it somewhat deterministically.
- [35:11]
Kieran
Yeah, we have done a bunch of this. And so the thing I'll tell you you'll probably figure out is very similar is like how do you route the demand to the right go to market tactic? But the thing you'll find is the larger the company, the less engagement you'll get with email only. Like you'll find that you have to kind of run these multichannel campaigns where you have to do email, voice, other things because email has become much more competitive because of AI itself. And I think that's where it gets pretty interesting to look at things like voice enabled agents and things like that as part of like how you actually reach out and communicate with those prospects. You're still kind of doing outreach, right? You're doing outreach into your inbound demand to book a meeting, I assume with sales reps. Yeah.
- [35:46]
Horatio Lopez
I mean right now the drop off hasn't been significant.
- [35:50]
Kieran
Interesting.
- [35:51]
Horatio Lopez
Yeah. Right away, you know, they come in, they have a question about the agent or product, a question about scalability on replit, which by the way, it can scale up at whatever needs you need. But yeah, I mean usually the person will come through and book a meeting.
- [36:05]
Kieran
Cool, Very cool.
- [36:06]
Horatio Lopez
We haven't had too much drop off from submission to meeting booked like in the steps of our funnel. That's the least drop off that we have.
- [36:13]
Kieran
Very cool. Last question, Horizio. Other than replit, what's an AI tool you think go to market teams need to use? And maybe other than clay, because we've had clay on a bunch of times.
- [36:24]
Horatio Lopez
I would say still a lot of people that I come across are still sending emails from Apollo or even their main domain. So using instantly and Smart Lead, Smart Lead, you know, the one that we're using in this demo has a better API in my opinion at least like when we first signed up. And that's just to get like, you know, the basic primitive of like your outbound motion, like sending emails is like something Everyone has to do. I think they have a pretty good tool for warming up domains, setting up domains that are not your primary domain so that you don't get burned and like land your whole company in the spam inbox forever.
- [36:57]
Kip
Yeah. What I think you're kind of pointing out for folks who may not be experts at email deliverability, sending email to opt in high permission contacts is very different than cold email. And you can send a great amount of marketing email and even non automated sales email. But if you're going to do high scale outbound to very cold contacts, you have to basically have a dedicated email IP and service to do that.
- [37:27]
Horatio Lopez
Exactly the last thing. I think we can speed run the completion of the AISDR in the same way everyone else will probably be building it anyways. You can come to replit.com guides and then AISDR Slackbot with a hyphen in between each of those last words and then there will be a link in there to use the template, which is basically the completed version. So if I click Use Template I'm going to bring this into my profile. The only thing that you'll need to do is set up your own API keys so you'll have to put in like your own Slack workspace keys, your own Apollo keys and your own OpenAI et cetera to set it up. And I've seen a couple people just start with this and then they get set up. In just a couple of minutes we're going to come back. Now this is the instance that's running in our Slack workspace. We're going to call our AI SDR Slack Bot.
- [38:19]
Kieran
There you go.
- [38:20]
Horatio Lopez
Create a campaign that connects, that is focused on three marketing leaders based in the US and then it's just going to go through the same process. So this is the AISDR and you can use templates and code. Yeah, basically use the same concepts. I'm also going to go ahead and deploy that web app where it will include a landing page for all the concepts that you need.
- [38:45]
Kieran
Very cool.
- [38:46]
Horatio Lopez
Again, like what's the 20% of technical knowledge that you need to gain to get 80% of all the leverage that you need to be able to take any workflow and turn it into a web app, especially in sales and marketing. So if you just take these concepts and have Claude or chatgpt walk you through, you'll be pretty much an expert.
- [39:02]
Kieran
Awesome.
- [39:04]
Kip
Yeah. So let's not gloss over this. Horatio has created a special webpage web app just for the show today where you can basically get the overview of all the basic skills you need to learn to be able to build tools like this and to work with a product like Replit and what you're saying, Horatio, if you just take this and then use ChatGPT, Claude, whatever your AI tool of choice is and kind of go through it and have that tool explain it, you will have kind of the core knowledge and be that kind of engineering led marketer, sales leader, entrepreneur that we've kind of talked about on the show today.
- [39:40]
Horatio Lopez
Exactly. Yeah. So if you just sit down for a Saturday afternoon, copy and paste all these concepts, have Claude walk you through, you'll come out an expert in a couple of weekends.
- [39:50]
Kieran
Very cool.
- [39:51]
Kip
That's amazing.
- [39:52]
Kieran
Thanks so much for coming on the show. This was a great introduction for anyone who wants to start to learn some basic code and to get a ton of upside on AI. I know people are going to appreciate it.
- [40:02]
Horatio Lopez
Yeah, absolutely. Thank you for having me.
- [40:04]
Kip
Thanks for being here. Thanks, everybody. We'll see everybody real soon on market against the grain.