← Marketing Against The Grain
Marketing Against The Grain
The AI Workflow That Lets 50 People Do the Work of 500 ($2B Founder Reveals)
00:44:58
Loading summary
Transcript104 lines
- [00:01]
A
Welcome to Marketing against the Green live from London. We sat down with Gamma's Grant Lee and Kristin Fracchia, who built a $2 billion company with just 50 people. And AI has fundamentally rewrote the playbook that they've used to grow and many other companies are using to scale their business. We give you the exact playbook of how we turn every salesperson into a closer. We get into some of the things that are working at HubSpot, Gamma talk about some of their most critical AI workflows that are helping them to scale millions and millions of users. And then we give you the number one problem we see in AI today, where humans have zero tolerance for AI. So stay tuned and let's get into today's episode.
- [00:57]
B
We've got the two lovely folks from Gamma here to join us as our special guests. And a few things I want to start off with. AI is fundamentally changing how companies grow. You guys just raised a big round today. Congrats. You guys have basically built a $2 billion valuation company. If I'm reading the reports, you don't have to disclose with 50 people, which is pretty remarkable. And so I guess to just start us off, what is the lessons of building like a high growth, high scale company in this AI era? That's different then. Like, I don't know, when we did it at HubSpot, it was, you needed a lot of people. It was very different than the kind of companies that are being built today.
- [01:33]
C
Founders oftentimes think about, when they think about innovation, it's like you innovate on technology, you innovate on product. But I really believe that founders also have a chance to innovate on org design, which is how do you build a company from the ground up? And for us, you know, we've tried to stay lean from the very beginning. We lean into things like hire generalists over specialists. We lean into this notion of instead of like traditional management layer, we have player coaches, people that are really close to the work, kind of mentor and coach those around them. And like a lot of these small things that have just been baked into our DNA, how we operate, the culture of the business, I think have helped us to scale to where we are today. And just again for us, just trying to think about doing a little bit different every step of the way.
- [02:13]
B
I think that's really interesting. And we're going to kind of talk a lot today about how we actually scale and build with AI. Kieran, you and I, we're always hacking together stuff with AI. We're trying to build a really AI first team and go to market at HubSpot. You all are empowering an entire generation of people to build first with AI. Kieran, maybe you could kick us off with like what's the current state of the state in AI? Like what is the AI landscape in terms of where do you think product market fit exists today?
- [02:43]
A
When I think about the internal applications and then the companies building for AI, the likes of Gamma, the other incredible stat is 70 million users. These companies are able to grow in a very different way because you're able to unlock something magical that a user could not do before. So there's this like real virality to the way those companies are growing. We covered Gamma's marketing playbook a little bit on marketing against the grain creators. The way you kind of do paid and we've scaled for the internal application of AI. We spend a lot of time trying to integrate AI across our go to market. I have been doing that for 18 months within HubSpot. I kind of break that into three acts. The first six months was a grind and we really didn't get very far because the AI models were not good enough. The next six months we started to see real signal in a certain couple of spots which really is like personalization, AI prospecting, things like that. The last six months as the reasoning models have got much, much better. AI has really started to transform the way we've done things internally. And so I think it is an incredible time to start to deploy AI across kind of go to marketplaces and we're going to talk a little bit about that. And I think there's still a lot of noise in the AI space and it's hard to like decipher like where I will be a perfect fit for use cases. So there's a lot of experimentation and testing, but I have never seen so many internal pilots in terms of what we are doing. So much experimentation and I've never seen companies grow so fast before that have really caught the zeitgeist on that kind of magical AI moment. So I think we are entering a year in 2026 when I'm going to start to see this as being like interesting and impactful to truly transformative in the way we actually grow businesses.
- [04:26]
B
Yeah, and I wonder from the Gamma side of things, like how is that mapping up with what you're seeing from customers using the tool, you know, the types of use cases, what's being built today versus maybe what was built six to 12 months ago and trying to paint for everybody the trajectory we think we're on and kind of start to extrapolate into the future a little bit.
- [04:44]
D
Yeah, I could go ahead and start with that. So, first of all, I have been with the Gamma team for, it feels like years, but I think it's been eight months and I have a year in AI. It's been about eight years, I think. And in that time I have seen a Gamma product grow dramatically as the models improve, the product improves together, exponential improvement. But one thing that I have noticed is that earlier on people were sort of playing around with it a bit. Right. It's fun. Make me a deck about fish.
- [05:12]
A
Right.
- [05:13]
D
You know, not really using it for serious use cases. This year we started to see more professional use cases and that's something in our marketing. We really wanted to get out there. We were using it for professional use cases. We use Gamma for everything internally. And so we wanted to share that with others. So we've supported that in our marketing, but we've also organically seen people start to adopt that more. And then the biggest change more recently since we released our API, it unlocks a whole new range of ability to attach Gamma to other tools. Right. So Clay to Gamma. I think Grant will talk a little bit about this in his talk tomorrow on stage at HubSpot Grow as well. You can take Zapier, make N8N and really attach all those blocks of data sources, the walls of text, and get really amazing visual outputs in Gamma. And so that's what's really interesting to me right now is how people are really integrating Gamma into their workflows.
- [06:01]
B
Yeah, I think what's interesting, if you're out there and you're thinking about this AI world, the creative work has taken a step function over the last six months in terms of the video models, the image models, especially the Google image and video models have gotten really good. We saw Wevy AI got acquired by Figma last week. That is a really great team. And that is a AI workflow product. But just for like creative use cases so that you can do a hundred different revs of a social post or ad creative or different particular cuts of B roll that you might want to test. And I think what you're basically saying is that's now just going to become default, that these classic creative use cases, making a deck, making a video, building an ad, are going to be workflow ties in the same way that email campaigns and everything else were over the last decade.
- [06:53]
C
It's taken us, you know, a couple of years to get here. Right. So we think about we recently launched Gamma 3.0, but go back, you know, two years ago, you have Gamma 1.0. I oftentimes think about like, 1.0, as this is for the AI tourists, right? They're going to come, many are going to leave, that's okay. But they're going to find what's interesting about your product and, like, be really crazy adopters of that. And then you get to your, like, 2.0 launch, and this is for your true sort of early adopters. They're still willing to put up with a lot of shortcomings of your product, but this is where they start pulling product out of you. They tell you all the things they wish your product could do, and hopefully you can listen. You build and iterate and then you finally get to the point where you feel like you can launch a 3.0. And 3.0 is the first time you're actually going really after the mass market. Right? Like, can these people that hate change actually embrace your tool? And that's the ultimate test. If you can do that, then you have a chance of being a tool that's going to be around.
- [07:49]
E
Hey, founders tired of stitching together tools that break when you scale? HubSpot for startups is the GTM engine that helps you scale faster. One connected platform where your data powers AI teammates that handle prospecting, content and support so you can focus on growth. Trusted by 35,000 startups and more than 300 unicorns, qualified founders get 90% off year one@HubSpot.com startups.
- [08:17]
B
Okay, I want to dive in on this with everybody because we have a lot of startup folks in the crowd right now. Here's my biggest observation. The gap between the startup world and the AI world and the Main street business world has never been bigger. It is enormous. Most people I talk to, they wouldn't even understood about like, five words of what we've covered so far tonight because they're just like, well, I hear about this AI stuff. What do I do? And I think you all have done things to drive adoption and learn how to get people who are just not that they're maybe skeptical, but they just don't even know where to get started. They do not have a foundation of information. And so maybe, Kieran, we'll start with you. And then I want to go to Grant and Kristen around, like, how should everybody here, if they're working on a cool AI product, think about reaching that kind of Main street audience that just doesn't get it?
- [09:12]
A
Yeah. And I love the three kind of Personas you put down in terms of Gamma's history, because I think that's the way I see any kind of AI usage where if you're a nerd like me, you're willing to put up with a lot of faults and you're willing to kind of iterate and try things. But if you're a person who just is trying to do their work, AI has to help you do that work better. You're not going to spend hundreds, hundreds of revs. So I can give you some examples. We try to get many folks internally to adopt AI products and AI things that we build. I think sales are maybe the hardest Persona because they are trying to hit a number. You're trying to interrupt them with new functionality. I can give you like one interesting story that I thought was fascinating. We had an internal use case where we kind of trained an AI avatar to replicate what our sales engineers team would do.
- [09:55]
E
Right?
- [09:56]
A
It's the entire training set. It can bring up the right visuals. It was a pretty incredibly knowledgeable avatar who knew everything about HubSpot. And that avatar would join a call and the rep would conversate with the person and when there was a technical question, they could call the avatar and the avatar would come to the forefront of the zoom call and answer that question. Now what was fascinating was the avatar was like better than the human because it had a far bigger knowledge base. It could answer any question, it could bring up the right visual, but if it was asked something that was not in its training set, it would not get that correct and would have to retrain on the call transcript. It would know it the next time. Now, if you are a person doing that job, you get some forgiveness. You get something wrong, you're coached, you get better. The human had zero tolerance for failure from the AI. As soon as the AI was wrong, they were like, I don't ever want to use this product again. So there was a very low threshold in terms of I do not trust this thing. Data was a big part. When we try to show sales reps or customer success reps or customer support the next best action, this is what you should do. Because AI is telling us to do this thing. It didn't trust it unless you showed its sources and data. And then the big thing for any startup, what we have found, again, this is really for internal use cases. So it may not hold true in all use cases is it is for the mass market. AI cannot be a different thing you use. It has to be within the current workflow. It has to be in how they already do something. Right. So you're trying to like build into their current workflow, not try to teach them a brand new workflow. And I think that is some of the ways that we've managed to kind of crack adoption internally in terms of a lot of the AI functionality that we've kind of built.
- [11:34]
B
What have you guys Learned? You've got 70 million users and so you've had to learn how to do this at scale. What are the big lessons?
- [11:41]
C
I think for a lot of categories, it's hard. Like a lot of this is outside of your control. Certain categories like coding, you know, it just very obvious you don't need to spend much to educate the market. Everybody's just embracing it. And to be fair, Copilot had been sort of priming the market for many years, and then cursors came out. You know, the cursors of the world would come out and the market's already ready for that. I think for, like, areas where text is heavy in usage, it becomes obvious like, there's a lot of adoption. Customer support. Legal is now adopting a much more, I think in our territory where it's much more multimodality, I think people are still trying to figure out how best to embrace it. And this is where a lot of the workflows come into place. And also just building a tool that feels familiar but different, I think you need to then figure out how to actually tell that story. And candidly, we're still figuring it out. I think we benefit from a lot of our early adopters helping us spread the word, but still, early days, overall.
- [12:34]
D
Yeah, I think there's a couple of things I know we've done at Gamma that have helped that might be helpful to others who are building an AI startup today. One of them is we have built Gamma to be very accessible. We've tried to abstract away a lot of the AI so that you can focus on getting the job you want to be done. And so ideally, you don't have to spend a ton of time prompting and fine tuning and finagling in order to get a good result. But we're dealing with LLMs here, right? You know, prompting matters. And so something that we've been asked for for a long time is to create a prompt guide, right? Just give people something to start with. So even though Gamma, I think, has done a very great job of solving the blank page problem, nothing is worse than staring at a blank slide. You know, at least you can put in some thoughts and get something out. But it's still, you know, getting started, right, Finding the right way to get started and being able to have a good prompt in order to get a good result, it's really important. So I mean, this, just this week we released our first prompt guide just to give some people some things to get started with. Because once you can see something that works, I think you have a higher tolerance. Right. For working through it later on. And the second thing, speaking specifically to our marketing at Gamma, that we've had to try again and again is to bring people back to the product. So one of the things that pains me the most is when someone says, oh, I tried Gamma a year ago, like, haven't come back. Like, you have no idea how much better it is now. And that's true for virtually every AI product. And you know, ChatGPT and Claude as well too. And so that's something we do intentionally in our marketing, is find the moments to bring people back. So when we launched Gamma 3.0 in September, we did do a big marketing push about it. We talked about the latest features and we actually saw a step change in our percentage of weekly active subscribers. So it was here and then it went like, this is the happiest chart I've ever seen in my life. And it stayed there. And so I feel like we brought people back. Right. We're able to interest them again. And so I see the model companies doing this pretty well as well too. And I see other AI companies doing it too, that seem to be standing out in the space. You have to be bringing people back constantly.
- [14:26]
B
Yeah. I think the other thing that we've learned, Kiran, that you didn't cover, that I would add, is the fastest growing AI product at HubSpot is our integrations with ChatGPT and Cloud Gemini. It's because we bring the product experience to the new operating systems, whether they be browsers, whether they be desktop apps, whether it just be, you know, a tab in somebody's browser itself like that. We are seeing the adoption and use of HubSpot, getting insights, building campaigns, you know, doing an outline to go build a slide, decking Gamma. But doing that with the data from HubSpot inside ChatGPT is also something that I think is a real strategic decision that any company building an AI has to think about is how do I need to integrate and build alongside the next kind of operating system layers that are really starting to emerge right now.
- [15:17]
A
I think the prompting one is a really interesting one because one of my biggest gotchas, you know, the thing I have to constantly remind myself on is the kind of difference between the consumer and me being in AI all the time. And one of the things I was really caught out by was when I was at Zapier, I was trying to overhaul the entire onboarding flow and this was like pre reasoning model. So I would just caveat where the AI was probably not where it was today. And I was so confident of natural language onboarding through the AI being much, much better than the click, click, click experience where the AI agent could build the workflow for you. I bypassed my usual 6 week sprint build minimum viable. I was like this is going to work. And what I realized was it didn't work, which was not good. It was embarrassing because it took like three months to do it and it didn't work for like an interesting reason. It wasn't like the capability of the actual AI to do the work. When I looked at the call transcripts or the chat transcripts with the AI agent, people didn't know how to ask it for what they wanted. They were like Facebook CRM leads in please, like how am I going to do this? And so I kind of realized that like today you can like ask open ended questions to better pattern match into your rag. There's ways doing that. It's why ChatGPT asks lots of questions to refine your own question. But I think that is a big part of it is like really staying close to the customer and how the customer experiences AI. Because when you're so in the weeds, your experience is very different from the person sometimes usually adopting it. The one other thing, I'll just say I wonder how you think about this. Kip and I have talked about this on the show. I think one of the biggest problems historically for AI creative tools is isn't actually the creation of the thing, it's the editing of the thing. The editing of the thing usually sucks and it's the reason people have a bad experience and leave. And I'm wondering how you found that as part of like the gamma journey and solving that.
- [17:07]
C
We sort of had an unfair advantage in that we were building the editing long before we had the AI, right? So we were building the editing experience so that it would be easy for the human to create slide decks easier without having a design background, design skills. And then about two and a half years ago we realized that the learning curve for even that was too steep. Most people don't have the time to learn a new tool. So we invested deeply into rebuilding the entire onboarding experience so that people could experience that magic Using AI immediately and the marrying the two has been really helpful. Now that the AI can pre assemble and give you that first draft, you as the human can still go in and do all the editing you want. I think that's been for us a nice little unlock for users.
- [17:49]
A
Yeah, I think that's really important.
- [17:53]
E
So here's what nobody tells you about building a startup in 2025. You don't need a massive team anymore. Gamma just proved it. They built a $2 billion company with 50 people. The HubSpot for Startups team teamed up with Mindstream to show you how the best AI tools actually work together. This guide gives you the seven AI tools replacing entire departments including Gamma, Notion, Claude and more. It includes copy paste prompts to launch products, hire employees and find customers fast. It's completely free and if you qualify for HubSpot for startups, you could get up to 90% off year one. Scan the QR code or hit the link in the description. Now let's get back to the show.
- [18:36]
B
Okay, everybody here loves AI and we also all love AI hacks. I want everybody to share their favorite AI use case, personal, professional. Just like the way you're using AI right now that you find just to be super valuable brings you lots of joy. Who wants to go first?
- [18:53]
C
I can share a couple one and then I'll share a smaller one. So one obviously for us that I'm observing, a lot of our early adopters of our API use is all centered around this notion of like customer communication. So you know, just imagine like you're an agency or a freelancer or consultant who, you know, have a lot of meetings with clients and you might use tools like Granola to like record all the meeting transcripts and you can immediately then take those transcripts, work with Zapier or make feed that into a tool like Gamma. All of a sudden you have, you know, the sort of presentation completely done. So such that by the end of the meeting in their inbox you have a recap of everything you discussed. You have all the deliverables agreed on, you have the timeline, you have action items and the client is super happy. They know you listened and there's a clear set of next steps. I feel like those are just going to be sort of no brainer situations. We have one agency that's basically running their entire RFP process. Doing this, which used to take weeks, can now be done literally in minutes, seven minutes, end to end granola data.
- [19:55]
B
Into a created deck with all of the follow up and RFP info, right?
- [20:01]
C
Exactly yeah. And then obviously Future State, even enriching that with HubSpot data, like there's all sorts of magic that could happen and that sort of thing where that level of automation just wasn't possible for us. So we're super excited to see things like that on a personal level. You know, I still use Gamma for sure, but then also tools like Midjourney, where we tried to sort of open source our brand a little bit, where we share kind of the SRF codes we're using in midjourney so that anybody can create the same brand assets. Me trying to pretend to be a creator myself going on LinkedIn and X, it's oftentimes really helpful to create things that feel on brand. So having this thing where anytime I have, you know, something I want to create, I can create a carousel that always feels on brand that for me feels like a hack. I was like, oh, this is great. This is super easy. I couldn't imagine trying to do it before. And now, you know, and any day I can go in and create that sort of level of, you know, quality that I've been looking for.
- [20:49]
D
Okay, Grant stole my hack. Okay, this is going to sound like another plug for Gamma, but meeting notes are the bane or meeting follow ups are the bane of my existence.
- [20:59]
B
They're everything. Not just you.
- [21:00]
D
We all know this is true. Everybody at Gamma knows this. I'm like, AI notate to Gamma is the best hack ever. You look great internally, you look great externally. You've got your granola, your zoom transcript immediately. Five minutes later, here's the recap of next steps. Like, how many times have you had to like, type in, okay, here's what you talked about and recap for somebody who wasn't there. And this is what we agreed on next. So internally it works great externally. So impressive to be able to send off through a client or prospect right after you get off the call. So that is actually truly my favorite hack right now. But I'll share another hack which is just generalized and I don't think people do this enough, which is using AI to write your prompts. So I think you've talked about this before on the show, but I don't see many people do this. It's so underutilized. ChatGPT or CLAW can write really great prompts for itself. So if you start with a task and you're like, I want an output that looks like this. This is what I want to accomplish and ask it literally to create the crop for you, I always get way better prompts. Than I would have come up with myself because I'm lazy and I'll just say positioning statement, you know, otherwise. Another thing I really love now is using Claude's skills. So I have Claude skills plugged into all of my Claude projects for various things in marketing. This is my new favorite hack because all of my cloud projects now pull from Claude skills when they're relevant. And my favorite new one is the one I trained to not write like an AI, which was actually a really interesting anthropological study as I was watching it think through what looks like AI writing or not. And so now I have my human copywriter editor. It's not perfect, but we all can identify AI writing new EM dashes.
- [22:30]
B
And this is dashes.
- [22:32]
D
You know, it didn't tell me it didn't identify EM dashes or rocket ship emojis at first until I was like, what about the EM dash? And it's like, oh my God, you're.
- [22:39]
B
Right, you're right, you're right, you're right. Icloud love the M dash.
- [22:44]
D
Yeah, just it was, it was really big on your overall parallel structure. I should share the whole list of what of the AI writing tells.
- [22:49]
B
But that should be a whole LinkedIn post.
- [22:51]
D
I would totally now my writing is at least like slightly more human, you know, I mean it's a step change more human than it was before. So I mean just having that as a foundation and that you're building on top of it is so important. So really love layering the the cloud projects, the cloud skills, getting all that context in there.
- [23:07]
B
All right, Kieran, you are deep in the AI wizard.
- [23:11]
A
You know my. One of my favorite use cases in store too ruined it is I have friends who don't know anything about AI and I used to create many videos mostly of Liverpool because I'm a Liverpool supporter. They're not a supporter. I apologize for people who are not supporters. But like images of Liverpool winning the premiership and photos of them in it and then videos of Liverpool win in premiership and I superimpose them in it. And then Stora 2 ruined it because I'm in these WhatsApp groups and I have like my tech WhatsApp group and then I have that WhatsApp group and when Stora 2 came out, their entire conversation for the entire week was why is my Instagram feed nothing but Stephen Hawkins videos doing like hula hoops and things like that? And they looked into it and realized that it was all AI. So a couple of ones I'll kind of build on the prompt one first and I'll go into a couple of others. The AI's prompt engineer is really good, but one of the cool things is if you kind of build onboarding docs for custom GPTs, you can onboard them to skills in the same way you would with an employee. So if you go to Perplexity and you say, hey, like create me an onboarding doc for someone, I'm going to teach how to prompt GPT5 to. You can take that onboarding doc and then use that to create a custom GPT for prompting for GPT5. So I have like a ton of different custom GPTs that are onboarding different skills that works really well. Cloud Skills is really, really cool. We built an app that we kind of showcase in the pod where you could upload a job description and you can have the app give you a bunch of skills based upon that job description that you can actually just use for your day to day work. A couple that I like. AI is really great at mimicking characteristics of people and that was where I was originally fascinated by AI. So I have a custom GPT that will change its thought process based on what I asked it, right? So I've taught it first principles. I've taught at these different thinking methodologies. And so when I want to have a real conversation, I would say and I walk that back in first principles and I walk it back in this, you know, thinking methodology. And it's really good. As a thought partner, I love it to red team ideas. I tell it to be really, really brutally honest. I give a prompt on LinkedIn where I showcased how you can make it be really brutally honest. And someone actually emailed me personally and said, I'm so thankful for that prompt. I cried, it was so mean to me. And I was like, ChatGPT, is it a great. Is a great ChatGPT coach? If you ask ChatGPT based upon memory. Look at my memory. What way could I have used you better for certain use cases? What ways am I underusing you? It's really good at being a coach for itself. The one a couple of go to market ones I'll just kind of quickly touch on there. A lot of them are my favorite right now are sales use cases. One of the big things that we want to do internally is turn every seller into a closer because 35% of a salesperson's time is spent selling. Everything else is spent on the stuff around selling. And so the one I'm fascinated with right now is if you kind of spend a lot of time with sales People, they have to do a lot of discovery and qualifications. So the first call, they're trying to figure out your budget, they're trying to figure out what you want the product for. And we are at the moment experimenting, having that first call be done through an AI agent. And so the agent can have that conversation. Then we run that conversation through like a LLM model, pull out the context, give it to the rep, and then try to get that rep to close deal in one single call. Now that's very experimental. We're like very early stage. AI has been really great for all of our kind of workflows. And personalization via email has worked really well in AI. We've generated about 30, 40% additional meetings each and every month by just personalizing the email for the person. So those kind of standard ones work pretty well. So there it's a couple you could keep going.
- [26:30]
B
Yeah, I understand. First of all, I have to represent the most underrated part of AI that never gets hype, which is just voice dictation. You're not using voice dictation AI, you are just wasting an hour a day. Use will of voice, you can use Whisper, you can use the new Claude app has good AI voice dictation, but that is complete game changer. And nobody talks about it, Nobody uses it nearly enough. I talk to my computer, my little office all the time and it looks ridiculous, but it saves me a bunch of time. The other thing that I would recommend, pick your AI browser du jour. I tend to like Comet the best right now. Whether you like Atlas or DIA or what are the others? A very basic kind of person. And one of the things I always do is I find somebody who does something in marketing that I like and I just try to figure out how did they do it and who did it, because I want to hire the person who did it and then I want to figure out the tech, the infrastructure, the strategy to do it. And what's great now with these AI browsers is that you can take an Instagram image, you can take a brand campaign and you can just open a few tabs and drop like the Instagram posts in, drop the company website in, and then basically just tell it what you liked and ask it to reverse engineer how they did it and they will give you the tech stack, they will give you the likely steps they did to do it. And then you can also then research who likely did it at that company and you can do all that in a couple minutes. And I used to spend hours doing all that and that's just like a. Just a hack to like figure out what a cool marketing tactic might be in the future based on just what somebody else is doing.
- [28:03]
A
Yeah, I saw someone do that with. Have you used the new Kimi K2 model? I've saw a lot of use cases with that and reverse engineering and research and it is absolutely incredible. It's like my tomorrow night is just Kimi K2.
- [28:19]
B
You have the funnest night.
- [28:20]
A
Yeah, yeah.
- [28:21]
B
You have the most fun nights.
- [28:24]
Okay, so those are a few hacks. We have a few minutes and I want to see if there's any audience questions I can keep going. But this is a very rare, like intimate version of doing this. If we do live shows, it's normally like hundreds of people, so we can take a few questions. Anybody have something that you want to make sure we cover? We got somebody right here.
- [28:41]
C
How do you keep everything on the same page? Especially also on the product decisions as well? Yeah, I mean, so how do we keep everybody on the same page?
- [28:50]
B
How do you operate? How do you run this?
- [28:51]
D
How do you hire excellent people?
- [28:54]
C
Kristen, does it help?
- [28:55]
B
That is a big part of it.
- [28:56]
C
Sure have to worry about any of it. I mean, from the very beginning, we've had a very transparent culture. So I think we do everything we can to preserve that. Today we still fit in one floor, one office. It'll be much harder, obviously, as we scale, but we have our little rituals. Like every Friday we have show and tell. That's a chance for anybody in the company to go up, present an update to the company. Right after we do kind of a fun ritual which is called Gammarama, which is anybody at the company can present on a fun topic. And at the end of obviously using Gamma, at the end of presenting, using Gamma, they can give us feedback on what they wish were better in Gamma. How can we improve it? So it really is like the feedback loop. Even at the most individual level, everyone's contributing. So I think, you know, these are the things we'll try to preserve. I think transparency is a core value of ours. And yeah, as we get bigger, it'll be harder. But fortunately, even at our scale, you know, the reason I think we've moved so fast is we've kept it super small. So 50 people, you know, in a prior life, at a different startup, same level of traction, we are 500 people. So obviously just the different dynamics there, we haven't had to yet deal with that just yet. Very cool.
- [30:02]
B
Anybody else? We got a couple over here. Hi.
- [30:06]
F
One of my favorite, like talks Ever. Thank you very much. I come from a marketing and branding background as I have a question for Grant but anyone on the panel might like. So this is the first time I started a startup about a year ago and it's AI and we recently pivoted to video, actually automated the workflow of classic TV ads and we've been discussing things some big corporates in my previous career and the one guy in a meeting said he believes you can teach a filmmaker how to prompt but you cannot teach a prompt how to make a film. And I don't necessarily agree with that. What is your view?
- [30:40]
B
Creative specialists who work with AI is kind of the spirit of the platform.
- [30:44]
F
Especially from the bigger corporates and stuff. The lack of knowledge of AI and the potential is missing. And like I'm struggling with this question now.
- [30:51]
C
Yeah.
- [30:52]
F
Because I don't know the answer.
- [30:53]
C
I mean the first question is are you trying to remove the human from the equation or not? Is the human still in the loop? Is the tool for the human to use or is the tool potentially to replace the human altogether to create a new workflow? If it's still for the human to use then I think the considerations are a little different. Like you're really trying to guide the human towards best leveraging the AI to get to the ultimate outcome. And the human's taste still factors a lot into that. If you're trying to remove the human altogether, I think that's where the paradigm is definitely different. And then of course then you have to think about who's the buyer of the product and, and like how does that actually integrate into everything you're trying to build. So I think I'd have to better understand kind of the specifics there.
- [31:32]
B
Kieran, you have a very strong opinion on this topic. Do you want to give it kind of filmmaker prompt? You need creative specialists. Do you need creative specialists?
- [31:42]
A
Oh, I think you.
- [31:43]
B
We hire more specialized people now than we did a year ago.
- [31:46]
A
We do believe the specialist versus generalists, but I believe the generalist is still a form of specialists that can just do more. Right. I think if you look at the example for weavy, you can have a creative team that is very small but they have deep domain expertise but they know the tools and then that means.
- [32:02]
B
They can't have non creative, you can't have non building network crazy image based prompts and everything. It just does not.
- [32:08]
A
I don't think we're there yet. I don't think there yet is good enough yet to just maybe not.
- [32:12]
D
Yeah, yeah. Something I believe in strongly is there's two types of generalists in the AI age. There's your tastemaker and your engineer. Both of those people within a company. It's true. At Gamma we have people who have just amazing taste. It's very rare to find the tastemaker engineer in the same Persona.
- [32:27]
B
And they're amazing people.
- [32:28]
D
And they're the people I would point to. They're just like this person is incredible. But they're all using AI. You know our creative director, I would describe her as just peak tastemaker. I mean she has amazing taste and she's doing animation mid journey and cling. Right. And then we have our engineers who are just doing complex growth analysis. They're pulling together a ton of tools and they're generalists. Right. They range across a bunch of different types of what would previously be considered to be like specialty channels or functions within marketing. But they have their top percent at one of those two things. And so that's something that I look for. And do you have that balance in your team as well?
- [33:03]
A
That's the perfect description. Where so marketing has traditionally been art and science. And you take those two things. What AI does is put you on either end of the spectrum to be really good, you have to be either incredibly good at the art part which is the taste, or incredibly good at the science, which is the engineer. And then the unicorns can sit in the middle. But AI means the middle part where you're kind of just average at any of those things makes you kind of redundant. And that's how I think about it.
- [33:25]
B
This gentleman up here, did you have a question?
- [33:27]
G
I guess on the go to market side it was interesting. You know, I guess a lot of his founders trying to get as many customers as we can as quickly as possible. I think you were mentioning there that HubSpot's trying this new agent where it does the sales cool for you. I think my only hesitancy with that. Do you generally believe that would take over a sort of good salesperson? I think what I'm trying to say is that there's only so much resources I have on the go to market side. Like can I get a prospect enough people, can I get into enough meetings? What is that best agent? If there is one that can help me out. But I do technically feel that for those first calls they're the most important ones that I should be on. It's almost the. You know that.
- [34:11]
A
Yeah.
- [34:11]
G
That when you're really getting down to that questioning a human psyche still I still think to actually sit in Front of someone and understand them is the best way.
- [34:20]
A
Yeah, yeah, no, I think it's a valid point. And part of that experiment was with the sales reps want their first interaction with a new prospect to be through an agent. Let me give you some, some things there. I, I think if I was a founder and I was trying to think through where do I deploy AI to? There's two parts AI. There's how do I use AI to acquire more demand and how I use AI to monetize more demand. I would say AI today has a lot more capabilities to better monetize demand, where it's actually disrupting how you get demand. If I was a founder, I would use AI as an inbound prospecting tool. Like it's really great for someone has come in, shown some sort of signal, and then you can grab that data and do some prospecting to them. Ibound is a little harder where that person is completely cold because it's getting saturated. To stand out, you actually need to have really great unique data sources. The part that I'm talking about is more like I wouldn't think of it as replacing the sales rep versus for us, because we have such scale, we are trying to get signal of how to make sure your call is the best call for you. So you still opt in to talk to that agent, but you still have the time to talk to the rep. So the example I would give is like the person books time to talk with the rep in two days time and we say, hey, in the meantime, would you want to speak to the agent to have a conversation so we can make sure that calls is enhanced for you? And there's a tip that we got from Shopify in terms of a UX pattern that really works for AI in that as long as you commit to giving that person the thing they wanted. The example Shopify used was it's really hard to get people to use AI for support because they remember the workflow chatbot days where it sucked. Right? And so Even on us, 70% of all chats on our chat go human. The first word they say is human because they know the AI is going to talk to them. And so similar to what we talked about earlier, Gamma's got like way better in a year. The AI as a chat agent has got way better in a year. So people don't realize that. And the tip from Shopify was just add a little wait time, right? To say, okay, we were two minutes before you get connected to the human. In the meantime, do you want to talk to the agent. It's the same kind of pattern we're using in sales where it's opt in, we don't force it on you. But it's much more not selling. And I have lots of like experience trying to get AI to sell. It's much more discovery, like just a conversation to extract what you want from that sales call.
- [36:32]
B
Okay, we're going to do one question here and then we're going to do a quick fire round of biggest prediction for next year. So everybody up here be thinking, well he has.
- [36:39]
C
So regarding this onboarding you mentioned like you tried that in Zap when you.
- [36:44]
B
Were there and didn't work.
- [36:46]
A
It didn't work but they have since done it in the last three months because the reason the models are much better.
- [36:51]
C
So question is like in some cases.
- [36:53]
B
You have some UI based actions rather than API based.
- [36:58]
C
So if you have some kind of both like UI plus API calls to.
- [37:04]
B
Achieve the goals for onboarding.
- [37:07]
C
So do you have any recommendation like any tools out there which are blend these in nice smooth way for you.
- [37:15]
A
So you mean like you have the conversational UI but you still need kind of traditional qlik click UI to do some of the things? Yeah, yeah. So I also believe that's true. I actually don't believe conversation UI is the best UX experience for all AI apps. I think in some cases it's easier just to do traditional click click, click ux. I know I give you an example of a tool you can go try that I think has an incredible experience blending those two things together, which is Max, who is founder of Gummy.
- [37:41]
B
I like that you call it Gummy.
- [37:42]
A
Yeah, but that's the agent. I love that product. I think he's done an incredible job of blending both a traditional UI interface together and like a conversational ui. So if you're looking to see like an example of an app that's kind of blended those two things together in a zapier style. That's why I'm using them. They're in the same market. I think they are a good example of that.
- [38:01]
C
Thank you.
- [38:02]
B
Okay, we're going to do a quick rapid fire before we end the show of biggest AI prediction for 2025. Kristen, can we get stuff? 2026. Sorry, yeah, I want 2025 to get the hell out of here.
- [38:14]
D
I'll probably be pretty good. I could probably be pretty good at that 2026. So I think one of two things is going to happen. One, either we are going to see such a change in progress in LLMs and none of this is going to matter. We'll have all new predictions and playbooks. I think when we get lazy. We talk about the AI age today. Right. But we're really talking about post 2023 ChatGPT coming out. Progress has been a bit slower this year. Right. I don't think we've seen that same type of like change from what it felt like before there was chatgpt to chat GPT. If we have that type of improvement again, just imagine how much everything is going to change regarding what we're talking about. But if that doesn't happen in 2026, then what I'm seeing paradoxically in the AI age is you have both people wanting to use AI more and also wanting to be human more. And so we lean a lot more into experiential events. We nail a lot more into influencer creator marketing. I'm seeing that start to ramp up more with other companies this year. I think it's going to get even more intense next year as well too.
- [39:08]
B
To bring back the human element, especially go to market.
- [39:11]
D
Yes.
- [39:11]
B
Is your human fiction. I love that grant.
- [39:14]
C
I'm gonna go somewhere complete.
- [39:16]
B
I love that. Yeah.
- [39:18]
C
We're gonna talk about autonomous vehicles. I think next year, autonomous vehicles, the growth rate's gonna start dramatically increasing such that it'll be. When you zoom out and look, you know, look back, it'll be the sort of inflection point and we look four or four or five years into the future, the percentage of autonomous vehicles will be greater than human driven vehicles and traffic is going to dramatically reduce because of that. And so we'll be able to get from point A to point B much faster because cars don't have to worry about hitting each other, worry about being bumper to bumper. And so hopefully I'm maybe just trying to make sure that this happens in San Francisco because it's a pain in the butt.
- [39:53]
B
I was going to say you get around it. Autonomous vehicles.
- [39:55]
C
So I really want Waymosis to take off. And so that's my prediction.
- [40:00]
B
I love, I love that one we've had.
- [40:01]
C
These are fun.
- [40:02]
B
Kieran, what do you got?
- [40:03]
A
I'll get to. First one is somewhat random, but I think because the open source Chinese models are so good, it is going to put pressure on all the kind of conglomerate US models to do something rash. Not rash, but to do something aggressive. And I think Google might put out Gemini 3, maybe launch it for free, lean into the fact that they can monetize that through paid advertising and force ChatGPT or OpenAI to actually have a free paid ads model. I think that's coming anyway because that's where they hired, I think the Instacart CEO. So I think ChatGPT actually have a paid model in their freemium tier will happen. I think they're going to have to do that too. Keep up with the fact that Google has such distribution leverage. The other one is this was meant to be the year of the agents, right? Like that was the whole thing. You're the agents. You're the agents. I will be honest. I think that AI is still not consistent enough to have true autonomous multitask agents. And I think 2026. The reason the models make it much, much better at being consistent, so that an agent can do three or four tasks repeatedly, which they really cannot do right now. It goes off the rails. And so we might actually see that kind of year of the agent come to fruition next year. Because from all the kind of signals, Gemini 3 and these models are going to be a step function better.
- [41:14]
B
These are all really good. I'm going to give you two real quick ones. The first that I said a couple of weeks ago, the podcast that I will repeat here, massive AI backlash is my big prediction is that this gulf between AI and Main street is going to cause a bunch of conflict and there's going to be a big wave of anti AI sentiment and there's going to have a whole host of implications in this market and the market you all work in. And I do think ChatGPT is going to launch ads. And I think if you're a marketer or you're a business, that's going to be a really great and valuable tool. And I think that will be one of the more interesting growth levers and discussions next year will be ChatGPT ads, the last one.
- [41:51]
A
At some point there will be a launch where us Europeans get it at the same time as you us.
- [41:58]
B
Change your regulations. I don't know. Okay, so this has been awesome. Thank you all for being here. We will see you tomorrow at Grow. Thanks again for joining us on this special live episode of Marketing against the Grain. Thanks everyone.
- [42:21]
A
Sat.