The GTM Podcast is available on any major directory, including:
Kieran Flanagan is the SVP of Marketing at HubSpot and Co-Host of Marketing Against the Grain. A longtime operator and investor, he’s at the forefront of how AI is reshaping go-to-market. With a background in engineering and years leading growth and marketing teams, Kieran now spends his time building, experimenting, and sharing lessons on how prompting, agents, and personality-led growth will define the next era of software companies.
Discussed in this episode
- Why prompting and context engineering are the most important skills for GTM operators
- How “vibe prompting” accelerates learning and output with LLMs
- The three keys to building AI fluency inside teams
- Measuring ROI from AI across sales, marketing, and operations
- Why every professional is now a manager (of AI agents)
- How websites will evolve into multimodal closing mechanisms
- The rise of personality-led growth in B2B marketing
- Why curiosity and persistence matter more than ever in an AI-first world
Episode highlights
00:46 — The 100x difference between good and bad prompting
Watch: https://www.youtube.com/watch?v=FDoT3ZVhwhk&t=46
03:57 — The rise of “context engineering” as a GTM skill
Watch: https://www.youtube.com/watch?v=FDoT3ZVhwhk&t=237
07:22 — Kieran’s 3-part framework for AI fluency inside teams
Watch: https://www.youtube.com/watch?v=FDoT3ZVhwhk&t=442
09:31 — Why “vibe prompting” is as powerful as vibe coding
Watch: https://www.youtube.com/watch?v=FDoT3ZVhwhk&t=571
11:00 — How AI boosts conversions & deal velocity in sales workflows
Watch: https://www.youtube.com/watch?v=FDoT3ZVhwhk&t=660
15:10 — Using ChatGPT memory as a personalized prompting coach
Watch: https://www.youtube.com/watch?v=FDoT3ZVhwhk&t=910
22:19 — Everyone now manages a PhD-level AI intern
Watch: https://www.youtube.com/watch?v=FDoT3ZVhwhk&t=1339
31:12 — The 3 biggest shifts coming to GTM: influence, AI optimization, multimodal
Watch: https://www.youtube.com/watch?v=FDoT3ZVhwhk&t=1872
37:42 — Why AI makes human creativity more valuable than ever
Watch: https://www.youtube.com/watch?v=FDoT3ZVhwhk&t=2262
43:06 — The grind, reps, and curiosity as the ultimate AI skills
Watch: https://www.youtube.com/watch?v=FDoT3ZVhwhk&t=2586
Key takeaways
1. Prompting is a must-have skill
Prompting and context engineering will define the next generation of knowledge work. Those who master them can outpace peers by 100x in output quality.
2. AI fluency scales through culture, not training
Prompt guides, Slack channels for sharing use cases, and AI hackathons help teams adopt faster than formal courses.
3. Measure AI by outcomes, not activity
ROI is clearer in conversion rates and deal velocity than in productivity metrics, which remain hard to quantify.
4. Everyone is now a manager
With AI assistants as “PhD-level interns,” every knowledge worker needs to learn delegation, training, and feedback loops.
5. Prompting is iterative, not instant
Great prompts require repetition and refinement (sometimes over days) to reach production-level quality.
6. Personality will beat keywords
In B2B, personality-led growth (podcasts, newsletters, YouTube) is replacing keyword-optimized blogs as the primary driver of influence.
7. AI engine optimization is the new SEO
95% of B2B journeys will soon start in LLMs, making visibility inside AI assistants more important than Google rankings.
8. Websites are shifting to bottom-of-funnel
As research happens in AI, company websites will shrink into multimodal closing tools with sales agents built in.
9. Creativity matters more, not less
AI is a powerful assistant, but human creativity is still the differentiator in breaking through noise.
10. Curiosity is the ultimate moat
The fastest learners (those who experiment relentlessly with AI) will be the new winners in GTM.
Brought to you by: Mutiny
Are you a B2B Marketer running campaigns for target accounts? Then you know the struggle: tedious and manual processes, endless delays to get things live, and sales feeling like you’re not doing enough.
That’s where Mutiny comes in. Mutiny is the fastest place to launch breakthrough campaigns for your target accounts.
AI agents research your accounts, build personalized landing pages, and scale everything from LinkedIn ads to sales handoffs, all in one seamless workflow. No more stitching tools together. Just smarter, more impactful campaigns powered by real data. Launch in days, not weeks.
See why teams who use Mutiny generate 3X more account engagement.
Book a demo at mutinyhq.com
Recommendations
- Newsletter (The AI Marketing Generalist):https://www.kieranflanagan.io/
- Podcast (Marketing Against the Grain):https://www.youtube.com/@MATGpod/videos
- Article (Vibe Prompting: Turn AI Into Your Personal Prompt Engineer): https://www.kieranflanagan.io/p/vibe-prompting-turn-ai-into-your
Referenced
- Mutiny: https://www.mutinyhq.com/
- ChatGPT:https://chat.openai.com/
- Claude: https://claude.ai/
- Gemini: https://deepmind.google/
- Perplexity: https://www.perplexity.ai/
- HubSpot:https://www.hubspot.com/
- Zapier: https://zapier.com/
Guest links
- LinkedIn: https://www.linkedin.com/in/kieranjflanagan/
- Newsletter (The AI Marketing Generalist):https://www.kieranflanagan.io/
- Podcast (Marketing Against the Grain):https://www.youtube.com/@MATGpod/videos
Host links
- LinkedIn: https://www.linkedin.com/in/sophiebuonassisi/
- X (Twitter): https://x.com/sophiebuona
- Newsletter: https://thegtmnewsletter.substack.com/
- Website: https://gtmnow.com
Where to find GTMnow
- Website: https://gtmnow.com/
- LinkedIn: https://www.linkedin.com/company/gtmnow/
- Twitter/X: https://x.com/GTMnow_
- YouTube: https://www.youtube.com/@GTM_now
- Podcast hub: https://gtmnow.com/tag/podcast/
GTM 165 Episode Transcript
Kieran Flanagan: 0:00
Everyone really has a PhD level intern that they can work with for free. Prompton is a think or swim skill. If I had to tell you that promptin is a skill to learn, I don’t think you’re the right person to be in the company. I’ll give you three great tips.
Sophie Buonassisi: 0:38
Why do you think prompt engineering, like the ability to ask the right questions and guide these models, will be such a defining skill for the next few years?
Kieran Flanagan: 0:46
There is like a 100x difference in your output if you really knew how to prompt the engines in the correct way.
Sophie Buonassisi:0:54
Two years ago, Jiren Flanagan made himself a commitment.
Kieran Flanagan: 0:57
There’s just no one that will know more than AI about me if you’re a go-to-market practitioner because no one is going to work as much as me in AI.
Sophie Buonassisi: 1:04
Since then, he’s treated prompting like a craft, spending days perfecting single prompt. That obsession led him to an even bigger insight. Prompt engineering, how you act, and context engineering, what you feed the model, are quickly becoming four skills that every single person will need. In this episode, we break down his personal playbook for up leveling your own AI skills, from vibe prompting to simple habits. And stay to the end for an unexpected creative talent he’s been quietly sharpening with and without AI talk. His first reaction when it came up.
Kieran Flanagan:1:34
I can’t believe Brian said it was of all the things Brian could have said about me. All right, let’s get into it.
Sophie Buonassisi: 1:39
Kieran, welcome to the podcast.
Kieran Flanagan: 1:41
Yeah, thanks for having me on. I’m excited to uh to be on the podcast.
Sophie Buonassisi: 1:45
Yeah, we’re excited to have you here. And there are so many different areas that I want to dig in today, so we’ll jump right in. But a big one is something that if anyone, which probably everyone here, does follow you on LinkedIn, they’ve heard you talk about. And it’s really that everyone is talking about using AI. But you said that, you know, the unlock is really how we work with it. Why do you think prompt engineering, like the ability to ask the right questions and guide these models, will be such a defining skill for companies in the next few years?
Kieran Flanagan:2 :15
Yeah, I think prompt engineering is still like the skill to learn. I remember when I first started to really obsess about AI, uh, really when ChatGBD came out around three years ago, and I talked to this really well-known CPO of the Fortune 500 company who was building all of their AI technology. And he was talking to me about how he had hired engineers who’d become better prompt engineers than the ones they had in Open AI. Whether that was true or not was like beside the point because what he was really telling me was there is like a hundred X difference in your output if you really knew how to prompt the engines in the correct way. And that was three years ago. We have always kind of thought about prompt engineering, which is the ability to kind of like ask the AIs in the right way for the task you want to complete and the outcome you want to get. And we’ve always, I’ve always thought like eventually do you not need that skill, right? Because the LLMs become so smart that they can just do the prompts themselves. And we can get into that. I think that is not true today, but they are certainly great assistance. And but for me, I I that’s stuck in my brain, which is wow, like if I really learn this skill, I’m gonna be so much, much better than everyone else. And I still think that’s how I feel about prompt engineering today, which it is a like critical skill if you are a technology worker to be able to understand how to work with these models. Now, the thing I would layer on top of that is there’s prompt engineering and there’s also context. And I think context is a really important thing, which is how do I provide the right context to the model so it knows enough about what I’m trying to do to give me a really great output. So I think that context, which is becoming its own discipline, context engineering, context plus prompts really give you the skill set you need to be an incredible modern day knowledge worker.
Sophie Buonassisi: 3:57
Super interesting. I haven’t heard a lot of people talk at length about the context side. How would you say those are different skills from each other? Like if any go-to-market operator, founder, investor is looking to upscale in both of those areas ideally, how do they differ from each other and how should people be approaching them?
Kieran Flanagan: 4:17
Yeah, I think prompt engineering is how I craft the ask to the LLM. Now we could say, well, part of the prompt is to give the model context. So giving the model context is giving it the right amount of information to be able to complete the task at the level I need. And I think all of these models get much, much better when you can give it the right, right context. And that goes for all of AI it we’re deploying across across our go-to market. Like if you can actually ingest the right context for the model so it really understands what you are trying to do and it really understands what good looks like, it can actually produce much, much better results. But understanding how to give it the context is a skill to learn. Like a really simplistic one for kind of most marketers is, you know, you you when I spent time with marketers about a year ago, I was really into is AI a good writer? And can AI replicate someone’s writing style? And can AI produce something that you could just copy and paste and no one would know? And I spent time with a lot of marketers because I built a tool at the time to kind of like see if that was true and look to see how they prompted the AI to create content because they were all telling me it’s rubbish, it’s bad. And you would go to them and you would look to see what they were doing, and they would say, write me a LinkedIn post about how to do lead generation. That’s it, that’s their entire prompt. And there’s some fundamental things wrong with that prompt. So, first of all, you’ve given it, you haven’t given it context on what a good post looks like and why that good post is a good post. Like you do some analysis, you say, here’s the context, here’s a post that did really well, here’s the context of why it did well, and this is the kind of thing I wanted to replicate to. The other interesting thing here, which is like a really interesting little tidbit about prompting, is in that prompt, because I say create me a quote unquote LinkedIn post, because the LLM has a training set with lots of LinkedIn content, and LinkedIn content is generally not good, it’s gonna create something not that good. And so you should not say LinkedIn, you should just say like a great post, right? Don’t give it the platform because it will try to skew towards that platform. But basically, if they had crafted a prompt to give it a context of like a post that performed really well, some context on why that post performed really well, and then the other thing, just kind of make that prompt a little longer to say, here are some writers that I admire and give the names and are really good, mimic their style and it would go off and like replicate that style. So there’s like ways that you can make it much, much better. But a lot of people do lazy prompt them. They just say, like, create me a post about this thing and 10 points, right? And that’s why you you put garbage into it, you get like really bad results back, and that’s what’s happening.
Sophie Buonassisi: 6:51
Yeah, the the context almost matters the most now. And I’ve heard you talk about AI fluency at length and how it’s gonna be the skill that I mean is is absolutely central for every single go-to-market person and founder. So when we think about prompting, are there frameworks or mental models that you know you coach people on to use when designing these prompts? Like you just gave a really, really great tactical one. We even take a step back. Like, how do people begin?
Kieran Flanagan: 7:22
Yeah, so I I’ve had a lot of good conversation with founders who are trying to make their team much more AI fluent. And there’s kind of I I always tell them do not over-engineer this. But plus, I’m definitely on the more cutthroat side. So I want to like the the let maybe the less empathetic side. So I basically tell them look, there’s three things that are gonna matter to make your team much more fluent in AI. First one, prompting is a skill they need to learn. Second one, inspiration is a big part of how people accelerate and expand their usage, which means if you have a shared Slack channel, you have people who put the things that they’re doing that are working on that Slack channel, and there’s just a stream of things that people are doing. That’s actually one of the number one ways that people learn within companies how to use this for their own discipline because they can see how other people are using it. And so that’s that inspiration part is really, really big. And then the third one is like mini hackathons. So hackathons used to be a developer thing we did, right? When we were for specifically for developers and engineers, I was an ex-engineer and developer, was never very good at hackathons because it was a really bad coder, but can I have a vibe code, which is really good. But hackathons, now we can all do them. Like go to market teams can do them. Marketing can build AI workflows, sales can build AI workflows. And so these hackathons where people get together and build things in AI is another great learning technique. But to come all the way back to the first thing in terms of prompting, I think prompting is a sync or swim skill, which means if I am a person, if I’m your manager, your founder, if I have to tell you that prompting is a skill to learn, I don’t think you’re the right person to be in the company because there is enough information to tell you that AI is important. Prompton is how you work with these machines. And every there is so much information out there about them. So if you go to OpenAI, they have these cookbooks that tell you how to do prompt and per model. If you go to Anthropic, they have prompting guides. If you go to Gemini, they have prompting guides. There’s no excuses, but I can give some tips for people to really accelerate their prompting skills in a quote unquote vibe way, right? Like, which is basically the LLMs are going to do a lot of this for you. And that’s the shortcut. I don’t think people realize that the LLMs, if you work with them correctly, can actually accelerate your ability to prompt pretty, pretty rapidly.
Sophie Buonassisi: 9:31
I don’t know if it’s out there already, but I think you just coined vibe prompting, Kieran.
Kieran Flanagan: 9:36
I know. And it’s actually what’s interesting is right, we have ViMarketing, Vibe, Vibe Marketing, Vibe Coding. We have anyone who hangs around a lot on LinkedIn, I’m on LinkedIn, LinkedIn a lot. We have Vibe Cometon, which is basically people are putting their LinkedIn comments and autopilot and having AI do them. The worst use of AI ever. I don’t know why people think that’s a good use of AI. But VibeProm is a good idea.
Sophie Buonassisi: 9:55
I think you put a post out on that yesterday.
Kieran Flanagan: 9:58
Yeah. Like Vibe Prompton is like as powerful as Vibecodin. I think it’s like a really incredible way to use LLMs.
Sophie Buonassisi: 10:07
Mm-hmm. It’s really cool because you have an engineering background and you know, now you’re essentially doing the same, same kind of hackathon setups and so forth, but in a go-to-market context. And that’s a really valuable skill that lends itself to every go-to-market team. I know we ourselves we had a little hackathon where we outlined essentially every single workflow that AI would be monumental or incremental in. And then we rated each, ranked each, and now we go through workflows where we build out AI and then for each of those and then progressively, you know, progress them on a maturity curve. But the importance is doing it regularly. So Yeah. How do you think that’s a great time? How do you think about ROI? Obviously, we all know it’s incredibly valuable, but do you think about it as time saved, lead quality, conversion lift, something else?
Kieran Flanagan: 11:00
Well, I think if you if you integrate AI across your go-to market, there’s places where it will actually improve your performance. And also there’s places where it improves your efficiency. And so I’ll give you a couple of examples. If you integrate AI across all of your kind of email workflows, so we automate a lot of the prospect and that our sales reps do, and we can, because we gather enough data and data, data layer, the data layer really matters to the performance of your AI use cases. The better the data quality, the better the AI outcomes. And when we integrate AI across all of our workflows and we personalize those emails to the individual, because AI does a really great job of that, we see continual improvements in conversion rates. So in that way, we can measure more and more meetings being booked for the sales reps and less amount of sales reps having to spend their time to do that. Now there’s other ones where we’re building AI functionality for sales reps to actually use when they’re selling, right? That actually helps them increase deal velocity, it helps them to increase their connect rates, it helps them to do all of these different things. So imagine you’re in our CRM and you have an agent that basically can summarize all of your deal details, they can summarize all of your previous interactions, they can provide you great talking points when you’re on the call to say, hey, these are the talking points you should hit. In that case, what we look at is if the rep used that AI functionality, are we seeing an incremental increase in close one rates? So we can actually correlate usage to incremental deals once. So again, we can actually have a good number around that. Now there are other teams where when you integrate AI, it’s hard, it’s much harder to measure. So our product marketing team at Hustlot is an incredible team for adopting AI. And they use AI in a lot of uh a lot of interesting ways. It’s just a harder thing to like say, well, because you’re using AI, what exactly is happening? And they’re not the only example. For the most part, people are using it to be more productive, but how do you measure productivity? What’s the metric for productivity? I was before I was in HubSpot, left HubSpot, went to be the CMO GM for self-serve at Zapier. And Zapier are a company that saved people lots and lots of time. And I saw that all the time, which is trying to say, well, you know, how do I actually, how do we actually showcase the amount of time we’re saving you because we’re automating all of your work? It’s a much harder thing to do. One of the ways you can do is look at usage. Are you actually using these AI tools and just measure adoption and usage and believe that it’s getting better? But productivity is harder than that kind of binary metric where you can see something good happening in your go-to-market.
Sophie Buonassisi: 13:27
Mm-hmm. Yeah, unless you’ve got time tracking across every single activity and every single FTE. It’s it’s a bit more talking about it.
Kieran Flanagan: 13:34
That’s impossible.
Sophie Buonassisi:1 3:35
Yeah, and it’s so hard to do that. Quick pause. Are you a B2B marketer running campaigns for target accounts? And you know the struggle. Tedious and manual processes, endless delays to get things live, and sales feeling like you’re not doing enough. That’s where Mutiny comes in. Mutiny is the fastest place to launch breakthrough campaigns for your target accounts. AI agents, research your accounts, build personalized landing pages, and scale everything from LinkedIn ads to sales handouts in one seamless workflow. No more stitching tools together, just smarter, more impactful campaigns powered by real data. So you can launch in days, not weeks. See why teams who use Mutiny generate 3x more account engagement. Book a demo at mutinyhq.com. That’s mutinq.com. Also in the show notes. All right, back to it. And you mentioned just before this that, you know, AI prompting, if people aren’t interested in learning about it, it’s not a good fit. Now, you yourself, I know you’ve written about the best way to actually learn prompting or learn AI in general, is to actually just do get your hands dirty, be building. What about for folks, you know, that are looking to learn overall and want to get better? Like what would you recommend for people to learn more, to do more, to just overall upscale their AI adoption and prompting skills?
Kieran Flanagan: 14:56
Yeah, I’ll give you three great tips. And I have I have this coming out in a Substack post around Vi Prompton. And so I can’t give you the actual prompts because they are quite long. And so I can’t just like read them out because people will be bored to tears.
Sophie Buonassisi: 15:07
We’ll pop it in the show notes for everyone. Yeah.
Kieran Flanagan: 15:10
I can give you generally what I’m doing. And so one of the things I’m obsessed about is memory. And Chat GPT, now Claude has memory. I suspect Gemini will have memory if it does not yet. And so memory is incredible. So when we have thought about these LLMs, we were like, hey, how do they have lock-in costs? Like, how do they have switching costs? And why can’t you just go from one AI assistant to the other AI assistant? And memory is the lock-in, right? ChatGPT knows so much about me, my results get better over time. But what’s really phenomenal about that is ChatGPT can be an incredible coach. And if you actually prompt ChatGPT to basically say, from memory, from what you know about me, look at my role, look at how I generally use you for the kind of workflows and use cases. And based upon all of that, show me how I could prompt better. So basically show me some of the use cases that I have used you for prompting, and then take that use case and show me a before and after prompt on how you will improve that prompt. So ChatGPT itself is going to start to coach me on how I can prompt via ChatGPT by looking at how I prompt in ChatGPT. It’s kind of like inception, right? And it is incredible. Now that works for anything. If you ask ChatGPT how I can just use you better based upon memory, it can give you really great guidance just how I can use ChatGPT. It can give you new use cases, it can tell you how you can use it much much more deeply. So that’s number one is because if you are using ChatGPT and that cloud has memory, you can actually ask them, if you know how to prompt, you can actually ask them to basically guide you on how to improve your prompt then with them. And it does a really great job. So that’s like tip number one is ChatGPT as a prompting coach. The second is actually the most valuable. I started doing this a long time ago, is every model for the most part has a prompt and guide. And so GPT5 has a prompt and guide that basically came out pretty recently. Every model has its own prompt and guide. And so what you would do is you would take the prompt guide and then you would take a custom GPT who is built on that prompt guide to be your prompt engineer. So I have a prompt engineer that’s a custom GPT for every single model that basically looks at the documentation and then provides me with the right prompt based upon whatever output I’m trying to get. So I would say to it, hey, I’m trying to do this task. I want to do it in GPT 5. Can you give me the right prompt? And it will give me a perfect prompt because it’s trained on that documentation. Now, the other cool tip there, if you want to go one step further, is I love perplexity. I use Perplexity Labs a lot. You can ask Perplexity Lab to create a quote unquote onboarding doc for prompt in for each model based upon elite practitioners. Remember that word, elite practitioners. It will go look for people who have real domain authority in the space, create an onboard and doc, and you can use that to create a custom GPT trained on that onboarding doc for that model. So again, you have a prompt engineer in your back pocket. So there are two. The third one is you can create, if you really know how, you can create prompts to do self-evaluation. And so you can basically have the different models evaluate a prompt and give you edits and tell you how to get better. So that’s why it’s true vibe prompting, right? I’m using the LLM as an assistant to get better. Now I’ll just end with this. I’ve been building a software with a couple of good friends, developers. I I basically was an engineer, wanted to be a coder, was not a very good coder, had got fell in love with vibe coding because I was like, I can build an app myself, kind of ship 30,000 lines of code, and then realize, but I can’t really ship production ready software myself. So I’ve got two developers, and I’ve worked on all of the, I’ve been working on all of the prompts. And I would just like when you when you think of vibe, you need to still do work. And so the LLM can give you a first draft, but you really need to work hard and diligently on the prompts to improve them, right? Don’t ever just you can cut and paste and it’s a pretty good prompt, but if you want great, you still need to actually edit and improve that prompt yourself. I’ve worked on prompts, uh single prompts for days on end, like days on end, just trying to perfect this prompt. And so there’s some tips that you can get much, much better just by using Chat GPT, building these custom GPTs, and using the LLM to do self-evaluation.
Sophie Buonassisi: 19:27
Days to create a prompt. That is incredible.
Kieran Flanagan: 19:31
Yeah, just like iterate and iterate and iterate. And Darmesh had a co-founder of HouseFot had this really great line recently that uh really sticks with me, which is the quality of outcomes is based upon the amount of rep repetition you do for that outcome, right? It’s uh it’s th those lines are in sync. That is how I feel about prompt, and the quality of prompt is is pretty much in sync with the amount of repetition you do. And I re if you repeat, repeat a little bit better, a little bit better, you can get it you can iterate your way to like pretty great results.
Sophie Buonassisi: 19:59
Which is like most skills, even sports, as you train, right? The greater repetition, the better skill set that you come out of it with. Now, when you think about prompting and developing these like these quite complex prompts that you’re working on, first and foremost, I know for you know more the beginner side at least, and something that I lean on a lot is just asking ChatGVD to create the prompt for me to be a little hack rather than doing the myself and then trying to refine it and trying to learn backwards. But when people are progressing and becoming more mature to the rate that you are running prompting, what makes a prompt system scalable? Like, how do you actually scale your prompt system? Because it sounds like refining these prompts like it’s quite time consuming. How do you actually scale?
Kieran Flanagan: 20:44
Yeah, I think I can I can only speak to some of the things that I do. There are like great tools now that do a prompt evaluation, prompt version and control. But I use Entropics Console and Open OpenAI’s playground. And so what they are is you can basically go in and create your prompts. They have really cool tools where like I can go in and say, hey, here’s my first version, and I can click a little button and say optimize and say, Can you make this better? And it will provide again suggestions to like make that thing better. And what I really love really is the version and control. So version and control for a prompt is I have that first version that I did. Now I have a second version that Claude has made much better, and I can run both of them, and I can see, I can give it the kind of inputs, and I can see based upon the output how much better it’s gotten. And so those two systems have worked really well for me. Like again, just using the OpenAI Playground and I tropic console. But there are probably a lot of like more sophisticated systems that allow you to scale your version and control, your kind of prompt evaluation and things like that. I just I just have not like used them.
Sophie Buonassisi: 21:52
Super interesting. And people a lot of time refer to AI now as your co-pilot, right? Or your co-worker or your co-founder. Do you feel like now with AI at everyone’s disposal, everyone is a manager or working in a team, even though they were or are an individual contributor? Has that changed the way that teams are overall structured?
Kieran Flanagan: 22:19
Yeah. I I think you have what’s interesting is everyone really has a PhD level intern that they can work with, right? Because that’s the that’s where the open AI model is. And so that’s kind of bananas. Like you, you know, you used to hire PhD level interns and you they were awesome. And now you kind of have one for free. And you don’t just have a PhD level intern for free, you have as many as you want because I can run multiple prompts all at the same time across multiple AI assistants. I do think it’s a new skill to learn. Like prompting is basically asking asking a smart person to do something for you. Now you have to ask it in certain ways, but there’s a lot of people that haven’t had to work with anyone before, right? And so, like just working with people and you know, giving someone tasks is brand new. And so everyone has someone, now everyone has someone that they’re managing, and that manage that person they’re managing is this AI assistant. So that is like how you can integrate that person into your work and start to really think about well, what is the things that that person can take off me? And it does take some amount of thought to do that. And because of that, where should I spend my time to get more leverage? Right? Like, where should I kind of 10x my skill set if AI is able to do a bunch of the things that I used to do? Because I think this is going to be much more important in the future for me to be a master at. And then I think the other way it changes it is like eventually you will have a team of agents internally, and knowledge workers will have teams of agents. And I think teams of agents, the skill you have to get really good at is how to train those agents, right? So when you deploy an agent, you give it a bunch of context, basically onboarding. Like you onboard it to the task and you tell it how that how what good is, but over time you have to continue to like teach it and tell it how to get better. And so this notion of having an AI trainer, I think is going to be a role in most companies where that person is really going to help train those agents to get better at their task over time and have someone that’s gonna manage those agents, deploy them, onboard them, and improve them over time, and maybe maybe eventually give them performance reviews and do all that kind of weird stuff.
Sophie Buonassisi: 24:27
Yeah, yeah, very true. Everyone is a manager. And I’ve read that you know people can manage personally about six to eight agents at at maximum right now. Whether that’s true or not. I mean, what’s your take? How many agents do you think is possible for one individual to manage themselves now? And then what do you think it will progress to in the future?
Kieran Flanagan: 24:49
I I think the I think it’s as good as the back end this so I’ve it’s actually interesting. I was talking to someone about what I think sticky use cases are in AI. And I think the management platforms for AI assistance is a really sticky use case. And so you can imagine you have a platform where you can see all of the work that agents are doing, you can you can train them within that app, you can onboard them to new tasks. So you’re like a real management platform for for AI, like a version of workday for AI, right? So dependent upon how good that is, it will increase the amount of agents that you can actually manage. Those platforms are far and few between. I don’t think a lot of them actually actually exist. So yeah, I think it I think it’s going to be predicated. But the other thing is I was mess, you know, I I always mess around with building things. You can have an AI manager who manages agents. I have an app, the app that I’m working on has a manager, and that manager is the one giving the other agents tasks, not me, right? And so eventually it’s like, well, how many are managed by the person and how many are managed by the the actual AI managers themselves?
Sophie Buonassisi: 25:55
Super, super interesting. And it it will be interesting to see if we see more workday platforms for AI agents emerging. You know, we’re seeing more like hey Manning Medina’s company for monetizing agents and so forth popping up. So there’s a whole realm of companies that are either emerging or or going to be that are coming out.
Kieran Flanagan: 26:15
Yeah, I I I I’m, you know, I try to be a pretty active pre-seeded C stage investor. I’m an investor in one that I have like a lot of faith in. So that’s like you know, I I invest in things that I have a ton of faith in, and that’s a use case I have a ton of faith in.
Sophie Buonassisi: 26:27
Very cool. And you use you touched around your personal use case for AI in a few different aspects now. You use it also as a chief of staff. What are the use cases you’d recommend to anyone for the most impact when somebody’s getting started with building AI agents? And of course, that’s so context-dependent to each individual, but are there overall like synonymous use cases across the board of go to market or founders that you found you recommend to anyone building and wanting to up level their AI use split room?
Kieran Flanagan: 27:01
Yeah, I have a couple of interesting, like really quick hacks here. Again, I wrote it by the an AI growth operating model that most people could roll out. And there’s a couple of things in there. So one of the easy uses I love it for is if you are running a team, one of the looks I like to have is it’s called a moment a momentum deck. And so it’s basically just one slide. I all of my gro all of my operating models are like admin-like, because I want people to work, not have to do admin. But there’s a slide that says basically, what did I ship in the past two months and what am I shipping in the next two months? And it’s the the thing is structured so every two weeks the same deck is uploaded, updated, and it’s structured in a way where it’s easy, where it’s easy for an AI to pull out information. So I can basically upload it to my AI assistant and say, okay, what have we missed? What didn’t we do that we said we were going to do? What are the areas of overlap? One of the things I look at is every team fills out a blocker and if they have a mic uh mitigation plan or not. And I say, well, what are the blockers that don’t have mitigation plans? So AI is the an ability to like help me keep on top of those things is really good. I do a similar one for KPI. So that’s a momentum. That’s a that’s a momentum look, which is basically how quick are we going, are we doing the things we said we were doing. The other one is the accountability part, which is like every month did we hit the deliverable we said we would hit? And again, it’s a single deck. Every month has a new slide, so it’s all in a singular deck. And the reason they’re in singular decks, just so people know, is one of the frustrating things if you’re using ChatGPT or Claude is you have to continue to like upload the document every time there’s a new update, right? So if I’m if I’ve got my August update and then I get my September update, I have to re-upload the document because it has the September update if I want to query it. Now, if you were querying multiple decks, like for each month, they have a certain amount of files you can upload. I think it’s 20 in ChatGPT, it’s certain similar in Claude. So you’ll just run out of the ability to continue to upload, like you upload the June one, you upload the July one. So if you have them in singular decks, it makes it much easier because you can just upload that one deck. Um, and so that accountability one basically was again, it shows did we do what we said we were gonna do? And it’s really good because I can just upload the doc each and every month, and then I can run a bunch of prompts to say, well, what areas are we missing on? What areas are we overperforming on? What what are the best opportunities the team has seen that we should take advantage on? They can even query it and it can be really your chief of staff in that way. So they’re they’re two of the best like it is a kind of like chief of staff slash project management. If you s if you structure your updates in in in the right formats, that’s that’s one of my best use cases. I love that use case. Um, the other one is if I have a really hard problem to solve, it’s a great thought partner. Now it’s certain model like I have you know all of the models, so I have GPT Pro. The GPD5 Pro, I can’t even remember what it is, like $200 a month one. I don’t even know what they’re called anymore. GPD5, really powerful. I don’t know what it’s called, but but basically, if I have a hard problem to solve, I give it all. All of the context about that problem, all of the historical decks, everything that I think is important. And it’s a great thought partner. And one of the things I ask it to do that’s really useful is red team stuff. Show me all of the ways that this is wrong. My thought process is wrong. It is a counterpoint to you where you think is the best use of it. Because what LLMs want to do is like reinforce your great, you’re great. Yes, you’re right. Because that’s who they’re they’re kind of like tuned to do that. So I force it to tell me I’m not right, be critical. And I that’s what I love it for. Because even in the work setting, we don’t really like being critical to each other, right? Like I know we have like radical candor and all these things, but people aren’t that good at it. AI is really good at it if you tell it to be.
Sophie Buonassisi: 30:40
Yeah. A humbling experience, that’s for sure.
Kieran Flanagan: 30:44
Yeah. Yeah. AI as a coach is a real great use case as well because it’s not biased. It will just tell you like you’re bad, get better.
Sophie Buonassisi: 30:52
Yeah, the the hard, honest truth, always. Right. Go-to-market. Yeah, this is revolutionizing the way that we operate individually, like we’ve been talking about. But how does that thread itself to the greater go-to-market system, meaning how we’re actually building and selling and scaling software companies?
Kieran Flanagan: 31:12
Yeah, I think there’s three trends that I think a lot about that are happening that I can give people a quick synopsis of. So I think we’re gonna have to build influence, not clicks. I think AI engine optimization is the number one skill to learn. And I think multimodal, your your entire website will probably at some point transition to like more of a multimodal experience, and I’ll go through each one. So forever we’ve been trying to like create content to acquire clicks, and that’s how B2B has worked. 80% of all B2B buyer journeys start in Google. Like Google has been the honeypot for how we’ve acquired demand for our business. It’s estimated, I think in 2027 or 2028, 95% of a buyer’s journey in B2B starts within an LLM. Uh and and the problem is that all of our clicks are disappearing. But you what you want to do is you still want to influence your buyer. And I think the way you influence your buyer is not through blogging, it’s through mediums like this. It’s through what I call personality-led growth, which is like I think B2B will look very similar to B2C, where we gravitate towards individuals, not brands. And all of the channels that are still growing and have great momentum favor the individual, not brand. Podcast, newsletter, YouTube, a lot of the social channels, they favor personality, not brands. It’s why some of the best founders, if you look at Roy from Cluy, what’s he really good at? Personality-led growth, marketing. A lot of the founders of AI native startups have real spicy takes, have real thoughts, are like really prevalent across social and people gravitate towards that. So I think your content program looks less like keyword optimization and blogging and looks much more like media and creator-led programs. So I think you’ll have a collection of creators and that’s how you go to market. The second one is that 80% started in Google, 95% will start in LLMs. AI engine optimization is how you drive visibility in ChatGPT and these different AI assistants because all of the research is being done in there. Now, when you look at the data, someone who’s gone through an AI assistant, let’s say ChatGPT, converts four times higher than if they came through Google’s blue links. And people would say, wow, that’s a good thing. And that is a good thing, right? They’re much more qualified. But why are they more qualified? Because they’ve done all of their research in ChatGPT and ignored all your marketing material. So by the time they come to your website, they’re qualified. They have a couple of questions, they’re ready to buy. But you have to be visible in those assistants. And so you really have to learn the kind of techniques for AI optimization to increase your visibility, sure, voice in these AI assistants. And the third one is multimodal. Because they come to your website, they’re ready for a sales conversation. But most people don’t want to do a sales conversation. But these multimodal agents that are able to do voice, do see your screen, do audio, and even sometimes like these kind of digital avatars, I think as they become much, much better, we’re going to see your website transition to a close-in mechanism. Right. Today it’s a lot of research. Like we want to bring our brand to life and tell you why you should buy. We all, I’ve done all that in the AI assistance. I want to talk to someone who can answer these final questions, but I don’t want it to be a human. So I think these multimodal agents, your website is going to shrink, and they’re going to be, you’re going to have these multimodal agents that can have real conversations and answer those questions. And then you could book time with a rapper decide to buy. So those three trends, which is creator-led marketing, AI engine optimization, website as a closing mechanism with multimodal agents kind of baked in, is the biggest changes. I some of the biggest changes I see in the B2B go-to-market playbook.
Sophie Buonassisi: 34:48
Those are drastic changes for go to market. So it’ll be a very, very interesting time ahead. But everything isn’t one-dimensional in the sense like it is still feeding the LLN sometimes, depending on authority and so forth. So it’s actually kind of bifurcating the process where it’s supporting you top of the funnel, but then it’s actually running the relationship side and I’m a closing bottom of the funnel. That’s fascinating.
Kieran Flanagan: 35:17
Yeah, that’s a really good point. Yeah. So like use so L how so what is one way that you can increase your visibility in these AI engines? And it’s basically to create lots and lots of niche content because the way we talk to an AI assistant is very different than we were taught to like search in keywords in Google. And so we’re like it’s like you and I having a conversation. If we were having a conversation about a software product, we’re not doing what we would do in Google, which is like best 10, you know, best SMB software product. And so it means you need to create, instead of like one page that optimized for three keywords, you need to create a thousand pages around a specific like part of your product. And so you do need a website that can cater to that. But the interesting thing is that’s the first example, I think. Maybe not the first, but one of the key examples of where you’re building something specifically for an agent, not for the human. Because the human is not going to consume that content. They don’t care about that because they’ve got their answer from the AI assistant and probably your multimodal agent. And we’re going to teach people, I guarantee this, right? We are going to teach people not to bother have to not have to read for themselves. People are just going to get lazier. I like we see it all the time. When something gets easier and faster, the person consumer gets lazier and they expect more. So they’re not going to go and do like an hour’s research. They’re going to just ask Chat2BT and then they’re going to talk to your agent. But you do need all that content for agents. And it’s like an interesting example of where we start to do go to market for the human and go to market for the agents. Because the thing I’m interested in, very interested in for B2B, and Google released update this week, which allowed agents to do payments, which I think is really huge. Because no one, no one wakes up in the morning and says, you know what I want to get really good at? Buying B2B software. I want to go and like really figure out how to be great at BM buying B2C BD software. So why wouldn’t we offload that whole procurement process to an agent who can do payments? And that way, then who am I even marketing to? And how do I like for like how do I get the agent to pick me? Right. I think that is a a tricky, like that’s going to be a tricky thing for software vendors to figure out.
Sophie Buonassisi: 37:14
Definitely. I love the point that you made about go-to-market for AI agents and go-to market for humans. It’ll be interesting to see how that actually happens and whether it’s full bifurcation or whether it’s integrated. But one area, you know, on the marketing side, we talked about websites and venian. The other side is the creator side. Some argue that AI will erode creative intuition. What’s your take?
Kieran Flanagan: 37:42
I think creativity is I think AI makes creativity more important than ever. I think the way we stand out above the noise is human and creativity. I think people will gravitate towards the reason I think creator led really works. I’ve been talking about it for two years, but the reason I think it really is going to accelerate is because people will gravitate towards people because that’s who we are. Like we’re not going to we’re going to trust people. We want to hear people’s point of views. We don’t want to get all of our content from AI. And I think AI, is it a cre a good creative tool? I think it’s a great creative assistant. It’s not as good as humans at creating genuine creative assets. I think the human sk the human skill to learn is still like creativity that allows me to stand out above the noise. So I actually think it makes that skill set much, much more important.
Sophie Buonassisi: 38:32
Yeah. And what about reshaping marketing overall? It’s crazy to say just two to three years. Beyond that is an even further time frame, but obviously marketers should be learning prompting. What other skills should they be learning? And what is a marketing team fluent in prompts and AI actually look like even one, two years down the line from now?
Kieran Flanagan: 38:57
I think marketing is somewhat unique in terms of a team in that it’s a collection of like niche teams, right? Like if you’re in a sales team, you kind of are a seller and you have the same work and the same career path. If you’re in the customer success team, the customer support team, whatever the team is, it’s kind of the same work and the career path is the same. In marketing, you could be in the product marketing team and the brand team or the demand generation team or whatever team, and they have a niche skill set and their career path may look a little different and their team size may look a little different. And so one thing I think AI does is probably force marketing to be less specialized and more generalist because AI can do this specialization. Because why do we have such breakout of all these niche skills? It’s because of the domain expertise, right? It’s really hard to be a great product marketer. Uh, you can’t, it’s really hard to be a great brand marketer. It’s really hard to be a great demand generation marketer. You need a lot of domain knowledge. So it’s hard to like do subset like multiple of those roles. But if AI has a bunch of that domain expertise, the marketer is like actually, I can be a much more generalist and do more work powered by AI, but I still know a lot about marketing. I have like domain expertise within marketing. So I think one of the shifts will be we’ll see more generalist teams powered by AI able to do much more. I also think marketing can take on way more of the customer uh journey because marketers are always automation for the most part starts with marketer. And so we have these handoff points today that exist because you know we have to hand the person over to the sales or whatever it may be. But I think as AI becomes more prevalent, it may just be that marketing can do much more of the customer journey because they can integrate AI, and AI is doing all of the qualification discovery. AI is doing a bunch of the work, and marketers are like managing the assistants and training the assistants and training the agents to do that work. Um, I don’t know where we end up in two years. I think the thing is it’s changing so fast. So, what I tell people is the most important marketing skills to have are be curious and be persistent. Curiosity, there’s never been a better time to be curious. There’s never there’s never been a more important time to be curious. I think it’s the number one skill set, the number one trait to look for, because if you’re every everything is getting rewritten, and for me, that’s awesome. I think I get really bored when everything is like optimization stuff, twiddling the you know, twiddling the knobs, just getting a little bit better. It’s way better like when you have to rewrite everything. And so people who are really curious will be able to do that. And coming back to the Darmesh quote, your quality of outcome will be dictated by your number of reps you put in, which is really the people hate to say this because everyone likes to say, well, like work-life balance, it’s the grind, right? Like the grind does matter. As Darmesh says, more is more, which means like working hard and grinding it out and learning is going to be a really important skill set to have uh in this time.
Sophie Buonassisi: 41:45
Yeah, if you talk to anyone that’s you know been on the other side of the mountain of climbing and acquiring skills, they’d probably say the same thing. Of building company, just say the same thing. You know, you look people emulate and replicate the end outcome, but we should really end and lead is the process. And in that process are thousands and thousands and thousands of reps. And now, you know, it’s never more important to learn AI, but it’s also never more fun. It’s never more fun to be a curious person. Like this was like the world at your fingertips of you can build, you can create, you can learn. It’s I think the best time maybe a go-to-market professional, the best time would be in tech in general and a burn career.
Kieran Flanagan: 42:26
A hundred percent. A hundred percent. I I totally agree. I think it’s the best time. It is the the it is the number one time because the URL it’s like every start of every it’s where like people make their success, their careers, right? Like a lot of my career was made by being one of the first to adopt inbound and product like growth. And so people have like these new paradigm shifts, they they reset everything and there’s a bunch of new winners. And I think that is why it’s so exciting because the new winners are not based upon like your title or any of these different things. It’s based upon your curiosity, your iteration, and your ability to like really learn rapidly and and really work hard.
Sophie Buonassisi: 43:06
Definitely. And I’m also really interested to see how the actual funnel evolves to your point of marketing might take it longer. It might not be a pass-off. HubSpot actually spoke to that recently at inbound, right? Introducing the loop instead of the funnel. Yeah. And just how it’s not it’s not a linear process anymore. The buying process has evolved. So that’ll be an interesting one to see how it it shapes up.
Kieran Flanagan: 43:30
Yeah, exactly.
Sophie Buonassisi: 43:32
And Kieran, you know, we talked about how you learn around AI and experiment, but are there any Shaver books that you have and really shaped your career of the years?
Kieran Flanagan: 43:44
Oh, I should probably have a good answer for this. It’s been like I’ve been so engrossed in AI for like two plus years. I’ve forgotten what I even read before. I honestly don’t have not read much at all other than uh uh work. I I like I consume a lot of podcasts, I consume a lot of like content, but I have not read a lot of books, if I’m being totally honest.
Sophie Buonassisi: 44:13
You have to learn from dynamic sources.
Kieran Flanagan: 44:17
Yeah, yeah. Like you have to, and I I think like there’s a time to consume and a time to work. And I I’ve really kind of leaned into the time to work. Like I’ve learned prompting, like, how have I learned a lot of things around AI? I have a YouTube channel, AI, I have a Substack in AI. I’m building product. I have a product coming out in AI. I work in AI every day within HubSpot. So like I made a commitment two years ago that there’s just no one that will know more than AI about me if you’re a go-to-market practitioner, because no one is going to work as much as me in AI. And that was like my that was the only goal I had. I didn’t have like any, there’s no financial or anything. That’s that was it. Like, and so I think there’s like times in your career where there’s like a good time for consumption where you’re really trying to figure out how do I master a new skill, and there’s just time to act. Um, and a lot of the content I consume is like in the moment where I’m trying to figure out problems.
Sophie Buonassisi: 45:06
It sounds like, I mean, first of all, I love that. It sounds like you spend a lot of time learning and upskilling around AI. Now I have a question, very important question for you, Karen. There’s a little birdie named Brian Halligan, give me a tip-pop, but you’re actually a pretty skilled rapper. So, how do you have someone to learn the rap skills while learning in high?
Kieran Flanagan: 45:27
Uh I can’t believe Brian said it was of all the things Brian could have said about me, me being a skilled rapper, how do I so so I am uh I I all I all I listen to is hip-hop. I used to hang out like in there, used to be like a battle rapping forum where you could battle rap people over text. Now, I that doesn’t exist anymore. If it did, it would be kind of interesting because ChatGPT is a great battle rapper. I’ll tell you the funnest thing I’ve ever did on like in in rapping. So Fiverr is a really cool platform, you can use it for a lot of things. And me and my bros, my brothers, who all enjoy hip hop, we used to battle rap each other by paying this person who did like Sesame Street puppets. So we would we would create the back, the back end track, and then we would have the puppet rap, and then we would send my brother would send like me the video, and then I would respond as a puppet and video. So I’ve done a lot of weird things around rapping, but yeah, I’m I don’t I don’t think I’m uh a very good rapper, but I’m an aspiring. I have started to like use ChatGPT to like relive a lot of my youth fantasies, and like one of them was to be a builder, which is which is it’s helping me do that. And then one of it is to write raps, but at the moment I’m just sending them to my brothers and they’re just offensively, they’re just like offensive things.
Sophie Buonassisi: 46:36
But my brothers incredible, incredible. Well, we’ll see. The worlds are intersecting, AI is supporting the rap dream, and it’s all coming together. Who knows?
Kieran Flanagan: 46:45
Yeah, I could have a I could have a I someone was one of my team was showing me they had a person that they love listening to and they were looking for concerts of his and they were like, hey, I just found out there’s no concerts because this this guy is AI and he’s on Spotify, he’s like really popular, and so who knows? Like I could maybe create a little AI rapper. Uh it did inspire me that I could create a hip-hop like artist and like just put him put that person out there.
Sophie Buonassisi: 47:10
I mean, I have heard you say actually that voice is one of the underrated utilizations of AI. And obviously, we’re probably talking about a different context, and we’re seeing a lot of technology come out leveraging AI for voice use cases, but that’s a great use case.
Kieran Flanagan: 47:26
Yeah, yeah, like the Hey Jen models and these new models, they’re just incredible.
Sophie Buonassisi: 47:30
Truly, truly. I love it. Well, you are, as you said, already one of the best people to learn AI from. You are committing to apply to actually be the best, most knowledgeable person in AI. And already so many people are following your learnings, and you’re one of the people of helping to shape and disseminate information around AI to people. So for anyone, if they don’t already follow you, where can they get in touch and follow you across all your platforms? And these will all be in the show notes for everyone.
Kieran Flanagan: 48:02
Yeah, I think the number one thing I got asked to do was start a Substack because everyone was like, hey, you share so much, can you just document it in Substack? Especially the prompts. Everyone wanted the prompts. And so I started a Substack called the AI Journalist. I’ve actually been amazed how well it’s gone about three months ago. And so that’s probably the best place to go because I suspect people want where are the vibe prompt and vibe uh prompt and prompts you talked about? Where are all these things? They’re all in the substack. So that’s where you can go and you can get it. The other one is with my really, really good friend Kip. We do a podcast where we cover this stuff as well. It’s called Marking Against the Green. So they’re kind of the two core places.
Sophie Buonassisi: 48:35
Love it. Those will be in the show notes. I’m a huge fan of both. Highly recommend. Karen, this has been fantastic. Really appreciate the time and you sharing with everyone.
Kieran Flanagan: 48:44
Yeah, thanks for having me on.
Sophie Buonassisi: 48:46
Absolutely. Thank you. Thanks to everyone for tuning in, and we’ll see you next week.
