AI-driven Content w/ Eoin Clancy from Airops- GTM Engineer School (Preview)

All right, fantastic. Well, hi everybody, thanks for joining. From all over the world, I'm Jared Waxman tuning in from the heart of Silicon Valley. Over the, over the years I've worn a few hats and, and that's been across product growth and demand gen and all that stuff comes into play here with go to market engineering. Our agenda for today is going to be pretty simple. I'm going to give. Matteo and I are actually going to give a little bit of overview of the go to market engineer school. But then we'll get into the meat of this which is going to be in talking about AI driven content workflows and previewing a little bit what the hands on exercises are going to be within the course and then we're going to open it up for Q and A. So yeah, well, let's say a little bit about ourselves. Ian, do you want to give a little intro? Yeah, sure.

I'm Owen, head of growth at Air Ops. Worked with Andy, who's top of your screen as well at Telnix for the last six years before Air Ops started as what was probably the buzzy term back a decade ago as a growth engineer at Telnics, did that for two or three years, went on to become like a PM and then ended up running the marketing. So that's how I came into contact with airops where we revamped our content ops, did a lot of programmatic that Andy led the charge on, really grew our organic growth and then jumped onto the airops ship because it was so good in the last few months where yeah, excited to show you a little bit behind the scenes today in terms of workflows and then more generally we'll sketch out whiteboard style how each of the tools see on one of your screens later how each of these tools can connect. So I'm a very much a whiteboard guy. So excited to dive into that with everyone.

Cool, thanks Alan. Matteo.

Hey everyone, Matteo here. Yeah, also very similar background to Owen. I mean I've been in startups in B2B for just under a decade and mostly as a founding marketer from pre C to Series B. And since a couple of years I now run my consulting studio advising startups like Ahrefs, Common Room, Hypergrowth Partners and also work with Owen at Air Ops most recently.

Fantastic. If you're not speaking, you can mute yourself. Thanks everyone. Yeah, and then myself I was leading growth at Auth0 and then running campaigns at Okta. And one thing for me is that at every role one thing's always state convoy Constant, which is a passion for continuous learning for myself and the teams I've led. And I found for me the most valuable type of learning happens in structured live settings, the kind where you can ask questions, share challenges and learn directly from folks in the field. So for me it's been programs like Reforge and Pavilion and Winning by Design. They've had a huge impact on my thinking. So those are some of the experiences that have influenced how we're building. Go to Market Engineer School. So yeah, you can see hands on learning, access to VETA tools. Some of the pioneers and masters in the space that are going to be doing the instruction here, there's going to be multiple use cases, you're going to learn. A lot of the people that I talk to, especially in Revox, marketing and sales roles are saying the same thing and that is that AI is moving so fast it is hard to keep up. And I think they're right. I mean the pace of change in Go To Market tools is absolutely wild. But trying to learn it all solo, I think that for some people that's a recipe for burnout. So we're trying to build something a little different. At Go to Market Engineer School. It's hands on, it's project based learning built to get you somewhat comfortable if not fluent in AI powered workflow tools that are reshaping modern go to market teams. So we're bringing together 30 top voices, these are founders, these are operators, educators, help you skip the fluff and get you straight to what works. So you're going to be completing 16 practical mini projects using some of the most powerful tools in the space. Clay, Lemless, Cargo, Octave and Air Ops. And this isn't just passive learning. You're going to be building the automation, shipping, content, refining Outbound and then getting feedback from the instructors along the way. Designed to be very beginner friendly, no technical background required, perfect for folks and rev ops, sales, content growth, PMMs roles. Just going to share quickly what's included and then, and then, and then we're going to go through and Mateo will go through some of the modules real quick. So what's included? You're going to get up to 28 hours of live small group instruction across four weeks. Replays available if you miss a session or two. And then I think it's now up to over 7,000 in free platform access including three months of Octave which is worth $450. Two months of Clay Pro which is worth $1600. Two months of Cargo Pro which is worth 3,000 five months of Lemless Multi Channel Expert which is worth $500 and then two months of Air Ops Pro plus with 20,000 credits worth $2,000. So you'll have a lot of time and all the features you need to test out, try and build skills and those capabilities and then decide what you want to do next. So with that I'm going to hand it over and Matteo, you want to go through some of the modules?

Yep.

So thanks Jared. As Jared just mentioned, we're going to have basically four weeks. The first one will kick off with using Octet to really codify your messaging and positioning. And once we've done that, we're going to like explore how to use that as the foundation for creating qualified account and contact prospect lists with Clay. Then we're gonna like move on to how like we're gonna use like the inputs or the outputs of week one on with and so basically our prospect lists to create personalized omnichannels campaigns and pushing them to Lemlist and set up all that outbound engine there. Whereas in week three we're going to build some AI agents in cargo to run basic GTN use cases like custom custom account research or again custom qualifications or engaging your best deals, lookalikes or qualifying leads from product usage. So again, and last but not least, week four we're going to do all things AI led content creation, deep diving into airops and I guess Owen will be a much better person here to give you an overview here. And that's what we're going to do in the rest of the webinar. And yeah, last but not least, along the journey we're going to be joined by these top operators that are expert in all of these tools and in these use cases. So when you are going to be in class building and you get stuck, which is part intrinsic, part of the learning process, you can always raise a hand and they'll be there on your site to get you back on track.

Fantastic. And then so obviously today we're going to be focused on week four. I put in the chat and you see down here we're recording this webinar. It'll be available at this YouTube channel later in the day. And then we have another webinar coming up next week. You can sign up for it at this link and with that I'm going to stop my share and Owen, you want to take control here?

Let's do it.

Give me one sec.

And then let's try and keep this collaborative as well. So like, as folks have questions Drop them in the chat and I want to be able to answer them. So maybe Mateo or Jared, as you see them pop up, that you can help facilitate. All right, let's do it. So what I want to go through here pretty quick so we can get into the fun stuff. It's just like a preview of as Matteo went through there Week four of the Go to Market Engineering school, which is where Air Ops gets involved. We have amazing folks from our team internally who are our solutions architects and then a few hand selected customers like Andy, who you'll see on the call as well. That will be your, your TAs and showing you the meat and potatoes, how this stuff comes to light. And in my head it's most important that for each of the classes you do with us, you have like a workflow that works, can be put into production and then deliver value. So I know that's true across all of the cohort leaders, but I think that that's like the real value that you'll get out of the course. So kind of why care? Why? Why are we here at all? What matters? So the kind of industry we go after is we're talking to content marketers and SEO professionals. But when you look at marketing or growth more generally, I really like this tweet I saw during the week from Ross Simmons, who's really well known in the world of SEO. But mediocre marketers will be replaced by AI. Good marketers will use AI to do good work, but faster. And then great marketers will use AI to do great work. F so like in my head, AI doesn't replace marketing. It's almost raising the floor. If you were a really bad copywriter, you can now jump into an LLM like Claude or anything from GPT and you become what I would say an average copywriter. So the people who are still in the top 10% or 1% can still stand out by being in that top tier. And that comes with honestly, just practice. So all of these tools are now available. And if you might have seen like GPT4O's release last week, like if you're not a designer, you now are. So you need to be able to like use these tools to go faster. We also need to know how to use them so that you can put them to the best use, which I think across this cohort you'll get real hands on experience with. But I think this is like an important principle where we come into play and what we strongly believe is that winning the battle for attention is harder than ever. So Whether you're creating SEO content or you are writing posts for LinkedIn or you're generating like outbound email campaigns, we all see so much content and like some of it is honestly just like AI slop. Like we said, it's very easy to now generate the bare minimum or an LLM is the average of everything that's out there. So it's just going to be average content. So really standing out, be it top of an SEO rank, be it having something that's like useful to say on LinkedIn or having like an article that pushes the conversation forward, it's getting more and more difficult. So through this course, want to be able to make you with tools, be able to like stitch these things together quicker so that you can actually get your higher quality content or message or brand out in front of the audience. So just giving you a flavor of some of our customers and how they do it on like the content side. Airops is mainly used for one of three big use cases. It's used for strategy, production and optimization of content. Some of these logos here you might recognize and we'll deep dive into deep gram in just a second. But for strategy, there's obviously jobs to be done on any growth or marketing or sales team that a human should be involved in. There's other ones where like LLMs are going to be your best friend and they'll do better work faster for you. So Lightspeed, for example, they used to spend five to 10 hours creating like a content brief, be it for like emails or editorial articles. They now automate a lot of that research. With Air Ops, you have a company like Deepground, which again we'll go into. But they had like some really good ideas of like how to put their internal data to use and get in front of their audience. So they ran like a programmatic play. So they created content. And then last but not least, Descript, they have a ton of articles and they want to get more value. They have amazing subject matter experts. They want to take those transcripts, push them onto content and then again get them in front of their audience. So they went down like a content refresh route. But what they're starting to see now is improved clicks from Google. Google, they see that the content's actually resonating more with people. So people are actually subscribing to their tool at a quicker rate from what they send. So this is just like a little taste of some of the outcomes. Hopefully you can realize out the far side of this school. I'll kind of glance over this Again, different use cases here we'll start to touch on. But if you just wanted to learn a bit more about like strategy production optimization, here's generally for our customers, how we break it down. But we're trying to align the four classes that we do within the school to give you a flavor of each of these just so you can then take what we give you and then go on to improve it and like curate it for yourself or your company. So giving you one of our best examples is Deepgram. Like I said, they had a ton of internal data and the way that they bring it to light was programmatic pages for you. Maybe it is an outbound email, maybe it's like LinkedIn posts from your CEO. But they had all of this information on like amazing startups that were using their platform for different use cases and they had a load of other contacts that they were able to scrape from like Y Combinator or other like startup hubs and they wanted to grab some of that traffic while it was available. So that was their programmatic play. And then similarly they run a podcast. They have like YouTube videos. And when you have an asset like that, you don't just want to leave it on YouTube or leave it on whatever the, the podcast store. You want to make sure that you're able to like get more and more and more value out of it. So they run a similar play to descript, which was taking those core pieces of collateral and turning it into more assets that they could push onto LinkedIn into their outbound messages and further on. So again, just a flavor and we'll start to think a bit more about workflows here and actually getting into production. There's kind of like four or five different flavors of how to stitch together a really good workflow. If you think about the, the dark, I guess dark purple color on this, this is like where the human input comes in into light. Number one is up the top. Who has the idea. It's probably not going to be an LLM like you as an expert and you as a go to market engineer. You're going to have a lot of stuff that like lives in your brain that you want to bring to light. So in their case it was a target keyword. You, it could be many other ide out the far side before anything gets sent. Like you still will want to have like a human review or what we call it an expert in the loop. And then again the last thing for their content to get out is published to a CMS or like a content management system for you that might be again smartly lemless clay. That's where like data comes from. Then if you look at this kind of purple color as well, you want to be able to like scrape information that's out there. So they looked at top ranking articles within Google, they did competitive analysis. So think strategy. And then they also were able to use some AI features. So now there's AI overviews in Google. There's also people also ask which gives you ideas as to what your audience is interested in. And then the red down here is colored as such because it's their internal data. They had stuff in like redshift, postgres, snowflake, they wanted to like take abstract and get into their content. Green here just shows you what I'll get through in a minute which is like LLM prompting. I think if you were to leave with one thing from the air ops cohort or air offset of classes within this cohort it would be like prompting techniques and that's where I want to spend a little bit of time today. But they're generating all of these like individual components for the page one at a time. And that's important. They're not just saying generate all of the content for me they're being very, very precise. And then last but not least, they're choosing internal links. So what's important in content is not just creating a piece of content that doesn't link to anything. It's going to give you no value. It's going to not be valuable for your reader either. So there's a lot of elements here that get stitched together and it's an end to end workflow. That's the most important thing to call out. And so you will see there that is difficult. It will be hard for any of us to just have that idea and put it down on paper. And when we talk to customers we talk about content engineering because what you've seen there is like pretty content related. So generally customers of ours, they'll work with our team, we'll help them build and codify their workflow and then we've like a pretty vibrant community as well where people help each other. But within the GTM engineering school I think these are some folks that I would like each of you to try and aspire to be. I noticed Mateo's on the list but you do need to learn within like a structured environment. It's great seeing people post like one off workflows on LinkedIn and unless really you can like see behind the scenes and like know where they Started how they developed it and like where it is today. It's very hard to replicate that. Like Joe Rue, an amazing user of like many of the tools you'll see in the cohort and he does great diagrams that he put puts with his content on LinkedIn. So he's one I would definitely recommend to follow where he starts to show you the entire process of like where he started, how he mapped it out, how he got a V1, how it got an output that he needed to like curate. So he's really good on a deep research workflow, which I'd recommend checking out. Sylvan is just like an expert. He's like top 1% of like GTM engineers. He does great content creation and one of our workflows and I think one of lemlist workflows pulls from his core concepts. And then what I'll touch on in a few minutes is more like social content. So going to be one of our classes, you'll learn how to create it. But I actually did a webinar with Matteo months ago and he's amazing at knowing the building blocks, like knowing where to pull really good hooks, really good content from how to stitch it all together. So generally what I want to convey here is that going through this class, you're going to come out of each individual lesson with workflows you can actually use. But more importantly, you're going to see behind the scenes of how some of these amazing TAs and other like co cohort leaders start to think. And we'll hash that out later in the whiteboard session. But like the principles, how to understand where to start, it's as important to map it out as to know how to build it and then once you've built how to optimize it. So again I'm just showing like the finished product on these ones here. But what you'll really get which is important is like the how to do it. Oh, what will I learn? That's a really good question. So from the Air Ops set of lessons, we're going to dive into the kind of like different buckets I mentioned earlier. We're going to look at content creation. So we'll look at how to create SEO optimized content. You'll bring some ideas. We'll help you figure out then how to rank within actually the SERP or like your search engine ranking, think top spots on Google or Bing. But we'll also start to show you how to rank and get cited within like GPT, Perplexity and others. There's A bit of a skill to it, but it's actually not that difficult. So content creation is really, really important. The workflow I'll show today is how we can, as I call it, influence others, how we generate social posts. There's many different individual components and again I'll start to show you how to stitch them together. Class three is how to optimize content. And again we'll go a little bit deeper on the creation side and show you how to not just get cited, but like get primary citations in AI overviews and other locations. And then finally, deep research. Taking inspiration from Joe Rue's example where he actually profiled me, which is a little bit scary, how you can take an individual Persona or company, use elements like perplexity, LinkedIn, content and more to do a ton of research on a prospect or a company that you can then use within these other tools like Clay or Lemlist or Octave to learn more about your audience or just deliver the perfect piece of content to them. So those are the actual lessons. What underlines all of those are the key concepts and like principles. I want to make sure that each of you comes away with prompting. Everyone can prompt, but not everyone can prompt very well. Like everyone can go to GPT and just be like, write a blog post for me on X. It's not going to be great, unfortunately. So what we're going to show you a little bit of today is how to master system prompts, how to start to use user assistant pairs, and almost how to know what model is best for the different tasks. Orchestration. We'll get into the whiteboard session. I want to show you like how to actually map something out, how to start with an idea, how to make it like actually come into light and start to stitch things together. The copilot assist. I think a lot of the tools you'll see in the cohort as well have like co pilots that will help you build cargo does in particular. But using Those are like 10x tools and you need to learn how to use them and how to get the most value out of them. Because when we watch our users, unfortunately they don't, they don't get the best value out of it. So I think that's like one key thing I want each of you to learn as well, like how to get these amazing tools to do a lot of the work for you, which is a great skill set. And then finally integrations, be it integrating air ops into other tools or trying to bring data in from other locations, be it via like API or direct Connects, that's like a very important concept that's going to take you very, very far too. So I might pause there, Jared or Mateo, if any questions came up.

Not. Maybe not, but does anybody have any questions they want to interject?

We dropped the link, but this is.

One that I just got turned on to is the I posted in the chat. But using Copilot, there's a lot of.

Steps you can accomplish.

Workflows where you just ask Copilot to build you a Python script to do something and it's going to work way faster than an LLM step, it's going to save you tasks, it's going to keep the workflow nice and lean. But if you haven't tried Copilot with Air Ops, it's pretty incredible.

Yeah, it is, actually. That's a good point, Andy. I'll try and show it as I go through the demo as well. But all these tools I worked with, Andy, I know his level of prompting, it's super high. It's the level of Mateo's doc that he shared out. But all of these tools, they're. They're only great if you can use them at like an amazing level too. So with that said demo time, so let's take a look under the hood. I think we're doing pretty good on time. How do I get into this? So what I mentioned in, I think it's lesson three that we'll cover is social post generating. So I want to give a shout out to, oh, I must have closed down the link if I can get rid of this. But as much as us coming up with ideas, or me in particular, I often get inspired by users. So another person in addition to like Joe Matteo and Sylvan that I recommend following is Mode. I personally had a good social post generator. He took it like 10 levels further. And then what I'm going to give you is hopefully another few levels above it. But I like again how he maps it out. But let's run through one or two examples. So this here is in Air Ops, what we call a grid. You can start to run stuff at scale. And what I mean by stuff in this case is social posts. So it requires a couple of inputs, as you'll see. So maybe I'll make this bigger on each of your screens, but I get to select a social post type and I'll show you where those come from in a second. I want to give it just like a short amount of direction and I'll touch on that. Then I give it like a page URL so if there is, I'll say a blog Mateo posted that I want to, like, have opinions on. I can paste that in there or I can just give it like a full topic description and ask it to do extra research. So best thing I can do is give you an example. So today you might have, if you follow me, which I recommend you do on LinkedIn. We have a webinar coming up next week with an amazing woman called Crystal Carter. And I was late, come into a meeting and I was like, I need to get this post up. So what I actually did was this is a load of content. We have like, I don't know, 2, 300 words on this page and I trust this workflow now. So what I did was my post type is social engagement. I just gave it, like a bit of direction. Rather than saying, like, who it's for or like, what we're just going to cover. I was like, promote this webinar. I gave it the link and what it generated was three different posts. So each of them should be a little bit different. They follow different formats. And again, I need to get to that in a second. But each of them will have, like a different hook to begin with. And then maybe they're using of emojis. Their use of content on the page will change. But it gives me three options that now I'm like 80 or 90% of the way to actually like, posting within like a minute, which is class. So you'll actually see this one if I go to my. Where is it here? This is just what I posted. I like, maybe put it. Put a little bit of like, bolding around it, but, like, it was 90% there. So just in terms of using the workflow, you will leave this cohort with in production, like, this is exactly what you'll get. What I might do is so this, this is our grid that shows you, like, how it actually gets used in the real world. The post type is something under the hood that I have stored LinkedIn post topics here. So you will actually see maybe this is like what Matteo's prompt template is like. But a lot of, like, thought leaders on LinkedIn will, like, share out their, like, templates. Or there's another great website, I think it's called Clio K L E O Mateo put me onto this. But they have, like, amazing templates. So maybe you have an idea for content and you still write your stuff manually. You can come on here and be like, oh, I want to get like, eyeballs. That's my intention. I can then see like an example, I can see the template and then they'll just give me direction. And like this is like taking if our starting point is a GPT like prompt of like write me a social post. You've been able to give it even this as context will go from maybe a 2 out of 10 to like a 3 or 4 out of 10. So it's progressive. But this in Air Ops is what I start to feed into my workflow. I give it context. So the one you'll have seen there is social engagement. The outcomes should have been like this. If I go back, let's look under the hood of how it gets used. Engineering school, it wants to load up for me. So I have all of your workflows pre built so you'll see them soon. But this one social post generator edit workflow. So this is detailed. This is like far beyond what again I keep harping back to but like go to GPT, write me a social post. How this works is I give it the post type. So the example there was social engagement. I feed in our brand kit. So again I want to be able to write like an Air Ops person or if I have a very particular tone of voice or I'm like matter of fact, I can feed all that in and that's just like attributes that live in a different part. That's like our week one session and then give it directions which is just again, imagine I give it a big piece of content from Matteo and Mateo's cover 10 topics. I just want to say focus on X topic or focus on Y topic and then basically to do more research on my behalf, I want to be able to feed it a URL like I did. In that case, for the webinar, let's just grab the content from the one resource I know that like stores it all. Or maybe I have like GPT 4.1 came out this week. I have an example of that. I know nothing about it bar OpenAI's release note. So I wanted to do extra research on it. That's where I could give it GPT 4.1 and ask it to actually go and do a perplexity step. So there's a lot of potential inputs that go into it. What then happens is I find the templates, so I find the hooks and I find the format. I just need to do a bit of formatting, which is a python step. Then I want to branch. So I can only do one of these types. I can give it a URL or I can give it like a topic to do research on. So if I give it a URL which is like this example, I need to go to the page, I need to scrape the data, I need to understand it and then I need to write my posts. That would be my workflow planning. That's basically what we do. We go, we scrape the page and then at the far side I have three posts generated and you generate them one at a time. And here's where I want to show you some of the prompting stuff that we'll cover on the cohort. So hopefully this is big enough. But here we get to choose different AI models. We have like something like 60 plus available, we have like popular ones and then your other standard providers temperature is one that you'll want to control. So maybe you want it to like wildly fluctuate based on like creativity that's configurable. But really the core elements of any really, really good prompt come down to your system prompt and then like user assistant pairs. So what I mean by a system prompt is this like upper box. This is where you get the chance to almost tell the model like who it is. So you get a chance then to say in my case you're a world class LinkedIn post writer. Why is that important? Well this LLM made by anthropic is trained on the world's knowledge. It could tell you about dinosaurs as well as write you a LinkedIn post. So you just want to like from the get go give a clear direction. So in this case I'm saying it's a LinkedIn post writer and then I give it more context about like our brand. So I pass in our brand name and like our about section as variables which you'll start to see here. And then I wanted to write really good engaging posts that like maximize engagement details. We'll skip over for the time being. The second key point of a prompt is what we call a user assistant pair. So again, if I just go to GPT and I say write me a post that is me giving it like one instruction. What a user assistant pair does is you're almost hard coding a conversation into the LLMs history so that starts to understand what you're asking and how it should respond. So again, gloss over some of the details in this prompt for now. But I have a user user prompt and I have an assistant prompt. So what I'm doing here is I'm setting up all of the context and then what I'm doing down here is giving it a sample response on my behalf. So what I'm chaining Together is here is my ask, here is how you're giving it back. And then I do that like three times. So again I just give it like another hard code one. I'm like, okay, great, you did a good job there. Give me another post about something. And it does seem repetitive. It certainly is. But it starts to train the model that I'm looking for diversity of output. And then these are already approved by me that they are like engaging on brand and they just to meet my like quality bar. So stitching. I think I have three examples of these together. Only after I set up a system prompt and those ones do I then ask it to generate the net new one. So in that case I'm passing in the scraped content. I'm giving it the directions of just like hey, promote this or maybe promote this one aspect of it. And then I feed in all of the templates that we pulled above. So maybe just recap of like prompting things that are important. Know what model is good for the job at hand. Two, make sure you set up a system prompt and then three, use user assistant payers to start to again curate the model in the direction you want to go. So that outputs one of my posts. What I. And again, I didn't do this overnight. What I started to tweak then was if I just used Claude Opus, which is really, really good at content writing. Three times. It was very much the same. It wasn't really going in like different directions. So at the end I want three prompts to choose from. So simple thing, but I just changed the actual model behind the scenes. So I do Claude three opus, I do 01 and then I use Claude's latest one and now I really start to see and you can see it here in terms of even just like format, like bullet points. I get like numbers and then I get like different emojis. It actually looks very different. And now it starts to use different hooks and formats that I had given it originally. Then we just format them. But that without doing a ton of deep dive on the other side is like how you can start to go from like idea into like something that can actually hit production. And I will just quickly touch on if I ask it to do research. The model you should always try and use is perplexity or I guess like you can use their web service too. But again here I'm trying to do something similar. I've set up my system prompt and I should have done it here but the chat in this case is just like what I'm asking it to do so I might pause there. Folks have any questions? Before I take one more look back at an example I did for for this group of people.

We have one question from Noel saying like do you have a cheat sheet of source of what models you use for different tasks?

We do internally. I need to clean it up but I can promise if you come to the class we can share that. That's a good question.

I also have one question for you Owen, maybe for the class, sorry for the audience. How would you explain in layman terms the expected output or how system prompts and chat prompts work or how are they meant to work together or how do people think about them?

Great question. I I would say if you have them done right, it's like meeting your best friend. There's a level of context when you get to today's topic that it it just like the conversation flows if you do not have them, it's more like meeting a stranger. Like what they might say, you might misunderstand it. You're like why are they saying that? Or like what are they talking about? So just on the two ends of the spectrum where you have it done right, the output looks really good or it's like talking to your best friend, you're just like on the same page. If you don't have it done correctly, you have like a long ways to go. So like not setting them up correctly 2 out of 10 response go straight to GPT, have them set up correctly 8 or 9 out of 10 and in the middle there's like lots of color and it's like lots of different contexts you could add. But yeah interested if anyone else has a strong take or strong opinion on that.

Andy, maybe like you have a different way of explaining this given your exposure to arops.

No, I'm more. I agree with what Owen's saying and I honestly just learned something. I could probably speak to this but I've been in the habit of giving examples in that first user.

Input field.

Rather than which now that I'm thinking about it kind of bloats my prompts gives it way you know, sometimes too much context depending on the LLM rather than you're kind of like tricking it into like hey, here's what you produce into Owen's point. Like this is you know, here's, here's one of my LinkedIn posts that I know performed extremely well. You have that as an example and yeah the way I understand it is it kind of tricks like if you're using let's say 4.0 like, hey, here's what I responded with. And then in the next one you're saying like, hey, great job, now do it for this topic.

Right, so exactly.

It's basically like feeding, you know, like, it's like chopping one big task into a lot, a sequence of broken down smaller tasks that are a lot more digestible by the LLM at once. And it helps basically refine the prompt in a guided, structured way. Am I correct in saying so, Owen?

Yeah, I think that's a great way to explain it. I just asked our copilot how they would do it and I think they like echo a lot of like, what you said there, Matteo, of like, the earlier pairs are smaller elements and they're just for like teaching or training and they're not the main task. The last building block, to use your terminology, is the actual task you now want to complete. And then the prior examples are just as Andy mentioned. Be it a really good LinkedIn post you had previously that got a ton of engagement or something else, you want to show really good examples further up to the stuff stack and list as well.

Great, great.

Yeah. And I can't stress enough how helpful Copilot is in both the cold start, like solving the cold start with creating like new aops workflows or understanding features about it in very accessible terms.

What I'll also say is like this again to use your framing Mateo, like the cold star problem. This is intimidating where you're like, okay, great. It's actually easy for me to generate the system prompt or the first message. I know what I want to do and Copilot can get you there. But how you generate the subsequent ones of what this looks like. For another example, A, as Andy mentioned, you can pull it from your own experience or B, just run the workflow once without this user assistant payer stuff set up, get the output, tweak it as you'd want to actually see it. So bring it from the 6 out of 10 to 10 out of 10 and then feed that in as your first example and then just compound on it, do it again, add your next one, do it again, add the next one. Honestly, that's what I did here. Like I just had one that I really liked and I like compounded and I got three examples. But between copilot helping you with the cold start problem and then actually adding in examples that it then generates, you can very quickly go from just like a concept of like generate the social post for me to this like actual, I would say like 10 out of 10 or close to it, like setup.

And cool.

I don't see any more questions. What I took as a. An example for this class. Well, I think I just ran it earlier today, but I wanted to create a post type that was a bold claim, which is like Mato style. I would say I wanted to make it actionable for a cohort of people learning go to market engineering. My topic description was what came out, I think Monday GPT 4.1. It's new model capabilities. So if you remember the workflow, what happens when I don't provide it a URL is that it goes down this side of the branch and it starts to do like perplexity research. So I actually have it ran here. Like what perplexity does then was it pulls some live data and it says like, like an overview, does like a detailed explanation. So it's like my research assistant. It can give like examples and then it obviously generates the posts. But maybe you'll see one of these live from me tomorrow. But it then like outputted a few. Sorry, it outputted three social posts that are in that bold claim hook format. So this one, for my taste, might be a little bit too long, but it does start to provide examples to you of like, okay, it can be used for complex coding. It's like, good at performance and probably low cost and then really good for, like, lightweight tasks. Will it write you really good social posts? Like, would I replace it as one of my writers? Probably not. I think this just how I want to learn about it is pretty good. If I want to shoot it off to my audience, I can also do it. But it gives me a few different examples. Maybe I need to go back and change the prompt input. But again, very quickly you're going from. Especially in my case, I knew nothing about 4.1. I wouldn't know how to definitely speak to this audience about it. But very quickly I can, like, generate something with direction and a concept that will start to resonate hopefully with all of you. So that's just an example. We can do a live one as well if anyone has an idea. But in my case, these are probably too long. I need to go back and like, tweak one of the prompts or change the format, but just end to end. You're going from like idea. The workflow stays the same, but the outputs then look very, very different. All right, let me just check. It's fine. Cool. What else did I have? Whiteboard session. We'll see if we can get folks engaged on this. But I know when I started Growth engineering. I'd see so many tools. This is way back in the days of like segment and Clearbit as like the leading tools where you just had to always code in Python to make them talk to each other. It's now a lot easier to make like your tech stack talk. It's often hard to know how or when they should be stitched together. Like these are all the tools from what I know that you'll get hands on, you get to work hands on with over the course of this cohort. But maybe you have an idea of, I don't know, social post or outbound email or lead enrichment for your CRM. I don't know what ideas live in your head, but if we had a few minutes, I wanted just to take a few concepts from folks, drop it in the chat and maybe just map out how you would start to think about maybe using these tools together to achieve what might be your big bad idea.

Quick question here.

Yes.

I don't know if you're supposed to step in. Is it okay if I step in and now? Yes. I mean so everything we discussed so far the past 30, 40 minutes was mainly catered towards creating inbound, like you know, posting, getting people eyeballs on your website or on your LinkedIn and so on to get people to come to you. How could we use this to do outbound? I guess I'm curious to see demos and that. Or like, I don't know, like a sample workflow there.

Yeah, maybe give me an example of like a channel that you would typically associate with outbound.

Let's say cold emails.

Cold emails. Okay, great. And then sorry, keep you engaged. Give me a little bit of context on like your company or maybe like your industry.

We sell to mid market and enterprise restaurant chains. We sell, let's say, review management software. I'm oversimplifying it but for the purpose of the conversation, that's bad.

Great. So just to use these examples to keep it like air ops agnostic, if I can get rid of this and I'll try and answer Kez and iya's question in the meantime as well. So I want to plan out a full campaign. I am just joining your team as head of growth and I want to do exactly what you said, which is find users in my audience and deliver an outbound email to them. So what I would start to do is I would probably use clay to understand who's in my target account list. I want to start with the, the firmographic and demographic you just said there, which is like maybe restaurant Chains in the US Over a certain size of, like, employees, certain revenue count. You can do all of that that very, very quickly in clay and go from. I'll try and add it here, but we'll go from list of companies to list of people. Okay, that's a starting point. What I might do in parallel. So maybe not exactly connecting it into clay, but what I might do in parallel is make sure that my core positioning and messaging is actually set up in Octave. I think people would have been on that webinar. Why didn't that copy for me? But I can very quickly put in what my domain is. We'll say reviewrater.com and I can generate what my key messaging and positioning is for target Persona. Okay. So now what I have is I know who I'm trying to target, and I have key attributes available to me of what I want to say to them. That's great. Now what we might want to do is write the outbound email and, like, stitch these things together. So what comes to mind is I might use Air Ops at that point and then start to do. Why didn't. My mouse is broken. Here we go. So now what I can do is I can stitch clay, which might be my list of people, and Octave, which is how I want to speak to each of them to Air ups. What I would do in that case is, let's say my one example is I'm going after. I'm trying to think who you might sell to the like chain or like franchise manager.

Franchise manager.

So that's like ikey Persona. What I would do in Air Ops is orchestrate or the franchise manager, like, what's my key selling point? Maybe it's cost, maybe it's speed, maybe it's something else. So I would pull from Octave the core message. What I might pull from Air Ops is do I have a customer story that's relevant to them? Do I have something else on my website, like a blog post or my CEO did a podcast about it? So we'll say core content. And what I would do then is I would take all of this in, as you saw, try to feed it into a model, create the perfect asset. I'd want to review it. But then what I would do is I could create multiple things. I create an email, create a LinkedIn message, put it into one campaign, and then I can then send that really easily from Air Ops to Lemlist. I don't want that other arrow. You get the idea. So we go from the concept of I want to run a campaign Towards a target audience. The target audience part is A, I need to pull that list and then B, I need to know generally what I want to say to them. I then need to create the content for your outbound campaign. So on a per person basis, I'm going to want to really personalize it. So maybe it's a franchise manager, maybe it's the cfo, maybe it's someone else. But I take in that detail and pass that in as one input up the top of my workflow and start to pull within the workflow what's the key message from Octave, what's the core content from maybe our website. Stitch it all together into maybe like two emails and a LinkedIn post and then shoot that via API or direct connection to Lemlist. Maybe you use Smart lead, maybe use HubSpot. It's all equal. But that might be how I would try and achieve that like task.

Okay, that makes sense.

Cool.

I don't know if we have time for one more, but does anyone have questions on that or other ideas?

Thanks so much. That was awesome, by the way.

Super. I'm going to jump back to I think last slide on my side. Like you obviously saw what we'll cover in each of the cohorts. Our lessons I think are available online as well, but out the fireside with Air Ops, you'll get full platform access that gives you all of these amazing integrations and more you get. We charge on like a task or like a credit basis similar to Clay. So as you do more complex stuff, it just costs more. You'll have plenty of headroom for within the cohort and for actually like scaling your stuff in production, joining this class. And then I'm not the only one who will be teaching you. Mr. Andy Muns is there and then many other people who are like better builders than me on our team. They'll also be our TAs and you'll learn a lot more from them as well. But I will pause there and let maybe Mateo and Jared take us to the to the finish.

Fantastic on. Appreciate it. Yeah, if you guys have any questions, reach out to any of us and we'll be happy to respond immediately. Anything else, Mateo, anything else to add?

No, again, like our. Our DMs are open on LinkedIn. This recording will be also available on our YouTube channel right after this or like today. And yeah, if you have any questions about the course now or feel free to ask or if not, you can reach out to Jared and I on LinkedIn. Just FYI, we don't have many seats left. I think, like, they're just under 10 now. And, yeah, the course starts in about 10 days. So you can catch also us on the next week where we're going to have, like, a webinar that is mostly focused on outbound and personalized outreach. And so we can. We can see you there.

All right.

Thanks, everyone, for attending. Take care. Have a great day.

See you later.

Take care, guys.