Media/Appearances/YouTube
What is GTM Engineering?
dnsUNFILTERED/Benyamin Holley/
Transcript
0:00:00 Today we have Benyamin Holley, the original GTM engineer. And and you could say self-proclaimed original GTM engineer, but I would also agree with this cuz I've been seeing your stuff for a while, followed following your AI journey, AI-pilled to the max, and I kind of wanted to just talk about it. How are you? I'm doing great, man. Yeah, thanks so much for having me on. I was joking around with you on Slack.
0:00:30 I was like, when I worked at Sift, I was really disappointed that you didn't get me on the I wasn't actually disappointed, but I think I was a little bit more relevant to your audience at that juncture, but hopefully we'll have a great conversation today and somebody will get something out of it. So, that is a a really interesting start because the audience of anything really, including this podcast, every other public forum has really turned into leaned into AI. How do we use it? What are we doing? What are some of the strengths? What are some of the pitfalls? And I think you've been on this journey long enough to kind of understand those. So.
0:01:18 Uh I do have some questions that are very Claude-esque. They're kind of Some of them are hilarious. So, I did a uh a deep dive on Ben here with Claude. Some of it's really funny. I gave uh Claude access to a uh a more uh let's call it rowdy Slack channel that some of the GTM gurus are hanging out in. And that was a a fun deep dive. Claude called it spicy, which I would agree with. So, there's like this uh core principle, like applying software solutions to sales problems. And I'm just curious, what are is your opinion on where do most teams get that wrong?
0:02:02 Well, I think um if you take like a just a step back and like the you know, that that is my philosophy on a lot of like sales problems, and it's specifically around like sales efficiency for like outreach or, you know, daily sort of like task management, I guess. Um what I think the problem is is that most people don't understand like what actually closes deals or opens deals. Uh even very experienced like sellers like honestly a lot of people that nobody in my current company, but like people in the past that I've worked with like they uh you know, they just don't really understand what drives business and um like if you're going to apply software engineering principles to sales problems, you know, AI coding has turned these things and has turned us all into AI product managers.
0:03:00 We're not really software engineers, right? Um so it's more like the planning and scoping phase and like understanding the problems that are you're actually solving is like way more important. Like the jobs to be done problem is like way more important than like the nitty-gritty details or like, you know, where to put a hyphen in a Python script or anything like that's like useless information at this point. So that's I don't know, that's that's like where I would start with like where people um get that stuff wrong. Uh shots fired to the devs in the audience. Hi there. Hi. Yeah, no, I mean, I love the devs. Like we don't like I'm not definitely don't want it to sound like uh that, but like that information has become commodified.
0:03:54 Like that skill has become commodified. Whereas like sales skills, understanding business, like having business acumen, being able to like look at a book of business like a financial analyst and like stuff like this. Like those are skills that maybe are becoming commodified a little bit, but it's more about like like sales is all about like vibes, you know what I'm saying? So like that's a hard that's a hard thing to like get a robot to do. Yeah, what So what do you what's your thoughts on things that do drive business then? Uh like like what do you mean? Like how do you get someone to close? Like your the vibe, you know? What is what is it about the vibe that uh moves a deal across the finish line.
0:04:47 Yeah, well, I think like sales is a couple things. Like it's it's the process of like forming this might sound very like hokey pokey or whatever, but like it's a process of like finding a building an emotional connection between two people like working together to solve a common problem or well, it's not a common problem, right? Because like if I'm selling you Sift or AirOps like you have the problem. I I probably don't have the problem, right? Um And to me like what closes deals is like the truth. You know, if you can bring that out with someone and like get them to realize that you have a solution for them and that you hopefully want what's best for them.
0:05:37 I think that that that's what the best sellers like have is like they actually do want what's best for people. Um because when you like approach a deal in that regard where you're trying to help someone make a decision whether or not and it includes you. Um then that's where you build trust where you like build a relationship and where you can be like truly consultative.
0:06:05 So like as far as like applying AI to that I guess one example might be just like the ability to like do real research on a company. This is like nothing new, but um you know, go read their 10K, go read like press you know, press releases by their CEO, like go look at their job board, like what initiatives are they investing in and like aggregate this information like super simple use case it's like go ask Claude like what their problem is. But like that's how you start. Yes, so you're I mean you're basically saying to be human. Or good human at least.
0:06:48 Yeah, that's that's the those are the well, the reason I'm saying is like those are the things that can't be like automated. It's all the other random admin crap and like reporting and and things that you know, though that's where like AI is truly useful like at this juncture um, to allow people more time to actually be human, right? So, you recently wrote about dopamine hit workflows. Um, and I'm curious of two things. First, what the heck is that? And then, how do you spot that pattern um, early as a GTM engineer and like how do you avoid it?
0:07:30 Yeah. Um, so it's becoming so easy to like go down this rabbit hole. So, I was having a having conversation with one of our SDRs who I've been like coaching them on like some vibe coding stuff and I had him he set up a uh Versal database, you know, yesterday or whatever and he's just like asking about all this stuff and I cautioned him and like, "Okay, so just just be aware because this is stuff that I suffer from, too. It's not like I'm above it. Um, that it is way too easy to get off the beam and like go do a bunch of stuff that feels really good and like doesn't move the needle at all. And like honestly, the workflows and like AI tools and stuff that I've built that actually move the needle are like super boring things.
0:08:28 Some of them are are getting cool, right? And like uh, you know, as GTM engineering and you know, starts to move closer to like software and product management um, or at least that's how my thinking is evolving. Predilection for a lot of people like myself is like go start a bunch of like half-baked projects that feel good. Like what feel the dopamine hit is when cursor uh, you know, just starts like generating rows. And it gives you that localhost link and you click on it and you're like, "Wow." Yeah, exactly. Exactly.
0:09:06 And it's like, "All right, well now we need to step back and like, what does this actually do?" And I think that's the difference between I'm theorizing here cuz I've never worked for a software engineering team, but like that's the difference between like product and like product marketing and then the actual like software team is that the software the engineering team may be a little bit more removed from the problem. And so they really need to be directed. And in that same way, it's like if you're vibe coding, you know, the layers of abstraction of sort of like been you know, we're getting farther back on the layers of abstraction that we no longer have to deal with. First, it was like we don't have to do the punch card.
0:10:00 Now it's all on a screen. Then we could go copy stuff from Stack Overflow, so we didn't have so we could start with like an open-source repo. And now it's like you don't have to do any of that. You can just tell the robot what to build and it just goes and does it. So um the localhost thing is a great example. Uh you know, if I was like doing stuff in clay some like coming up with like some super crazy like email that's like crazy personalized, but it's like that's the doesn't really matter that much.
0:10:41 Like it's finding people who need what you have, who are experiencing the pain that you solve for and then like delivering them like a succinct, you know, message. Like not doing some weird personalization about like the weather in Houston or something, which I and I've done all that. So bald head and ginger beard. [laughter]
0:11:04 That's so funny because in Clay, I believe you can pull their profile pictures and then you could probably have GPT describe the picture. I don't know if that's what they did, but that's what like when I see these things, I'm like, "Oh, I wonder how you do that." And um that's not a good approach, but it was super interesting. And that and you know what? The thing is like it got that guy's attention is a thing. Like people like, "Somebody [snorts] sent me this dumb email that like talked about with some weird personal thing and I just didn't like it." And I'm like, "Okay, well, it didn't work with you."
0:11:50 and posted it on LinkedIn. Yeah, and everyone else to see too. Dude, you can't buy that. Like you cannot buy generating that sort of attention. So it doesn't always work and it's going to piss people off sometimes, but like mad respect to whatever insane outbound guy was doing that because it's probably hitting with other people who aren't posting about it. Like you don't know what's up. Yeah. It's like there's [laughter] that that was that was good. This is a great tangent. So, there's like this spectrum that's extreme of like totally absurd, you know, cold outreach email or call as like, oh, you know, I only work with all ginger people.
0:12:36 The other extreme is like super boring. This is our product. This is what it does. If you're interested, let me know. And like it's okay to try both of those things and everything in between, I would say. I mean, yeah, everything's a test. It's like, you know, A B C D to Z test on on all this stuff. Yeah, no, and when I was at Shift we did some really interesting outreach to like VCs and you know, other people and stuff sort of like in the early days. Uh that really worked.
0:13:14 Um and I won't go into detail about what it was, but like I did send some really crazy messages to some of the VCs and we ended up booking them and like actually getting a you know, our deal out of some of the stuff that would be like it's like crazy risky. But when you have nothing and there's like nothing to lose, like you can just experiment. At AirOps I tried to be a little bit more conservative.
0:13:46 Um but what I'm realizing is that uh what my like two cents on this stuff is like you should just be yourself and if you're super boring, that's okay. Like you don't need to be like some insane person to be successful. Um and if you're a literal insane person, then you should just like lean into that because you're not going to be happy like trying to fit yourself into a little box. Yeah, and if you haven't already, go go find Benyamin Holley on LinkedIn. His posts are hilarious. Cuz I think you're closer to the insane side of the spectrum.
0:14:28 Dude, people are like, "Oh, how do you come up with all these ideas?" I'm like, "I just like let the intrusive thoughts win." Um so, whatever is going through my brain, which is like always gotten me in trouble. It's kind of continues to get me in trouble, but not not too much like these days. Sometimes it's a win though, who knows.
0:14:54 You don't know, yeah. Yeah, you don't know. So, I heard you also mention um that everything or AI tooling like Cursor, Claude, are basically search engines now and that you were doing some research on like API that you were looking for and you just did some research, bought the thing, and then you were using it uh moments later and I'm just curious of like the gap between business uh systems availability through AI tooling um and you know, getting people to use it, find it, and use it through AI versus traditional methods of like search engine optimization and cold outreach.
0:15:35 Like how important is it to be where the people are, which is in AI tools like even ChatGPT, Claude, Cursor, whatever. Yeah, so this is one of the reasons that I joined AirOps. And if you don't know what AirOps does, it's essentially it's a end-to-end growth platform that helps your brand get cited and discovered in AI search tools or AI conversations. Um and when you think about the surface area of those tools, like they're becoming very fragmented. So, um one of the reasons I was excited to join AirOps is because um you know, there's going to be multiple winners in a category. It's not going to just be us and like one other company or whatever.
0:16:24 But um so, that was like exciting. But when I think about like all of the different places where people are using AI, like um you know, Amazon has their own proprietary AI search for like, "Hey, go find me socks." or whatever. Um pretty sure uh Shopify does as well. Um and so, it's not just chat GPT and Claude Um or or cursor, right? But think about all these tools are search engine services. [snorts] And so cursor nobody's tracking cursor citations yet, but I'm hoping that maybe Arabs could be the first to do it. If you're at a competitor company, don't take this idea. Just stop listening to the podcast. [laughter] We'll edit We'll edit this part.
0:17:12 Yeah, exactly. And yeah, so one thing I'll say is like when you ask like how important is it to be where people are looking at this juncture, traditional SEO is still the best predictor of AI search citation. That's based on you know, millions and millions of data point that we've compiled. However, every day the delta between how AI search works and a traditional Google serp query works is growing. [snorts] And so whatever company can predict you know, how that's going to work in the future is is good you know, going to win this space. But anyway, um yeah, I when I'm using cursor I think about it like this. Um people whether they want to admit it or not like shop for software tools the same way that we do B2C tool like consumer goods.
0:18:08 I mean, the principles are the exact same and I worked at a company called a few years ago that was building a telecom marketplace based on this principle. And one of the like you know, our value prop was like just buy telecom circuits just like you would buy stuff on Amazon or whatever. And it was really like hitting on a fundamental truth, which is that like B2B buyers, while they obviously you know, have you know, a different type of attachment to whatever tool they're buying. They were all Amazon pilled, right? Like the buying experience of B2B is like it trends down to like the lowest common denominator of like B2C purchasing because that's what people people buy way more socks on Amazon than they do telecom circuits. Right?
0:19:01 So that's like that's just what they're used to. And now it's like I use ChatGPT for like um help me get some stuff for my yard or whatever. Like go find the stuff on like go get me some links. Like people do the same thing with software. Like so all that to say um yeah, you do need to have your website optimized. You do need to for AI search and there's some slight differences. There's some nuances there, but we don't have to get into all that. But like um to me it's very important like and Claude Okay, think about this too, right? Like if you're working on something in Cursor or Claude Code or whatever, it has access to your code base.
0:19:52 Like it can read your code base. Then it can go read the API documentation for whatever tool it comes up with and make like a real recommendation. That's probably like way better than whatever the salesperson would But the salesperson's not reading your code base, right? Unless they're like really smart. Um but you're not going to give it to them. You probably shouldn't give it to Cursor or any of the AI tools to be fair either, but that's also another discussion. Um So anyway, yeah, it's it's the same thing and the AI's going to make a better recommendation. Yeah, that's actually really interesting and I I was just trying to think if if business customers or consumers use the AI tools more to like shop, what do you say, as a generic term. I would say it's probably even or maybe businesses using it more.
0:20:52 I I yeah, I don't know. [laughter] What maybe. I know that uh agents Okay, do you know John Barrows says? It doesn't matter, but like I was just curious. He's just a he's a sales guy on LinkedIn. And he talked about something a few years ago. Gosh, it has to be a few years ago by now. It's time to time flies. Where he said that eventually uh buyers and sellers would both be AI. That like and I don't think that we're going to get ever truly get there for like enterprise deals, but even like mid-market software um where like if it's something that has to integrate into Salesforce or HubSpot, like your cursor and Claude Code can go read everything in your Salesforce and make a determination on whether or not this tool has the correct API endpoints for like your Salesforce setup.
0:21:51 It It'll give you a totally personalized recommendation. That's again better than any salesperson. So um I don't know if people are using it more now, but I think that like so like I'm 33. I'm really dating myself here, but like uh the kids, the 23-year-olds who are 10 years younger than me at this point, um I don't think that they're going to even think twice about it. Like a lot of my younger colleagues are like they're like learn way faster than me at any of this stuff and they're probably smarter than me, too. I don't know, but like I don't think that they're going to want to go through traditional sales process for anything that like could be put on a company credit card that doesn't require an ACH deposit.
0:22:45 Interesting. So yeah, I guess it makes total sense because, you know, the younger generation is always native in whatever that thing is. Like native on the internet, you know, when I was a kid, there was no internet. [laughter] Um and now when when we were kids, there was no AI and, you know, generations behind us. So yeah, it makes sense. And you're right. I think that AI will buy from AI in the in the future. How long that future is from here, who knows, but probably faster than we think. I don't know either. Yeah, so this reminds me of like other funny like anecdotes about like people like oh AI's never going to replace plumbers. I'm like I don't know, dude. Have you seen the Optimus robots?
0:23:38 Plumbers aren't safe. Yeah, my my early career, my family owns a electrical contracting company. So, I'm electrician. You need any lights, you know, advice? Obviously, I can't use my hands from here. Okay. But, uh yeah, happy to lend some electrical advice, have my license, everything. We'll remote you in on the Optimus robot, and you can fix my lights from wherever you're at. Actually, no, that would be a good I'd like I'd like the doctor dial in on the robot things. Well, you've seen like the um some of the personal like assistant robots or whatever? I don't know what they're like do house chores. They literally have somebody in like the Philippines like remote in, Mhm. and they'll train the robot on your house and and the tasks that you need done.
0:24:32 So, And that's exactly how self-driving cars started is like they were gathering intelligence from all the new sensors that were there, and then eventually, they used that crunch that data, and now can determine this is a stop sign. This is a stop light. That's a weird stop light in Pennsylvania. You know, they have the edge cases kind of ironed out, and now they can turn on like full self-driving. And the same what thing will be true with these AI robots that do house chores chores, which I'm totally looking forward to. Dude, yeah. See, that's that's what AI really should be for is doing my laundry. There's like some meme on Twitter or whatever. It's like uh you know, I want robots to do my dishes so I can make art.
0:25:26 I don't want robots to make art, and I still have to do the dishes, which is kind of like where we're at right now. Oh, that's perfect. All right, so I did ask um the uh Clay community Slack uh some questions for you. One of them was, if you didn't have unlimited Claude tokens, so how would you find success if you didn't have unlimited Claude tokens? Okay, so First of all, we do get unlimited Claude tokens here AirOps. It's a great benny in 2026.
0:26:01 If you're interested in having unlimited cloud tokens, you know, hit me up. We're hiring for a ton of roles in the US and London right now, so um but second I won't say what I said on the Slack channel, but let's just say that like if I did not have unlimited cloud tokens, I'd be very distraught. And what I would do, I mean I was a really good seller. I've still in sales, right? Like I technically I work for a marketing team, which I've never done before, which is a whole other discussion, but um I would just hit the phones. I would find like I would probably like go work for HVAC company or something like sell HVAC stuff.
0:26:51 I don't know. This is totally random. um because yeah, if I wasn't doing what I'm doing totally dominate the marketplace. People would be like, "Oh my god, I have to buy this HVAC system." [laughter] Yeah, I don't know, electrical services. Yeah, electrical services. I would do something like more like localized building relationships like, you know, in in my region. um cuz I do good at like dinners and you know, public speaking events and stuff like that and uh Yeah, I don't know. That's what I would do. I would just hit the phones. Like I would just be like straight old school like Wolf of Wall Street sales person.
0:27:36 Cuz like I said in the Slack channel, there's a lot of ways to make money. This is just like what I'm doing right now. But I'm I'm like really super enjoying it, so um but it is all kind of like predicated on my like technical ability as well. Like my ability to like leverage AI tools and like enable the rest of the company to like do the same. Yeah. So like instead of uh Uber for dogs, Uber for dog food, you'd be Wolf of Wall Street for HVAC or electrical. Yeah, [laughter] you you can call me on that for sure. show. That's a reality show for sure. Yeah, I don't know not I have anything in common with like HVAC buyers, but I guess there's only one way to find out.
0:28:30 Right, you'd find it. You'd find it. Yeah. Uh so you so you talk a lot about um Clay, and I'm actually not super familiar uh with Clay uh with like GTM stack. Um but you were commenting that uh static workflows need to be like institutionalized across the company and run forever, not like a random uh type of thing like a one-off. Can you explain a little bit of that about that, and you may have to explain Clay for the audience.
0:29:04 So that's Yeah. Okay, so um Clay is essentially uh think of it like an eight-end for sales, um which is maybe not a complete one-to-one comparison, but it's basically like a spreadsheet with a bunch of APIs up at the like the top row that like do certain things that like enrich an email address in the pre- preceding cell, or they will um write an email with like an AI tool, and like it'll output it in the um succeeding cell, and things like that.
0:29:39 And so like you'll have this whole spreadsheet where you can like find people on LinkedIn, uh read their job history, enrich their email address, validate it, and then like write an email, and then like have another cell where you actually put it into the like the email sequencer, and it'll give you like outputs like yes, successfully put into the email sequencer, and things like that. Um I feel like people know what Clay is now, though. Like it's uh at least like in anybody listening to this, they should at least like heard of it, honestly. Um but so in the conversation that you're referencing is a kind of one that's coming up a lot, like oh, should I just use Claude Code?
0:30:29 Should I just use Clay? Because like Claude Code is obviously, you know, a terminal coding app and all this kind of stuff. And um you know, you can do a lot of the same stuff that you can do in Clay inside of Claude Code, but there's a lot of problems with like working in the terminal. Like first of all, you can't see anything that's like really happening. Like, if I I do this I build lists in Clay code like almost every day. Um and so I've noticed a lot of shortfalls, right? First of all, I hate managing sessions in the terminal. Uh somebody told me to use tmux. Uh I'll try it, but I've just been I've been trying to switch back to the desktop Clay code app because it's a little bit more user-friendly.
0:31:25 Um Anyway, the um And think about this, right? Like, when you're doing something in Clay, you have a total output on everything and you can audit every row. If you're doing something in Clay code for like list building, like, the only way to audit anything is to uh open the CSV in LibreOffice, which is what I use because I don't have Excel. Um And like, see what the outputs were and you can't really see like you can't just can't really see what happened. And there's there's a like visual layer that's missing. So, that's obviously like a very That's like one use case.
0:32:08 That's like if you're list building. The reason like I do it in Clay code is because it's just so much faster. And at this point, I've gotten the workflow down uh to because like all of the lists that I build are like a little bit different. So, it's not like I can just you know, have a static list like qualification table that like puts something somewhere. Like, it's like every time I would have to do it I have to build a new Clay table with like new parameters, new filters, and it's just it would be a huge pain in the ass. Like, I don't want to navigate a UI like that.
0:32:54 Um As far as like static workflows go, like, I'll I'll give you an example. So, um I built a job changer tool uh for the SDR team. Basically, like, it uh we have a tool that allows us to see people in our book of business who changed jobs. [snorts] Um it'll filter that based on a couple of things. I built all this in Cursor, by the way, which is why I'm explaining this. Um because this is a static workflow that I actually built outside of Clay um because there was like so many different moving parts um that a coding agent would just be preferable to my use case.
0:33:40 And also like like the way I'll do AI coding is like my code my prompts are like the dumbest prompts ever if you ever read them. Um So I'm like, "Hey you, please go fix this." "Why can't you do this for me? Why are you telling me to input stuff into a PowerShell?" In Cursor, you know, it's it's hooked up to common room, it's hooked up to this the Slack bot, it's hooked up to like email validation. It's got all these API keys like hosted in the Vercel project. And so it can essentially run the same type of workflow that you'd run in Clay, but it's just like way more flexible. And obviously, you know, it's an AI coding tool.
0:34:30 So everybody kind of should know what those do, right? Um But the other problem as far as like institutionalization, like one of the things that I'm working on is we're doing an enterprise TAM refresh. So trying to find all these new companies in the enterprise, add a few more data points to them in HubSpot and things like that. Like that really has to be done in Clay because other members of the team are going to be using this. And my boss um he said uh everybody needs to be singing from the same hymn sheet. So I've been like that's been like going through my mind recently.
0:35:14 And if I just have it like off in like some GitHub project somewhere, first of all, most people aren't even going to know how to access that. And they're definitely not going to want to. So those are like sort of the drawbacks and tradeoffs that that I guess you could make from like a I don't know what you want to call like a process design standpoint. So I hope that answers the question. Yeah, so you got two really interesting points here. Um First, that visibility layer. And we've seen this at DNSFilter. So like uh there are some visibility capabilities in Claude, Cursor, whatever. It can make diagrams and some things like that.
0:36:01 But the but that may not be preferential Is that a word? I don't know. But for that specific, you know, report or data set. Um so those visibility layers like Clay or some dashboard that you have probably won't go away. The there will be a new layer kind of layered on which is the text base version from uh the AI output IDE or CLI or Claude Code desktop or Codex. Um, and the [clears throat] second thing is the you can make like live artifacts in uh Claude Code now which are cool.
0:36:39 But what you're pointing at is that they're for you personally for you. And that is great, but wouldn't it be even better if that was somehow uh available to the whole company? And that is where you get like the multiplayer game of AI that uh I've been heard heard being talked about because right now it's kind of a one player until I run out of tokens. And how do we get that to the next level of everybody can access the same stuff at the same time? Like thinking of like uh editing in a group on a Google Doc or something. Like we're not quite there yet.
0:37:24 No, but that's interesting. Like I know like ChatGPT came out with like team sessions for a chat or something like this is probably like six or seven months ago at this point. And I I had it never use it. But um but yeah, like one of the things that I looked for a while back. The tool like this probably exists, but I was looking for a tool that would somehow integrate into Claude Code that would allow me to see the spreadsheet live for all of like the changes of a list build that I was doing so that I wouldn't have to like okay, Cloud open up the folder. Open up the file.
0:38:10 And it's kind of like how Google Sheets has like their like chat window on Google Sheets now or whatever with Gemini. This is not a me complaining about Clay. Like we're very happy customers. I've been using Clay for years. So but like one of the things that I wish was a little bit better um is like they have this tool called Sculptor which is like an AI assistant on the side panel, right? Like kind of like the Gemini Google Sheets thing, but it's just like kind of limited right now. And [snorts] if I could do everything in Claude and clay that I can do in Claude Code with the same speed and efficiency, like I would start translating some of that stuff back to clay because the the
0:39:04 [snorts] problem with clay is that like building a workflow in clay without like true AI assistance like you'd get in cursor becomes way more time-consuming and there's so much more testing, whereas uh cursor, you know, it'll run a terminal session, it'll monitor the outputs, it'll diagnose the problems in its own workflows, and then self-correct all by me saying the dumbest prompt ever, which is like, "Hey, go fix this," you know? So, that's kind of like where we're at with like these tools and where I'm at. If somebody has a better suggestion, like DM me for sure. I'm I'm all ears. Yeah, definitely leave in the comments. I'm I'm definitely struggling this with uh myself is like, how do I share the work? And it's like you can't you can share an entire chat if you wanted to, but uh that's like almost a snapshot in time.
0:40:04 Like for example, you were talking about boring creating the boring stuff earlier. So, uh earlier this week we had a like a cross-functional meeting of like status report kind of thing, and everybody has like one slide-ish that they uh present, and everybody's slide was made with Claude Code and looked totally different from everybody else's. So, our uh our uh CML was I was talking to him yesterday kind of getting some tutorials on Claude Code usage, and he's like, "What?
0:40:37 Everybody's slide like we got to do something about these slides." He's like, "What's you know, Gamma or there's some bunch of other slide creators." And I was like, "We can just make this in Claude." So, I've been fighting with Claude this morning on using our brand system to build a slide deck based on Claude's own PPTX skill, and it's actually decent, but it's not uh good enough for like public consumption, but internally should be fine. But like that sharing capability or capacity is the big limitation that I'm seeing right now. And we're finding ways around it with sharing skills, sharing plugins, having a MCP gateway, which is really cool if you haven't heard of that before. Um Now, what is an MCP gateway?
0:41:28 So, an MCP gateway, so you got a bunch of MCPs that you're working with, um Cloud and whatever. They usually have their own marketplace, which is really cool. And we have a bunch of those enabled. But there's some that don't exist there. Like Salesforce, for example, Zendesk, other stuff. Um like a YouTube, you know, thing doesn't exist in there. But you can make [clears throat] your own, or you can get one from the community somewhere. But that's a little risky. You should definitely check those out before you add any MCPs, but an MCP gateway Yeah, exactly. So, you can use an MCP gateway to now host all of your MCPs. They're all going through the same place. And if you tie it to your identity provider, um you can secure it with that identities of your employees.
0:42:25 And then you can give specific permissions on which tools are available. Uh basically like which API endpoints. So, it's a really powerful kind of connector layer, and gives us the ability to share. And we were having an argument this morning, you know, a lively discussion about um dev, you know, engineering, they want to uh have a lot of freedom to test out new MCPs and new skills. And having this gatekeeper in the way is um problematic. So, the I the concluded resolution that we recently we're talking about is just they're part of the team that is implementing that system. So, they have, you know, right access, management an you know, at least someone dedicated on their team to be a representative of let's say in that gateway system. Yeah.
0:43:19 So, yeah, that's where we're at. Is the MCP gateway would that be like Composure or like um there's like some other one that's [clears throat] Yeah, the ones that um I saw were Mint MCP, which is really good and actually the one we landed on. And then there's another called Runway. Um there's a few others. Uh some of them are open source you can self-host, which our dev team is kind of using and now going to migrate to the kind of sanctioned one. But yeah, it's a pretty powerful thing to implement at any company, I think. Yeah, actually uh speaking of the brand thing um our uh brand director Jess Rosenberg uh built something really interesting for specifically for what you're talking about where basically it had like entire specs of all of the colors, fonts, everything that we use.
0:44:17 It even had guidance on like we use like square edges um cuz you know how everybody want now is like the square circle morphism. Yeah, exactly. Like even the Riverside chat that we're on here has like rounded corners on the cameras. Um which is like fine. I don't really have anything against round edges, but um but she did a really good job like enshrining our visual distinctiveness into something that an AI coding tool can easily understand. So, it's funny when I went back to the one of the SDRs that was asking me about like these dashboards and stuff that he was building.
0:45:00 Um he was like, "Do you have on Jess's brand kit?" And um Jess, if you're listening to this, I'm sorry before you hear this, but I was like I was like, "Why do you care about that? You just this dashboard for you. Just like don't worry about it." Um but anyway, he did it and it looks really good and it's on brand. Which I do think is good. You know, I was like, "Dude, don't worry about it. It's just for you." And it it looks just like all the other dashboards and like slides that we would use for anything.
0:45:41 And so it is a small thing, but like I I know like marketing folks really care about that stuff. I mean, she did a really good job on it, too. But, it's interesting like if you don't have a skill for that, um I would definitely like look into it. We have a we had [clears throat] a pretty a very beautiful, extensive PDF of our brand system. Um so, what I did is I actually took that uh and translated into markdown essentially. Yeah, exactly. be fed into any new skill that we create um that is like a presentation of some sort or like some type of visibility layer. And I used it on um the new Claude Code design tool and all these plugins we're making. So, everything consistent brand identity, I think agree is important.
0:46:37 At least so you like kind of feel like you want to feel DNS filter when even at your own personal dashboard in Claude Code, but yeah. Cuz it's about the vibes. The vibes. It's all the vibes. So, you've been saying you've been doing um less terminal work at AirOps. So, the rest of the team can of do that stuff. And And sounds like you're giving them a lot of training. But, what is that How do you then um institutionalize their work um in practice? And we kind of been touching on this, but I'm just curious if you have any you know, like instead of Ben builds everything, what's happening now?
0:47:23 Um yeah, so first of all, I have like nine terminal windows up that you can't see, but uh I haven't been doing less terminal work. What I've been try But, I do Here's uh the difference. Um I cuz I get asked this all the time. Like people um they're like, "So, when do you use like Claude Code versus Cursor like versus Clay?" So, like um Claude Code would be for like one-off work, list builds, right? Or a query like, "Hey, go look in Metabase and see what the common thread is amongst like all these uh customers." Like, this is all kind of work This is all like I I don't want to say data engineering is too strong of a term.
0:48:14 But, like this is all like data work that would require somebody who like knows how to work a pivot table, which I don't know how to work a pivot table. I don't even know how to work Excel. Like, I could not put in one command in Excel. Um but with Claude Code, I'm like, okay, go cross-reference Metabased, HubSpot, and Gong, and go find like common threads on enterprise accounts so that when I'm doing outbound messaging, I'm actually saying the right thing and not just whatever I think is right cuz nobody cares what I think, right? They care what they think. Um and so that's like stuff that I do in in Claude Code.
0:49:01 Obviously, the list building in Cursor is where I'd be building static applications that would be hosted somewhere, and then obviously we covered Clay. Um but as far as like where this is going, um so I just invited a couple of the SDRs to our growth team GitHub um because uh how do I say this? In the past, people were not doing best practices around like GitHub stuff. Let's just say that. Um and uh what we It is not even just about like security, right? It's if somebody's vibe coding something that's really useful to them, we want that in our team GitHub.
0:49:43 And this is something that I started to sort of get at at Sift. Like, I had built a GitHub I I had was in the GitHub organization, and I was hosting like growth team things on that GitHub. Um and when you think about like where sales teams are going, like, I don't want to uh some people are going to disagree with this, and I I would agree with their disagreement just but like the way that I want to build out like the organization here at AirOps is like we treat um everybody here like a junior developer intern.
0:50:24 And if they vibe code a dashboard, it should go in our GitHub because now that's tribal memory, right? If we want to build something truly unique to AirOps, to our company, that only works for us, that's usable for the next 10 years or whenever we IPO or whatever. uh Um, then we have to start collecting all these like little vibe coded projects, right? And so the the really simple thing is this is probably not this is probably like super obvious to anybody who like has a real education in software engineering, right?
0:51:02 It's like everybody put everything in the GitHub. Like let's start there, you know what I'm saying? And then if somebody asks me they're like, "Hey, like how would you do this in the dashboard or like uh how would you improve this MQL automation?" Oh, I'm like, "Okay, well, I'll just log into the uh GitHub organization on Cursor and like use a cloud agent to like go do exactly what you just asked me about." Um, and hopefully like build on these problems and now you're building like recursive self-improving loops and tribal knowledge and like uh motes that are actually compounding um and like it just accelerates more and more the more that you do it.
0:51:50 Um, so I don't know if that answers the question, I think. It totally and it brings up a really good point that if everyone could use GitHub a little bit it would be so much easier. Like I'm I'm in this uh scenario too like how do we get the knowledge, codified knowledge in like markdown files on GitHub? Like that's the the company brain could easily be done on GitHub if everyone used the same if everyone knew how to do it and all the little projects they were working on like they're, you know, isolated uh uh conversations could be summarized as like one line of status in a project folder. And I I do have a very specific way that I work in like my directory structure, which is get committed, but it's just me on my team.
0:52:47 So, if every team could do that, I think the world would, you know, the AI world would be a little bit easier because AI would have all the context it needs. Um, and it does help that you can like dig into Slack and calendars and email. That's all helpful. Um, but yeah, if we can teach everyone, the sales, marketing, HR to use Git or GitHub, I think we'd be in a better shape. Yeah, absolutely. And it's not hard.
0:53:20 Like, you you know, people have like on the sales scene, they like ask me like, "Oh, how did you learn this stuff?" I'm like, "Dude, I'm not that smart." Like, it is not complicated. And first of all, and one of the other things, too, is like literally, like if I tell them like, "Hey, you should just copy and paste this Slack conversation into Claude Code, and it'll just do the stuff for you." Like, it'll literally do it for you. And like, yeah, it's like you had I had to learn a few things, like, you know, how does a database work? Like, just like stupid basic stuff that like, like I said, any It's like stuff that's like on the first page of any CS 101 college textbook, you know. Um, and it's not hard.
0:54:15 And it's just like it's just I think people are like scared to fail or whatever, which is like understandable, or that it's like intimidating. It can be intimidating, right? Um, but [snorts] for me, I was like, "Dude, I don't know, man. I got to learn this stuff because this is like how AI likes to work." Like, that's that's just workflow I'm really trying to conform to. It's, you know, I need all the help I can get. Yeah. Conforming to the AI work list. It's a weird world we're in here. Yeah.
0:54:53 So, we're the audience of this podcast and Dean is both of her we have a lot of MSPs, and I know you've done a lot with MSPs. I'm curious, um, you know, the speed that they typically move, if you're advising an SDR team or BDR team at an MSP, what's one thing that you would have them do like in this week to move the needle? So, also people ask me about this, and I have to preface this by saying I've never worked at or sold for an MSP, so my advice may be worth exactly whatever it sounds like, but for the sales teams that I did talk to and like, you know, you know, I've like consulted with some of the companies in the in the channel and stuff like that.
0:55:48 Um I think like one thing that every sales team needs, whether they're an MSP or a SaaS business, I mean, this is stuff I'm working on right now. It's not like we have it all figured out, right? Um it's just like get a better list of accounts. And um there was a workflow that I worked on with um uh Jeffrey Newton, who did sell for many MSPs, um who actually knows what he's talking about. Um where we went and used Surfer and like if you're an MSP and like let's say you sell to dental offices, if that's your lot in life, you know, I'm sorry, but like um let's say that's your bread and butter. Um somebody's got to do it.
0:56:38 Um like you could go use Surfer and find all of the dental offices and like a 100-mi radius like super easily because Surfer just like scrapes Google Maps and all you'd have to do is plug in your Surfer API key to Claude Code in the ENV file or or wherever [snorts] and be like, "Hey, go find me all the dental offices within a 100-mi radius that have more than 20 employees." And then it could either Google search to find whether or not they have 20 employees or if you have Apollo hooked up to it or LinkedIn, uh some other like LinkedIn enrichment tool. And like I bet you would increase your TAM by like 10 to 15%, which probably for a lot of people would probably be way more.
0:57:32 I'm being like super conservative there. Um that would be the first thing that I would do if I with any sales team that I would work with. It's like, "What What does your list look like? How can we increase the coverage by like using other data tools um and not just like, you know, buying lists from like dudes in Bangladesh." The dudes in Bangladesh are probably doing the same thing that you can do with Surfer. Yeah, so surfer.dev and there's other services that um kind of run Google queries for you. But yeah, that's actually really good advice.
0:58:12 Like start with your your list and you can build your own like ZoomInfo type of thing essentially. And another unlock is if you take that API key and add it to your MCP gateway and then put it into the IDE or Claude Code, then anybody can use it or you can determine who can use it. But anyway, another unlock. Yeah. Well, that would be
0:58:39 So now you you got me thinking about the MCP gateway because the SDRs are like asking me about like plugging stuff into Claude Code and I'm like uh I don't really just want to like give everyone an API key to like store in like their notes on their phone, you know? That is exactly the problem an MCP gateway solves cuz I was right there. I was like, "Hmm, uh the AE." I'm like, "You know, you could connect with Salesforce. You could do it with your API key, but you managing that and having some code running on your machine that does the API calls and all that. I don't think is a possibility. So, I'll send you my affiliate code for mittens and CP.
0:59:30 Okay, appreciate it. I'll have to get one first. Hey, Jake, you I need a I need a my code. Anyway, um you've been taking like really big swings. I'm curious what is what is the next big swing that you're really focused on? The next big swing like at at like AirOps as far as like stuff that I'm doing? Let's say things that you think would be valuable to everyone. Like in the in the AI world, GTM world, or at AirOps. Honestly I'm trying to think about like what I can talk about.
1:00:09 I I I kind of keep coming back to like the um this MCP gateway now because like when when people ask me about this stuff, um like when I was at Sift, I would run through our book of business in Salesforce on Claude Code. And so I had um you know, the Salesforce CLI, uh which is an incredible piece of software engineering if anybody hasn't used it. Um we're on HubSpot here at AirOps and it does not have the same capabilities. Uh Hubs- uh we've gone on a very small tangent, but basically like HubSpot is a true SaaS application where all the code in the back end is opaque. You can't really access it that well. Um so there's a lot of limitations around what you can do.
1:01:02 And in Salesforce, Salesforce isn't even really a CRM. It's like three data warehouses in a trench coat. Like all the code is exposed and you can like manipulate it in very interesting ways. I tell people I I could provision security protocols for users in Claude Code for Salesforce users. I was like, "Dude, this is crazy." Like everything that you can do in the UI you can just do in Claude Code. Anyway. Um when I think about like multiplying and creating leverage across a team, um that's what I think of. Like I would go through the book of business in Salesforce on Claude Code and basically I'm like, "Okay, we have these I'm going to go through these 10 or 15 accounts today. We're going to hit each one of these people.
1:01:56 You're going to do like deep research on them and then we're going to write them a customized email." And this was like one by one. And maybe yeah, you could probably automate this, but the the point was that I needed to actually like have some human review on this. And um like do some intentional prospecting, go like one of their posts, like stuff like that, right?
1:02:24 Like um like I would love for people to be able to systematically go through their book of business in that fashion. Um because like the kind of the way it works like for most companies if they're doing this stuff at all is like they'll do like mass research and then like put somebody into an policy sequence or something. And obviously from a time committal perspective, like revenue operations and like maybe a GTM engineer if they have somebody like that on their team, like they're probably better positioned to do the sort of research and data gathering that you would want to like use in outreach.
1:03:08 Like, "Hey, I noticed you're hiring like five SEO people. Like what's the story?" Or hey, I noticed that your SEO team got cut in half last year. You know, could AirOps like help you scale out that function again and without adding headcount? Um And the the reason like I would want people to be able to do this on their own is because like the sales team is always going to be closer to the problem than I am, right? They're always going to have more intimate like informational leverage. They're always going to be like their vibes are just going to be better.
1:03:50 Like, we're working with a you know, these people and they'll always know more than I will about like what to say to people. Um and if you can like give them the same tools to like create information and leverage and like do outreach at speed to people, um that to me is like sort of like the next frontier that that's that will require uh two things like on my end. [snorts]
1:04:20 Um the hiring won't really be on my end, but like you know, hiring people with a predisposition to doing this sort of work who also aren't going to get distracted and bogged down by shiny object syndrome and like get distracted by like vibe coding like useless nonsense. Um and yeah, I don't know. That's sort of like where my mind goes because like I have a lot of skills like I can go create pipeline on my own and it's like I need other people to be able to do this as well. Um so that requires like cultural and technical technological like enablement. So, that's kind of like where what I think of. Yeah, that's what I was thinking.
1:05:08 Well, you enablement is really the the key ingredient there. Enabling people to be more like you. [laughter] Yeah, well, be Yeah, native. They should be themselves. They shouldn't be like me. Right, right. Your skills, their personality. Yeah, exactly. So, I got five rapid fire questions. Ready? Yeah, I'm ready. Cursor Cursor or Claude Code? Um both. Good answer. Smart lead or email bison? Definitely email bison. Shout out to email bison. You know what's funny? I have no idea what either of those are. Well But Claude does and you do. You use email bison anyway. Okay, you email bison, got it. Uh one sales convention you you would just die Wait, hold on.
1:05:55 One sales convention you wish would just die. Oh god. Um Please say business cards. [laughter] Oh, no. Um People getting drunk and making career-ending moves. [laughter] Woah. Woah. That's the for the That's for the after-dark special. Oh, well, everybody's seen it. It's not like we don't know what happens, so. Yeah, that's true. All right. Honda VFR 1200X or what's the next bike you're eyeing? Uh Transalp 7 Transalp 750 or a Tiger 900. Awesome. I'm going to look at Okay, look those up. And then, this is a funny one. The one AI hot take you defend to a room full of skeptics. And it's funny because you're hot You would defend your all of your hot takes in any room. Absolutely.
1:06:45 Um What's at the top of the list, though? Uh the the top of the list is Jeez. Um Oh my god. Um Well, what's coming to mind is sort of like what what I said earlier, like don't go learn Python, in my opinion. That's just my That's just my hot take. Obviously, there's limitations to that, like maybe you should learn Python, but um I'm skating where the puck is going, and the puck is going to a world where the limited amount of memories that I can hold in my brain, like knowing how Python works is not going to be useful. That's a good mic drop to end on. Thank you, Ben, for uh hanging out and
1:07:33 [laughter] and uh you know, telling me telling the audience about GTM engineering. I think it's like a a thing that no one really kind of understands. They kind of um uh have heard the term and or kind of know they need somebody like that on their team. Uh but I think you're living that reality and the the rest of the audience could figure out um a little bit clean into what is kind of the next wave of of SDR uh AI-enabled SDRs and marketing folks. So, thank you. Yeah, thanks so much for having me on. And uh really hope that this was like helpful to people. And if anybody has any questions, you know, feel free to DM me on LinkedIn.
Mikey Pruitt has Ben on dnsUNFILTERED, episode 54, while Ben was at AirOps. Teams go wrong applying software to sales because they do not know what opens or closes a deal, experienced sellers included. AI coding turned the job into product management. Where the hyphen goes in the Python script is the cheap part. Sales is still vibes. What closes, in his telling, is the truth: help someone decide, and the decision might not include you. The simple research loop is a 10-K, the press, the job board, then ask Claude what the problem is. AI earns its keep on the admin, so there is time left to be a person. He had just warned an SDR who stood up a Vercel database. The workflows that move the needle are boring. The hit is Cursor spitting out rows and a localhost link. Weather-in-Houston personalization is something he has done. Finding people who have the pain, then saying it in one clean note, matters more. At Cyft he sent VCs messages risky enough that he will not describe them. Some became meetings and a deal, because there was nothing to lose. At AirOps he pulled that back. Be yourself. Boring is allowed. He treats Cursor and Claude as search. In the AirOps data he was citing, traditional SEO was still the best predictor of an AI citation, and the gap versus a normal Google query was widening every day. A buyer with Claude Code and the codebase gets a better tool recommendation than a salesperson who never read the code. He also says you probably should not hand that codebase over. John Barrows had said buyers and sellers would both be AI. Ben does not buy that for enterprise deals. He does buy it for a mid-market tool that has to fit Salesforce or HubSpot, and for a 23-year-old putting software on a company card. On the tools: he was building lists in Claude Code almost every day, because every list is different and a new Clay table is a pain. Clay lets you audit every row. In the terminal he was opening the CSV in LibreOffice, since he does not have Excel. A job-changer for the SDR team lived in Cursor, on Vercel, tied to Common Room, Slack, and email validation, because it had too many moving parts for a table. The enterprise TAM refresh had to live in Clay. His boss wanted everybody singing from the same hymn sheet, and a GitHub repo the rest of the team will not open does not do that. He invited the SDRs into the growth repo anyway. If someone vibe-codes something useful, he wants it in GitHub so the next question can go to a cloud agent on that repo. Claude Code cross-referencing Metabase, HubSpot, and Gong is how he checks that the outbound line is what the accounts think. For an MSP sales team he puts a caveat up front: he has never worked at one or sold for one. The move he would make this week is a better account list. With Jeffrey Newton he used Serper, from Claude Code, to pull dental offices inside 100 miles with more than 20 people. He would bet that lifts coverage 10 to 15 percent, and he called that conservative.
Notes
The workflows that actually move pipeline are boring. The dopamine hit is Cursor generating rows and handing you a localhost link.
Weird personalization, weather in Houston and the rest, is something he has done. A short note to someone in the pain matters more.
At Cyft, risky notes to VCs booked meetings and a deal because there was nothing to lose. At AirOps he ran that tighter. Boring is allowed.
List builds lived in Claude Code because each one is different. Shared work, like the enterprise TAM refresh, had to live in Clay so the team could see it.
The SDR job-changer ran in Cursor on Vercel, with Common Room, Slack, and email validation. His prompts were along the lines of 'please go fix this.'
He does not know Excel. LibreOffice on the CSV was the audit. Metabase, HubSpot, and Gong in Claude Code were how he checked the outbound line.
SDRs got invited into the growth GitHub. A useful vibe-coded dashboard belongs in the repo, or it dies with the person who built it.
An SDR asked for Jess Rosenberg's AirOps brand kit on a personal dashboard. Ben told him not to bother. He used it anyway, and it matched the other decks.
MSP caveat, in his words: he has never worked at one or sold for one. First move is still the list. Serper, dental offices, 100 miles, more than 20 people, a bet of 10 to 15 percent more coverage.
EmailBison over Smartlead. Cursor and Claude Code, both. Hot take he would defend in a skeptical room: do not go learn Python.
Get new episodes in your inbox
