10x Your Marketing With These AI Workflows | Kieran Flanagan (Hubspot)

Hey, Kiran, welcome to a show's podcast. Hey, I'm excited to be on. Thanks for asking me. Awesome. So a lot of people know you from your LinkedIn and from the podcast that you are doing, marketing against the grain, but you also have a full time job. Basically, your SVP marketing AI and GTM as your LinkedIn says at HubSpot. So to set the context, I wanted to learn just a little bit about what your responsibilities look like in the troll at HubSpot, what kind of team do you manage, what you're responsible for just to set the tone for what we're going to talk about. So I was at HubSpot for 10 years. We kind of grew quite a lot from, I think, when I was there, maybe two, 50, 300 people to 7,000 people. And then I went and left to did some time at Zapier, which is a really great company, whereas some people say Zapier, and then I'm back at HubSpot. I think the way I work is somewhat unique in that I kind of take on these two-year missions. So that's what this way I think about work is like, there's a really hard problem or something that needs to be done. I like two years because it's not so long that it's way in the future, but the two years is long enough that you can actually make real impact, like do something transformational. I think my role in HubSpot is like, I break that into three missions today. And it's, we are integrated in AI across our go-to-market. So we believe AI was going to be transformational. When I came back in May 2024, we stood up seven large, we called them P0, and most important bets to make an AI. And the way that we think AI was going to augment go-to-market motions and really changed the way that we grew businesses. And the reason that's important for us isn't just for internally that we want to do those things for ourselves, but we want to learn what works so we can integrate them back into our platform because we have a go-to-market platform. That's one. The second one is we have always tried to over-deliver for small businesses. That is a really big part of our brand trying to help small businesses kind of act like larger businesses, give them the tools to do that. And so we kind of want to rebuild our business a little bit in terms of our dime market strategy across product across go-to-market across a bunch of different things. And so that's another one of my missions is kind of really making sure we have a work-class a business for the lower end of our market. And then the third one is we are known for in-bign. In-bign was a transformational story. In-bign was really built on the premise that don't interrupt what the customer is interested in, be what they're interested in. Now that still exists today, the fundamental elements of in-bign, why we Brian and Darmech created in-bign still exists today. But how you go to market, use in in-bign how you do in-bign has drastically changed. And we want to kind of stand up a new playbook that helps our customers grow in this AI world from the learnings that we've kind of learned over the last two years or the things that we've learned over the last years. We call that loop market. And so that is how we change our own go-to-market motions and then we how we teach customers to replicate that for themselves. And so they're the kind of three core missions I have today in HubSpot. It's like over a whole go-to-market motion with AI, really transform the lower end of our market and then transition from in-bign to loop market. And or low loop market is like the premise of in-bign is still the same, but how you execute. Well, let's not go into loop marketing just yet. Yeah, too too too too many too many phrases. Yeah, let's focus on transforming marketing with AI because this is the actual topic that I want to discuss. Yeah, just quick question, how big is your team? So my direct team is 350 plus people. That's crazy. We'll talk about using AI for managing the team down the line. But yeah, the first question I want to ask you, since your job is literally kind of figuring out how to use AI to transform marketing, what are like top three things that were automated with AI at HubSpot in your marketing? So what comes to mind like what were the most effective, the coolest thing that you've have to automate it with AI? So in terms of like instrument and AI across our go-to-market in terms of what marketing did or or basically how marketers work because I think there's two parts of AI. There's like AI gets instrumented into how your customers experience your bio-injury and then there's like just ways that we work. So I can give you maybe some of the biggest the kind of things that we've done that have really impacted our overall results. Yeah, let's try it. Yeah, let's dig into that. Yeah, and so you know the first one, they're all they're all going to be the things that people have talked about, but we kind of just go into the weeds of them. And so one of the big ones where we were pretty early in instrument and AI across all the prospect and so we have all of these email flows that prospect into people who come to HubSpot through all these different ways and we book time with reps. And so previous to AI, those email workflows existed and the personalization was done on segmentation, right? So B2B forever has been a segment-oriented way of marketing. Hey, like you have the this job title, you have this company in this geography of this size. So this should be an email that you like. And email that AI has fundamentally different because you can actually personalize at the contact level or the even the company level, the contact level and truly create something personal for that person. And so we instrumented AI across all of the email workflows that we have. And that led to a 30 plus percent increase in meetings being booked every single month. And there was meeting closed at the same rate. Like they were they weren't worse close rates actually in some respects, they are better close rates. Now we have started to instrument. We take one chunk of learnings. So we learned for these income and leads and we prospect into them. We book meetings. We learned we could personalize really well. Then we integrated that into how our BDR team worked. Our BDR team is a really big team. They look into our database and they extract demand and they book time for sales reps. Now there's a kind of there's a lot of like miss misunderstandings about AI. If there was a like early, early, early post-aligned LinkedIn would say, hey, your BDR team is going to get replaced. Well, the average BDR team what they actually do is they do sequenced campaigns to book time with prospects and they do that through email calls and social outreach. It's not just one channel. It's usually the combination of all these three channels that are great to meetings. AI cannot automate calls yet. Right. You're not going to have AI call someone. And for the most part, we don't like to automate things in people's social inboxes because it's a little more personal to them. So AI in that respect can actually increase the productivity of the BDR team and allow them to book more meetings, but it enables the human to do that. It does not replace the human. So for example, AI creates the sequence for a BDR. It creates the call transcript because we're able to gather all of the context about that person, create a great call transcript, creates the email sequence, creates the social outreach, but the BDR is in control. And so we are automated in all of that work throughout our kind of sales work. Now that sounds like, well, that sales work. Well, I actually think what you'll see is because marketers are the true automators, they're going to start to do much, much more of the buyer journey. And that's like one example. Can interrupt you and talk a little bit a little bit about tooling. So how exactly is this done? Like, how do you so what you're saying is that let's not focus on outbound, let's focus on inbound. So leads are coming in and you're saying that emails to those leads are automated because right now at a Trev's, we're a little behind on that. People like our sales reps are sending their own emails and we're also looking for ways to automate it with AI. So we're using HubSpot by the way. So we're happy customers. What do we need to do to automate our own emails to people kind of who come and raise their hands? We want to talk to your sales team. What needs to be done to to have AI book those calls for our sales team? So we have, so first of all, your data set. So you need to have like pretty great data, hygiene, and have a good data set that allows you to truly personalize for that prospect. And so that could be a combination of HubSpot data or external data that's like very applicable for your market. So how do you guys do it? Let's talk about you. How do you do it? We do it. We have a, so we are a 9000 person company. So we are not a, we are not our, you know, sweet spot in terms of customers, which means we customize a lot of HubSpot for our needs. So we have a product and engineer in team that has dotted line into that AI group that I told you about in terms of our overhaul and the good market motion. And we customize a lot of HubSpot. So we have a custom prospect in tool built internally for us because we have. Okay, let's talk about us again because for you, it's like very custom and you have a large team, but for someone like a Trefs who is using HubSpot and we have those leads coming in and we want to automate our emails that were sending them. What do we do? So again, like you, like If you're doing it through HubSpot, I think the first thing is, do you have the right data set for your customers? And so again, it's a combination of HubSpot allows you to capture data, it allows you to enrich the contacts to companies. But maybe there's another data set that is very unique for your audience. Like one example for you maybe, there's a unique data set to tell you if that company has a search problem or a visibility problem on the website. You can use it. It traps all data. We can see how much traffic their website gets. We can see if the traffic was growing down lately. Yeah, we can come up with our like internal signals of search data. Yeah, LinkedIn data is another one that really is pivotal when you're trying to personalize something for someone. So a combination of data sets that make your email unique for you and really personalize it for that prospect. Then you can personal-- when you are customizing that email, one of the tips is to condense the-- or when you have a prompt, right? So the way this works is you would say, hey, AI, when I give you this information, personalize an email for that person based upon that information, right? So that could be a singular prompt to craft that email. The way to do this, actually, that we've learned over time, is to modularize your prompt. So you would have a prompt for the subject line, which is its own prompt, a prompt for the introduction of the email, a prompt for the body copy, a prompt for the CTA. So like, split out your prompts, makes them much, much better, and can create a more personalized email. And then the other thing I would say is, it took us a long time to figure this out. So when we look at-- when we integrate it, AI across all of our workflows, I have this rule when we're doing something new that you have to demonstrate that it's going to be of value in six weeks or move on. So it tries to keep you really focused on what's going to be impactful. That took us like four months of real iteration. And really, what we had to do was sit with the best sales reps and truly understand how they crafted that copy and be able to replicate that through the prompting. And so there was a lot of trial and error before we got to the success that we've had to date. Back to how someone like us would set this up. So yeah, I can see how you can develop a custom solution to tap into HubSpot data, to enrich it, to pull HF API to get some information about their website and blah, blah, blah. Do you know any kind of plug-and-play solutions for this kind of thing? Like AI as DR agent, which would plug into HubSpot, where you would set up those prompts. Like you said, you can have a custom prompt for subject line, custom prompt for even, I don't know, introduction of the email out of the email blah, blah, all the rules. Or right now we are at the stage where all of these has to be custom built within-- No, HubSpot, you can customize your emails. And we have some developments coming that allow you to have more control on how you do that. No, I don't know. Like your account manager would be able to take you through that, rather than I can specifically take you through where we are and what we've released. But you can definitely customize-- we are teaching customers how to do a version of this. The only reason we customize it is because if you just take our sales team, it's 800 sellers, right? If you take our BDR team, it's 400 BDRs. So-- The numbers are crazy. Yeah, so like the-- A lot of it is just like there's like added complexity when you get to that size. So we're still using HubSpot, but we have to augment it in certain ways for that team to be able to use it because it's such a complex-- we have a very complicated business. But if you're just like, hey, when an inbound lead comes in, I want to trigger an automation workflow. I want to personalize that email using prompts. We have the tools to be able to do what-- at the most basic form do what I said. The one thing you would have-- I'm sure you're doing this, right? The one thing that is a really important part is like what combination of data allows you to truly personalize that email for that person. So when you were 12, 80 months ago, if you just did AI personalization, your email would stand out because it's like new, different. You're at the kind of bleeding edge of what folks were doing. Now, most people are kind of doing that. And so the way you can stand out from every email feels quite personalized. The way you can really stand out is like trying to get a unique angle or data set that really speaks to your audience. And there's been some pretty good examples of companies that I've seen like ARFs could do using different types of traffic sets and things like that to really customize and personalize that email. OK, so again, full disclosure, I'm not super familiar with HubSpot. I'm only learning the ropes. But what you're saying is that the automation in HubSpot already allows for integrating AI to help you craft and send the email automatically. Oh, OK. OK. Yeah, yeah, yeah. So it's mostly about-- Everything I said you can do within HubSpot, but there may be some additional data sets that are unique to your audience. But I think-- and if you're a very, very large company like this, there may be some customization that you want to do that kind of suit your go-to-market motions. OK, got it. Let's switch gears from email to content, because content is a large part of marketing. I would even say that email is more on the sales side. And content is more on the marketing side, at least, in my world. Yeah. So what is your HubSpot experience so far in terms of using AI for creating all sorts of content? OK, so I think-- let me just see if I can get this up. I want to show you something. I think AI is a great content-- assistant, not a great content maker. But I can give you two quick examples. And I'm going to see this up here to see if I can show you. So we have internal tools, but I'm going to show you-- I actually built some of this last night just to illustrate this, because I knew it was called this podcast. And so let me see if I can make this work. All right, so what do I mean by-- like let's give one clear example of AI's ability to-- I would say remix content, but actually, I think of it as a pretty good ideation partner. And so one of the treasure troves that most people know about now, but I was really on this 12 months ago, is like, I just thought it was amazing that you had all these transcripts. So YouTube-- because if you think about it, what's in the training set? So the AI-- it's all the text-based content for the most part. Now, open AI have trained SORA, theoretically, on YouTube data. But these videos are not transcribed in the data set. So there are real treasure trove of content that you can remix into different types of ideas. And so I want to give you a quick example of what I mean. But that-- so if we go grab the transcript for one of our videos, this was one we did with the PM of Google AI Studio, Logan Kilpatrick. Now, there's an automated way you can do this, but I'm going to just do it the manual way, because most people would be doing it this way. I'm going to grab this transcript. And then I'm going to actually see the changes. So this is an app version, but I could show you how to do this in cloud. But like, think about this. This is what-- the way I think about this is you or anyone on your team will have a personal suite of internal tools when it comes to content that fit the way that they want to work. So this is one of the ways that I like to work, which is I take that transcript. And it basically-- I have taught the AI how to extract what I call viral talk and points. So what are viral talk and points? They're kind of based actually on how the YouTube algorithm works. And so it extrapolates them into what viral talk and points would make a good 10 minute, 12 minute video, or would actually make a pretty good social post. I can show you how that works. So it pulls it into three different categories. Looking for anything from that video, anything that makes a really great educational post. An educational post is there's one real tangible lesson that is unique in terms of how it's articulated because there's a clear example of it. So I think if you pair education with examples, it works really well. Spicy take is like the unpopular opinion. The counterintuitive take that works really well on YouTube. In fact, maybe 2L on YouTube. And then data nugget, it looks for like one really clear and crisp data point that it can craft this kind of viral talk and point around. So the first thing it does is it will go and start to categorize. It looks in the transcript. It starts to categorize these. And then it gives me these kind of talk and points. And so it will say, hey, AI generates unthinkable strategic ideas. What if AI gives you ideas so radical you never think of them yourself, Gemini three is June just that. So that was a conversation we had that AI is not very strategic. It gives you the quote, it gives you the proof and it gives you the value. So how do I use this tool? It's just an ideation tool, right? I see a really great video. I watch that video and I'm like, hey, there's so many great ideas in there. I'm going to have AI start to categorize that idea. And it pulls out these, I have actually eight of these different talk and points and viral talk and points. And so it pulls them into like these educational data not good spicy take. And I can show you the next part of this where you can have AI craft a first draft and tell you what I think is valuable and not valuable about that. But I'll pause here to see if this is actually making sense. And the point I'm making here is AI is very-- I think the difference between AI and traditional software that maybe people miss is AI is very personal to you in terms of how you like to work. So this app you might like. look at it and say, "It makes no sense to me. That's not the way I work." But this is the way I work. And now I don't have to buy an off-the-shelf piece of software and try to make it do this. I can just literally, this here, I just, this was like 10 minutes of work, right? And so, like imagine what you can develop with another kind of hour or two hours and customize it for your needs. - I have two thoughts in this regard. - Yep. - The first thought is, we're recording this on Riverside right now. And I kind of expect someone like Riverside to be super interested in developing something like this for me so that I wouldn't have to wipe code it myself. But the second thought, kind of answers, and you kind of answered why the first thing won't probably be possible because every person has a unique working style and every person would expect this tool to work a certain way. So it's almost the value of this tool for you is in the kind of system prompts that you're putting in there and that you're refining based on if the output of this tool is matching your expectations. So yeah, I guess the tool itself is easy to wipe code these days like tools like CloudBull get to incredible MVP's very, very fast. But the directions that you give to AI to extract those insights and to process those insights is what makes your tool unique. - Yeah, and so like what you're saying, so basically the thing you're saying is what I believe, right? So AI works best when you have deep domain expertise, right? Because what I was able to do is I created content for years. When I left to become a, when I was a software engineer and I transitioned from software engineer to marketing and within two months, I had a blog set up. I always love creating content. It's part of what I've always loved doing. And so I was able to systemize the prompts to show AI how to look for these things. Now AI is getting better that if you basically took this transcript and just went to Claude Opus 4.5 and said, look for the most spicy takes that would perform pretty well on social. One tip here is don't say LinkedIn because when you say LinkedIn, what the AI does, it looks to see what is examples of content on LinkedIn and the average piece of content on LinkedIn is terrible. So you don't want to use LinkedIn as an example there 'cause it will try to create something that is for LinkedIn. But if we take, just, if I kind of just carry on this here example, so again, I actually have a lesson Claude but I wanted to show you a more visual version of that 'cause it would work pretty better for your audience. But if I show you the Claude instance here and so let's say I decided to actually take one of those, one of those takes, let's see how I've got one here. Okay, so continuing on this workflow, I have some things here I think I can show and that's pretty mess. Port of my own financial advisor seeing people looking at this show what I do here. Okay, so let's see. All right, so basically I have these different projects, right? So I will say my AI usage is incredibly messy and that's something I need to clean up myself because I have a Claude personal account, HubSpot account, Gemini personal account, HubSpot account, an open AI personal account, HubSpot account. So I probably am a little messy in terms of how I use this. But like that's, so if we go back to the kind of thing I just showed, we kind of took one of those talk and points and let's say it was a spicy take. Then I have a project set up that's really specialized in turning these kind of talk and points into a first draft. And so it will take one of these talk and points which we just, literally just copy and paste. There's a copy button on that app. You paste it in here and it will create a first draft. Now people will say, "Hey, can you not just create a first draft in that app?" Yeah, there's a version of that app I have and it will just click above and it will create a first draft. But I actually just like Opus 4.5 and Sonnet 4.5 is a writer. So I have to take in there and put it here. Although I do have this built into a project as well. So that thing I showed you where you can just add a transcript and take the talk and point. This is it here. It's just as a non-Ui one. This is the one I actually use, but it's just not as pretty to show when you're trying to illustrate the point on a podcast. But you can see here it's the exact same thing. And so what is this useful for? Do I think AI is a good content creator? No, I would never use AI to ship content. Because I continue to believe we are going to a place where AI will force people, they're not force people. AI will mean that the content we want to consume will be much more from individuals and brands. I actually think AI is going to be the, somewhat the death of brand content and continue to increase the popularity of individual human creator-led content. And if you want to live in that world, you have to understand how to create great content. And AI for the most part can create good content, but it's still somewhat average. But it is a great content assistant. So whenever I'm like, what are some good ideas? I can get them from my head because I am always thinking through content ideas or I can take them from a great video, do a couple of first pass and see if there's anything worth building upon. So there's one in here that I thought was really good. I remember looking and I talked about it, which was this one here. So this one here was really interesting. I remember having that discussion and basically was the future of one of the biggest marketing skills that people learn in 2025 is to brief other teams through prototypes, not through like briefing docs. And I turn that into a post and that did pretty well. Like people enjoy that post, but that came from this idea, right? So what percentage of your LinkedIn posts roughly are created out of this kind of workflow where you have some video, video podcasts you've done or anything and you plug it into a tool like that. It gives you a lot of ideas. You pick one of them and you publish it as a LinkedIn post. What percentage? - Yeah, so that workflow would be, so that workflow is basically, just go through that workflow. It's like, I will see a good video. I'll watch a podcast. I'll say there's some pretty great things in there, some good ideas. I'll turn it into a list like this. Then I'll figure out like what are ones that I think I could create a unique piece of content on and I'll put them in a back catalog. And then I will at some point say, actually I'm gonna create something around that. I would say 30% is coming from that. - 30%. - Yeah, a lot of my LinkedIn content is not that thought through. It's like, I have a conversation. I have an insight and I immediately, I immediately have to write it there and then because the only way I've ever been able to stop thinking about something is to write it out. That's in life in general. - That's basically the entire reason why I'm asking you this question because I'm like that too. Oftentimes my LinkedIn posts, they come on a whim. I have this idea and like, I need to share it. I need to put it out there and I don't feel I have a shortage of ideas of what to share on LinkedIn. I'm sure you are the same. I have like a huge Google document with dozens and dozens of things that I want to share and not enough time to go and expand those ideas and put more meat into them, make them more meaningful. So it's almost like, but at the same time, I like the process that you're saying you might watch a video by someone else and then plug it into this thing and it helps you to pinpoint things because it's almost like makes you more productive because oftentimes when I'm listening to podcast interviews by other people, I'm like, is this a good use of my time? What am I getting out of this? So it's almost like you are making content while consuming content. - Yeah. - So you have consumed content, but you also created content out of it which kind of makes your time better spent. - Well, there's actually many uses of this. This is just like one use. It's really good for creating shorts because I give it to our production team and they use it now to like clip from our own video because it pulls out like the best clips, right? 'Cause it's built for the YouTube algorithm. I did a version of this with my brother 'cause he's doing really well on TikTok and it has this like map to TikTok, right? And so like it pulls out these clips. The other way is it's actually a great learning tool because what you're doing is creating a catalog of talking points and you can say, and it gives you the timestamps if you, I can show it again, but it has the actual timestamp. So I can look through any video and I can basically say, okay, like this is actually one of the things that you wanna listen to. I don't wanna listen to the whole thing but I'll just jump ahead to that timestamp. And the timestamps can be crafted for anything you want. Like if you say, the only thing I wanna learn from videos is these three things and then you can tell the AI, hey, look for these three things, it will give you the timestamp and you can say, I'm just going to go straight to those parts. - I used to use the notebook Alam because it has kind of native integration with YouTube. - Yeah, it's amazing. - But I think notebook Alam cannot give you an exact timestamp for some quote. I think it cannot do this. - Gemini3 has a deep integration with YouTube. So if you actually put the latest, they release, which is what I'm using. If you actually record this and you put it up and you say, how could cure and dress better for this podcast? It understands what I'm, it understands aesthetically, the background what I'm wearing can actually suggest that Kipp and I built a greater for our own show. And it pointed out a bunch of different things that we had to improve, [BLANK_AUDIO] like one of the things was really interesting. We've really dialed in for YouTube, but we have a big audience on RSS. And so what it said was, "Hey, when you're talking about things on YouTube, you're talking about something you're showing, but you're not articulating it for the RSS audience." It could actually see that. And so we've started to change to say, "Hey, like what we're showing is this, this, this, this." And so we can give you like real feedback. So the timestamps, I haven't, I've checked every single timestamp, is accurate? No, but I actually suspect they are, because it's pretty deeply intertwined with YouTube. - Let's go back to the topic of the system prompts that are powering your wipe code tool that pulls the different angles of snippets. So you have educational angle, you have a hot take, you have data point. So this kind of a broader question is about prompting engineering because they have interesting thoughts about this. But first I want to ask you, how sophisticated are the prompts there? So do you just say to AI, just pull me something educational, pull me a hot take? Or how much do you expand your instructions of what you consider to be something educational, what should be a hot take? - Yeah, how does it work? - Let me see actually, I'll show you a couple of things because prompting was really the thing that I had a tunnel average on. Like I spent a ton of time trying to learn to get better at prompting. I can actually show you just one of these prompts seen cut a little bit of the site, but just it gives you a, it's not gonna show you everything. Okay, so this is a public one. There's versions of this. I'm actually, I'm not trying to do a marketing thing and promote something, but I'm actually building a course version of how to build a team, an internal content team. So this is a V1, but I'm not, obviously publicly showing like the V2 or V3. But you at least get a, you least get an idea. I publish this for my sub stack audience. And so this is the prompt to extract those talk points that I showed you. And so you can see what it does here is it basically, you know, it gives you a task. You can actually, one of the things I didn't show you there is the version that I use is you can input your audience role. And so I have an audience profile and I will give it the audience profile which structures all of the talk points based upon who my audience is as well. But you can see here, I just say, well, if it's default, it's just a generic kind of default persona. Then it will go through here. It has categories, definitions and criteria. So this is the educational post, right? And this is like what it's looking for. So I give it, I give it enough for it to be able to go through that video and look for, so let's look at this one. Again, this is V1, I've iterated in this a lot. So it's like a lesson framework or high to step that teaches something actionable. Qualifies if it contains a teachable concept, framework, methodology or process, at least one concrete support in detail, right? So this is what I told you is like, I've always found when you pair the educational thing with an example or a statistic or something, it works really well, enough substance to derive a clear application. So you had, I had to play around with it a lot to just give me actually the meat of the thing. And then it structures it in a very certain way whereas a title, a hook, because again, this is like structured for a little different inside. The core inside, how you apply it, the value. And then it gives you the source. So it tells you who said that when, I was playing around with this, the emotional tag for the audience. So if you have the audience profile, how does it make the audience feel when they've gone through this? Now this is a little bit unaccionable right now, but I've been playing around with this because I would like to figure it out. So this is just the prompt to extract those three talk and points. Now I have one that extracts eight talk and points. And to give you an example here, this is another one of my AI assistant seats, but they all kind of look like this. If you look down the left hand side, look at these. And I have like lots of variations of these, but you can see I've built prompt engineers. So I have a prompt engineer for different, all the different models. This here prompt engineer is pretty interesting actually. So cursors prompt got leaked. And what I had, what I did was had worked with AI to like figure out the core things that cursor have done well in prompting and then built an engineer that can prompt like cursor in terms of how they constructed a link. Very similar here, clouds got leaked, but you can see I have like different prompt engineers all the way down here. - So let me ask you something interesting about those prompt engineer. So yeah, I heard on another interview, a podcast interview that you've done that a lot of models they actually release their prompting guides. - Yes. - And what you did, you kind of uploaded those prompting guides in your custom GPs, agents, whatever. And basically they are referencing the best practices of prompting to generate a prompt for you. But an interesting thing that I've heard on the podcast interview with the founder of perplexity, I think it was on Lex Friedman or somewhere. And he was saying that what they want to do is they want to kind of even out the playing field, democratize the outputs that people are getting from perplexity and from LLMs. So that like if you who is experienced at prompting would ask LLMs something, and I would ask LLMs something, but in very simple terms, LLMs should make the results that me and you get more equal. Because otherwise you would think, oh, this LLM is amazing. And I would think this LLM can do nothing. It's terrible. So what he was saying that their perplexity in the background, they were refining your prompt. So they were taking your simple prompt and they were actually using their own prompting guide internal to improve on your prompt. So why do you need those custom prompt engineers if LLMs are actually interested to refine your prompt and prove it and give you a better output? - Yeah, that's a very recent thing. And so like if actually there's like a couple of developments in that. So if you use the latest chat, GPT model, remember we had all of these different models we could select from. And chat to the five is actually figuring out your intent and then trying to match it with the right model in the background. So if it goes, hey, you want a deep research, hey, you want creative, hey, you want something else. Now, it's not actually doing a great job of that. If you look at all the external feedback on chat to GPT, a lot of people are saying they are not getting the model they actually want. I think intent is a really hard thing to infer. We try to do it a little bit in HubSpot when we, you're talking to our AI assistant and we're trying to figure out as a sales conversation or a support conversation, that's just two things. And that's really hard to infer between those two different things. Chat GPT is 800 million users, asking a range of different questions. So they have to try to figure out what model to assign your question to. And that is a very similar thing it's trying to do is like give you the best experience for the least amount of thought on your side. On your side. The other thing you'll remember at some point, the model, so they aren't doing it as much, but they brought in this thing where they would ask you additional questions. So before I do this, here's like one, two or three questions. The reason they're trying to do that is they're trying to pattern match in the back end, like what do you really want? So that is an example of them trying to improve the prompt. I will say the latest models, I agree that actually over time, it does not, like the, the, just asking AI to do something, like Gemini 3, Opus 4.5, just asking it basically to do something and me trying to craft this big complex prompt, the actual margin of difference in value you get back is starting to decrease. So a lot of these prompt engineers have been built over the last 18 months. I have not actually built one for the latest Opus 4.5 model because I just haven't found, it's just so good. Gemini 3, kind of similar, I haven't only built one for that. Do I still think there's value in prompt engineer? I do, but probably like really at the upper echelon of like what folks are trying to do. Just going by like AI researchers, I follow and listen to, I think what he's talking about is the average consumer does not want to go to perplexly and try to figure out. Yeah. You're a season CMO and here's your task and here's your output. They just want to go like blah, blah, blah. And that's going to improve their usage. And I do believe for the average user, prompt engineer is not going to be a thing because the AI models again, so much better at trying to figure out what you actually want. Here's another interesting, interesting thought I wanted to run by you because you're advocating a lot using AI kind of as a feedback mechanism. So you would share your, I don't know, thought strategies and you would tell it, ask it to critique it. Yeah. So what I did because there's the saying, if you ask for feedback, you would always get it. So what I did, I found, I searched for a sonnet by Shakespeare. I pasted it into chat GPT and I said, give me criticism, what can be improved. And of course, it told me how can I improve a sonnet by Shakespeare, which kind of is genius in itself because he's universally considered to be a genius. So how do you not fall into trap of getting just never-ending feedback? And like, it seems like a perfectionist trap. If you ask chat GPT to kind of find holes in your thinking, find holes in your strategy, there would always be holes. It would always suggest something. So how do you know where to stop, when to stop? Yeah, I think I actually think, so AI is rewarded for given answers. And so that's one of the problems with AI is like part of how they train it as you only get rewarded if you give answers. Whereas I think they've now realized that part of what they want AI to do is say, I don't know, right? Because if you want AI to become the dominant search, even like the dominant search player, I'm interested to get your perspective in that. But if you want, If you really want AI to replace the blue, the template links, you will have to have some additional, you will have to have AI say I don't actually know give me some additional information and maybe I can actually get you the right answer and they are actually starting to integrate that and how they train AI to reward it for for that kind of feedback. I think in terms of that, you cannot outsource like how do you know in and off so if I examples would be I have different AI assistants that are trained on thinking methodologies and so I have one that's like will red team my idea red team as they will take your solution and take the inverse of that and argue back at you. I will I have another one that does first principles thinking like and so whenever I want to just like think differently about a problem I'll put it to the AI or the solution I'll give it to the AI and say like run that thinking methodology but could I do that forever and could it continually give me feedback on that yeah and so the thing I advocate for is you can't AI again I think is most beneficial for people who actually have real domain expertise and can do critical thinking and if you don't have those things there's a great graphic about vibe code and I'm going to do that. I think about vibe code and it's really like vibe code and is no different than people being in the casino playing the slot machines and that you're just like one more pull of the thing and it will definitely fix this bug and like who wins the house always wings which in this case is cursor or a cloud code but you have all these vibe coders like pull them the thing and trying to fix their bugs. Not that different from like the average person who doesn't really know strategy and doesn't really have domain expertise saying oh tell me something more tell me something more and it's just like this vicious cycle and I do think we're going to over index into AI so I think internally and tech companies people are probably shipping more memos and they've ever shipped because everyone's like oh it's have AI do that you can't yeah like I don't know how this I don't know how any of this pans out but there was a great research study last year that really maybe sit up and take notice because I use AI a lot that people who used AI. We're real power users and we're kind of outsourcing a lot of it became stupider over time and and so I still believe that I am the I am the critical thinker I am the person with domain expertise and AI is assistant me and I know when to use AI and when not to use AI and I don't use AI for like I thought I had to think through some really hard problems yesterday I don't think I used AI for any of them because I was pretty comfortable doing it myself. Yeah that's that's a very good point let's let's go back to discussing our podcasts I know that not too many people in the marketing world around their own podcast but we can use it kind of as a metaphor for creating content. Yeah because a lot of the things check out you do you do research you create a piece of content you promote a piece of content so whether it is a podcast or video or an article or webinar or something else so I'm wondering which parts of your podcast when you are working on an episode are automated or assisted with AI like where does a I sit in in the process of starting from an idea we want to record the podcast about this. And up to published and promoted. Okay I'm going to give you like a really honest answer here because this is like this is like the you know the market the agency who does really great work for their clients and then you say, wait what do you do for your own marketing like nothing. I don't think we actually talked about this this week we have not really used AI enough on our own podcast and because this is not kept up to the time job is not podcasters. I think the podcast has suffered recently because we have been so busy and so a lot of our episodes are Kipp and I we watch up each other every day and like hey have you seen this I have a good idea and you actually literally just turn up and try to do it and that's actually not what we want to do anymore. So we actually have started to think about how we can instrument AI better across our own podcast so I think there's a couple of ways the research is a big way. Research is actually the best way so we want to we we've kind of done a lot of the eight we've done like AI use cases for the last year Kipp and I naturally like to just to kind of like riff on things and tell stories but because everyone's wanted the kind of prompts in the use cases that's really what we've done over the last 12 months we want to get back to kind of talking about marketing as a whole stories as a whole riff on ideas and so what we have what we were setting up is like many assistants like I showed you one that can kind of we tell it like here's something we're thinking through is there any like good examples of stories that really fit into that you know we fit into that and let me I can give you a real example I posted a LinkedIn post this week that I'm going to do a podcast episode on it's come it came from an episode that I'm working on this week where I was trying to talk about a coffee I had with a CMO and that CMO was telling me hey like I spent 15 years learning all these things and I have these new marketers on my team and they ask AI for the playbook and the playbook is like what I would have wrote I for them it's like really good. What is what why am I even useful and what makes good marketers marketing and I thought that was a good episode I thought like marketers to win tend to be the ones who can know the formula and follow the formula and actually AI is putting on a force us to like figure out what is the new things we can do that is different from everything else and I was going through that and I asked my assistant AI assistant the story one any good stories that can really help me bring this to life and they told me the story of a high jumper called dick foesbury and dick foesbury was this high jumper back in 1968 before he came on he you know he came into the sport I didn't really know this but like people with Somersault over the bar and they Somersaulted over the bar because they had to land on the feet because the grind was hard and he was like you know what I'm going to do it differently I'm actually going to jump over and land on my back and he was able to do that because their school was the first school to have foam mattresses so foam mattresses was the technology for him that allowed him to this crazy new technique and then he won Olympic gold and within 10 years everyone had switched to that technique and I was making the case that kind of similar to marketing AI technology we have to be free from practice practice and take risks and so with that example AI found me three I would not have the time to go find three great stories but it found me three stories and I was I had the domain expertise to know well I could take that one and craft it into this and it really brings that message to life right. So that's one way is research to like figure out what are the what are the what are the when I think about content I think about it as like a modular thing so let's say you have an hour and I have podcasts but there's three like modules that you always want to have in that maybe it's a application maybe it's a story maybe it's an analogy. And so you can have a research assistant that helps you bring them into the overarching content. Yeah. Can I can I add to your to your story about CMO asking why why would my team need me if they can ask AI I asked the same question to Amy Lee Kramer. We recorded an episode a few months ago because she's a startup advisor and I asked her how are you going to have your job like a few years from now if like any CEO of whatever why they need to start a advisor if they can ask AI and they I would probably do a very similar very similar thing. So she had her own answer but I had my own my own answer to that prepared which I will share from my perspective it's two things it's first of all responsibility AI doesn't care what happens to the advice it gives you it has no responsibility. It just it just keeps giving you advice like we like we discussed. And the second thing is initiative like a good advisor they would come in and yeah they would answer questions that CEO or other people on the team have for them. But they can also look into things themselves and say yeah you didn't ask for it. But I think that this is the problem you're having that needs solving and I don't think that I yeah you can also argue that you can you can tell a I look at my website look at my CRM go through this. And give me a lot of advice but then again you'll be just overloaded with the amount of advice. Yeah and there's there. So yeah this is more or less my answer. Yeah there's there's another one which is like you you you you have deep deep domain expertise and you probably do things and do things for the team and give feedback to the team and answer questions for the team and think about problems in a certain way that you actually don't know why you do what you do. It's called like unconscious I can't remember like it's like you you consciously and you yeah and so the unconscious part is really hard to get because you may have because I suspect you've never really wrote it down like you wrote down like the formula is that you know but there's something deeply ingrained in you that makes you you it makes you really good at what you do. And you haven't actually wrote that you haven't actually wrote that down and AI is struggling with that is called there's a you'll hear this all this year this will be the new AI agent. You know talking point which is context graph and context graph is why does a person do what they do and the and we've never I don't write down why I do what I do I just do what I do and AI can't understand that and so that's why you see a lot of these agent that companies AI agent companies fail is because they don't actually have the why of why something is being done it has like the outcome that you're trying to get towards but it doesn't have old and you want and the why of why you may certain decisions why you do certain tasks and so there's something deeply for an advisor they just have nuance and ways of working that are not going to be documented anywhere else and I still think that is a really valuable thing but AI is good at doing is like hey what is the best way for me to run email sequence for a company of the size and give me some good example. and it will just collect everything that's online and give you a pretty good playbook. It will give you a good playbook. It will give you an elite playbook, not really, but I'll give you a quick tip if you ever want to see what it would do if you give it an elite version. Give it some examples of actual people. So if I say, "What's the best AEO strategy that I could possibly run from my company?" I would give your name and several other names, and it would ring fence to use, not the average. And I don't think people do that either. Yeah. Let's discuss using AI for team management. Because like you said, you have a very large team working with you. So what are some cool, I don't know, use cases, workflows, prompts, maybe tools that you're using to help you be a better manager? Yeah, I'll see if I can give you the. So this is the hardest one for me to give examples of, obviously, because it's all in HubSpot and it's all a kind of proprietary information. But I wanted to show you just because I think people, we get blindsided by the influencers, the LinkedIn influencers, the AI influencers, and we're looking for the most shiny thing, where a lot of the value of AI is the most mundane thing. And so what is one of the hardest things that I find hard to keep in top off? It's not just I have this, that's like a direct team. Then I have like two very large cross-functional paths. And so I have a very large breath of things that I'm trying to stay on top of. And so I have an an order in the amount of like, "coms coming at me," right? And I can't make, I can't make, I can't keep on top of it. I also don't think I'm a very good process orientated person. I like to just do work. If I could never do that stuff, I would be very happy. So like this is like a really simple one. Now there is like a little set up, so you have to have AI connected to your Slack, follow Gmail, Google Docs. So I'm doing this via Glean. You can do it via a cloud. There's a lot of tools. Opening AI will connect to these things. But this is like really simple. So I have an agent. This is just like, I kind of pulled out this because I can actually show this. This is a real piece of feedback that could give me. But I can't show everything. But what's it doing? So what it's doing is if I say, give me my, summarize my week, summarize my day, summarize the last two weeks, what it will do is it will go find topics. First of all, go and find different topics across these channels that are coming at me. And then it clusters all of the comms per singular topics under that topic. So what I mean by that is, this is like one topic. This is our quarterly business review in 2026 alignment. It goes and finds, it's gone and found all of the different comms that I've had on that topic. So you can see it's gone on my Slack channels. It's gone on my fellow, which is our meeting note taker. It's gone in DMs, my Slack DMs. It's gone on challenge DMs. It can go into Gmail as well. And then it takes the key points. What are my actual actions? So this is what I do every, start of every week, what happened last week and one of my actions for that week. And then who needs response? This one I really need help on because there's people tagging me, asking me things in Gmail, in Slack, people have asked me to follow up on things in meetings. So this gives me everything I need. And then I just like implement that into like, what do I need to do that week? So that's, that's like number one. Okay. So the other one would be AI can be a great chief of staff. And so if you have a certain go to, if you have a certain operator model you use, I wrote a post about this, but I have a specific operator model I use and the operator model is built to increase momentum, like speed of execution and drive accountability. And it has this kind of set thing it does where we have these kind of roadmaps, these portally roadmaps, we have a biweekly, a biweekly sync that we do that basically goes through. What did you ship over the past two weeks? What do you plan to ship in the next two weeks? What are any kind of blockers you have? We have a scorecard that says every month that we hit the targets we hit, we said we were going to hit or not. And then we have this kind of retro perspective we do where we kind of internalize learnings. Now, what's really incredible is if you, if you have all of your roadmaps in one dock, all of your biweekly sprints in another dock and all of your kind of accountability metrics in another dock, you can just give AI this and it can start to actually build that some pretty cool things for you. And so let me just see, we have this thing that everyone loves as well, which is a quickly touch on this because every, every company I talk to when I'm going through this, love it. It's a bullpen. And so what we do is I believe in companies that we've kind of veered towards making decisions by consensus and we've lost the ability to have healthy to bait may not be the same in your company, but in a lot of technology companies, that's the way it is. And so bullpen is basically anytime there's an alignment problem or a big problem to solve. We stand up a bullpen. That person who stands up the bullpen, they provide three solutions and that solution, each solution has a pro and con. And then we just get the people in the room who can make the decision on that thing to debate it. And then we ship a decision there and then so we can move on. And so that's part of my operator model. But what I'm getting to here is, so I give AI all of these things and I can actually ask it for some pretty cool things. So I can say, let's see what's this. Let me go past this because I have lots of different things here. Okay. So basically like what across all of these different teams like that are supplying these updates, what are the current blockers? And so I can ask it like, what are what are the things where we have blockers, but don't have mitigation plan? What are the things where we're trending below our targets? And it can basically give me back this right? So it can basically say, Hey, like this is the current list of blockers. It will say what is the work stream? What is the blocker? Do they have a mitigation plan? What is the target date? The plan to actually have that mitigation plan done? Is there any, is this over? Based upon where we when we said we would have this done? And then what support do they need? This here KPI gap to target is basically going to tell me which work streams are trending behind where we want them to be by the end of the month. So again, it gives me the work stream. And then it gives me where they were last month and then tells me like, what is the gap to target and do we have a plan to fix it? So this risk assessment is do we believe we'll fix it or not? And the AI is trained to figure out like, are you actually going to miss this? This is going to be read or are you actually going to hit it? This is going to be green. So it's like, if I just give it all of the data, I can start to like ask it any questions I want about how all of these teams are actually running. And I find that I find that like super valuable because that those teams are providing all those updates that's context and my chief of staff can give me any look that I want. And I can I know we're going to talk a little bit about data. I can show you what that looks like and some of these interactive data maps. This is this is amazing. Even even the simple use case of using AI to go go through all your Slack messages of a week and kind of categorize that for you and show you okay, you've had progress here. You helped push projects here because a lot of the time as a manager, I feel unproductive because I spent most of my day on Slack answering questions, helping people and blocking people. And then you look, what did they do all day? Yeah, what did I ship? I helped everyone, but it doesn't feel productive. This is why you want to go like and publish something, create a podcast, do something on LinkedIn. But if AI agent would provide for you, okay, so you have made this many decisions this week, you help like brainstorm new features, you created five new ideas for the blog team, blah, blah, blah. And suddenly you feel productive. The only the only question I have for you is that giving AI access to your company Slack, a lot of people are afraid that AI would be training on their proprietary data. And your hub spot, your Slack is like a treasure trove of a lot of proprietary information. And you give different AI models access to your Slack? Gleene, and I definitely can't speak like we have a very big legal procurement team that brought in here. And so they've gone through all of the, yeah, we're we do not give data easily. And so it's pretty, I trusted if we're using Gleene, it's gone through a lot of like hurdles and loops to be able to use that data. Cloud has like all the enterprise seats like cloud does not train on internal data. Opening, I just not train on internal data for their enterprise licenses. And they also, I know cloud has a connection to Slack that we're starting to, I'm starting to look around. So all these tools will have enterprise versions of what I've just showed. Like you should be able to do it with most of whatever seat you're using. And if you're using an enterprise seat, obviously an enterprise seat, they do not train on your internal data. Okay, we're already hour in. So let's do a quick rapid fire round. I have just three questions. First one, do you have any hot take as of today? About AI and marketing. About anything. What's your favorite hot take? What would people disagree with you about? What would they disagree? What would they, that's okay. So I had not thought what would they disagree with me about? Well, isn't this a hot take? Isn't this a definition of hot take? Yeah, no, I'm trying to, I'm trying to think through what is a, what is what is a good one? I think one of my hot takes is what I, what I've heard is what marketers are really worried about is AI is going to like replace them and do all of the creative work. I think AI actually makes marketing more important and more, and more of a creative discipline than it's ever been because what marketing, like marketing, if you actually, like one of the things I really love actually is, reading old books around copyright and like copyright and is one of my favorite topics and really I liked the 1930s 1940s 1920s and back then it was like literally pretty simplistic. It was like here's my audience and here's like a small message I can craft for them. So it really was like true marketing in terms of do I know the audience and can I craft something in a simplistic way to you know engage with that audience and the proliferation of the internet was really cool for marketers because it gives all of these new channels right we were able to dominate search and paid and it gives us ways that we could be more important to the company I think more measurable and actually and actually show the value of our work but one of the problems is we've kind of lost the creative flare in some ways because we look at the data and we say well we should just keep optimizing these things because we can incrementally improve and that's where the data tells us and I think a couple of things will happen I think AI for and and and because of that marketers have become slaves to their tools right like 80% of a marketer's job and I was like using all these different tools and then 10% is admin and 10% might be actually creating something and I think one of the things Kipp and I talk talked a lot about is like AI is their nascent for marketers in that it will allow marketers to offload all of that tool work and admin work to AI and the mundane things but you can keep the creative things like I created my first kind of video that would view three dot one last week and you know the idea still really matters like the video was good not great and the tool did all the work but like it gives me so much more time to like craft the actual video like really think about the idea so I think AI is a creative bazooka for creativity and marketing and I think that's going to be a really good thing and I think AI probably makes marketing more important and that's another thing I would say is go all the way back to the start of our conversation there's two things I think will happen. I think AI are the automators in most company. And so they can actually start to integrate AI across the entirety of the buyer journey and what you'll find is like a lot more of that will come into marketing and when everything every single vertical is going to have more software than ever because it's so much easier to build and so what's going to stand out I think it's companies with the best marketing and companies with the best position and companies that people trust and have authenticity and so I think that makes it more valuable. I think I need to rethink what I qualify as a rapid fire question because that question didn't have a simple answer. Okay another one. How did you hire your best or one of the best team members because the question of hiring is always interesting. Side projects. I'll be really quick. Side projects. I love people who do side projects. Have side quests have like hustle like side businesses. So you saw someone with a cool side project and you like. Yeah like do they do they're doing they're doing their work and doing other things I've always found those people are like really entrepreneurial curious learn you things very very rapidly. Awesome. And one final any cool hack or tactic that worked super well for you let's say in 2025 but it's not something we discussed on this podcast so far. You know the one that comes to mind but I don't think it I think it's very specific to me is like we had a we had a executive offside in San Francisco last year. And I was really trying to figure out what are some interesting ideas I can bring to that and one of the cool things I did was I give AI all of our technical documentation and all of our product documentation and basically created an assistant that acted like a CTO and one like a CPO. And I basically said hey like disrupt this company and build a business that has better functionality and features that is different cheated and it came up with a tool that I took and built upon and we prototyped and not a launch. I think that's pretty cool. You haven't launched it yet. No it's launching in April. You probably can say what it is because these these would probably go out. It's a two called hot machines. Yeah so it will turn your it will bring AI so like the majority of small businesses are using Google Sheets as their CRM. It will bring agents cleanly bring agents into your Google Sheets and allow you to easily use them so think about like I have some contacts in there and I want to create a email sequence. You'll have an agent in your sheets and it'll be able to like do that for you you'll have an agent that will be able to do a bunch of different things for you in right within your sheets you never have to leave there. It reminds me what clay is doing. Yeah it's a little bit like yeah like bring it bring in the bring in the agent experience into where you are. Kiran this is without a doubt my best podcast episode about using a yeah in marketing very actionable very specific very little fluff almost no fluff so thank you a lot for taking the time and thank you for being so specific and sharing so much. Yeah no I'm a big fan of yours and the company and I appreciate you asking me on.