From Vibes to Workflows: How AI Is Rewiring Modern Marketing

0:00They are here to be productive, creative, but they don't want to prompt all that stuff. find which model, like which model do you use? Oh yeah. Oh, Kling is great for skin. VO is great for product. So long story short, we believe at the brief that we can have an environment where you can basically create things, integrate all the models, pick the ones that are best for specific use cases, and do this intelligence on top of AI. Welcome to the Talking AI Podcast, where we talk AI with both experts in the field and early adopters. I'm your host, Matt Page, and we're here to demystify AI for you so you can get some value from it.

0:41Let's talk some AI. AI has unlocked insane creative power. We get text, images, videos all generated in seconds, but for most marketers, it still feels disconnected, a bit random, and frankly, overwhelming. And my guest today, Thibaut Embrett, Chief Product Officer and Growth Officer at Debrief, wants to unify that. And he's led growth at Adobe and GitHub. So he's seen a lot. He's had a firm receipt to where AI marketing workflows are heading in this new age that we're in. So we're talking about Vibe marketing and how to turn what feels like a random act of vibes into a repeatable end-to-end workflow.

1:18But welcome to Talking AI, Fatima. Yeah. Thanks for having me. So excited to be here. Yeah. This is going to be a good conversation. We have not hit the Vibe marketing topic yet. We've done a lot of the Vibe coding, a lot of the AI tools, a lot of agents and everything. and that comes into play here too. But I'm curious, just like vibe marketing, the vibe coding, the term vibe marketing means a lot of different things to a lot of different people. How do you define it? How do you think about this movement that's happening? Yeah, so as a developer myself, I think I do vibe code and I think vibe coding is real.

1:50And it's just super interesting to see what's happened with code generation. And I think what's happened, what's going to happen is we're going to see the same disruption that's happened with code generation happening with marketing. Because, and when I worked at GitHub on Copilot, when we realized that LLens can generate, they're basically using text token and they can generate code, that's the first thing that happened. Like the first disruption and wave of innovation has been code gen. Because it's an extension, it's just a conversation, code is text. The next thing that's going to happen is the models, the video image models are getting close to perfection to generate image.

2:32I would generate anything with AI for, let's say a person or a product placement. Six to nine months ago or a year ago, you would have been like, this is AI slop. This doesn't look good. Now it's getting harder and in some cases it's going to look perfect. So I think what's happening is just a question of time and developers that now at the brief, everyone uses cursor to write code and supplement their brain. The same thing is going to happen with marketing. Now, the thing that's different is that developers to get support or to generate something, you can have a conversation and you can ask. With marketing, you can do it.

3:17But with marketing, when it comes to generating something visual, it's harder to express it with words. When you want a concept to be, when you say, hey, I want you to fix this function. I want you to grab this, call this API, do this and fix my SSO talk, whatever you ask in code. It's pretty simple. It's pretty natural. But when you want to express that you want a kind of like a diffuse background or you want something of like vector style or you want something, you have to be a professional designer to express that to the LLN. So that's why the first thing that we're seeing is that a lot of marketers say, I don't want to prompt.

3:57And not just marketers, a lot of designers that we've talked to at the briefing customers or prospects tell us, yeah, of course I prompt, but I don't want to because it's I have to and I do, but can you actually help me not having to prompt? And so I do think that vibe marketing for me is designing things at the speed of thought. Designing things so that you can just express something, show something. And I do believe in show rather than tell for designing because I'm sure it's the same thing you've seen in your life. When you don't know how to design and you want someone to design something, what do you do?

4:4199 % of people will say, I want something like that. I want something that looks like this. You know, you're going to go on Pinterest. You're going to save something. You're going to say, oh, I saw this logo. I took a picture of this shop. I really like it. Can you design a logo for me like this? Very few humans are going to be like, I want something like Russian 1944 style, like a vector bay, whatever type of designer you can think of. That's just not how people, most people want to design, except if you are a professional designer. but even as a professional designer and same thing for a marketer, I think the power of live marketing is you just show, you bring a bunch of images and you say, give me something like that.

5:25And it starts generating something. And you say, yeah, but make it more like this. And it's generating. And then you start just like sculpture, creating something, except that in the past, you had to know Photoshop. Quick break in the pod. If you're listening to this podcast, chances are you've been thinking about how to actually use AI inside your business. And that's exactly why we built the AI Opportunity Finder. It's a free tool that helps you uncover high impact, tailored AI use cases based on your business, your goals, your pain points, and your industry. No fluff, no generic use cases, just real ideas that fit your business and they're ranked by ROI potential.

6:02It takes about three minutes to run and it's like having your own personal AI strategist for free. If you wanna try it for free, check out the link in the show notes or go to hatchworks.com backslash AI dash opportunity dash finder. And now with Vibar, you don't need to know Photoshop. You give me about 10 rabbit holes I want to jump down, but that's a really interesting point there. And I do believe it's like the experts with AI have the most power to do amazing things, but that's such a nuanced point there because I experienced this too. Like I want some type of designer thing, but not knowing how to describe it in terms of how the LLM is going to understand, that is a gap.

6:45And to your point, if I can just show, hey, like this thing with a mix of this, there's a ton of value in that. I think what's interesting too, I've noticed with marketers, A, you can do so much more, so much quicker. But while we're on the topic of vibe coding as well, I've even seen some marketers starting to do really cool campaigns with like vibe coded things. Yeah. Which they never would have been able to do before unless they could have bribed one of their developer friends on the team. And it's like bottom of their backlog. So it's like this new nuanced tool that these teams have to create really unique and compelling experiences.

7:23On top of just the standard copy visual video ads, all of these different things. The lines are blurred. Yeah. Product designers on my team write some code. The designers are able to now get to almost working prototypes that they ask designers to review and integrate. We have salespeople on our team that generate websites for Outbound. They generate the website with Lovable and they're like, yeah. And they put some ads and they use that as a presentation. And so the lines are blurred. And I think that's amazing with AI is that it's just lowering the barrier to entry. for humanity to so many things.

8:08Love created a new category with what they did. And so, yeah, it's fascinating. And back to your point of reference, having that lovable prototype is just, again, something you can reference to. I was looking back the other day, they turned one, I guess, since they officially became lovable. I've created a hundred different things. Yeah. But I have to go back to the point you mentioned earlier. You mentioned GitHub Copilots. And so you were there when that product came out and I believe you were leading or in some form on the growth side. What was that like seeing this totally new thing emerge?

8:41Yeah. And just marketing this totally new way of working. Any unique experiences from that time? I'll tell you, I got to be honest. Yes, I was working on the marketing and growth side, but that piece of genius, because if you look, if you listen to Cursor and the founder, that's when they saw what was possible with Copilot that they were like oh we got to create something because there's and it's but back then in the team that started that it started really as a very kind of like skunkworks project that few people knew in the company and then to be honest with you at the beginning I thought that was really cool but I and I won't speak for the product team because they had the vision to build that too.

9:28And that's when Nat Friedman was the CEO working with Microsoft. Nasatya had made that investment. There was just a lot of foresight in that decision. And the team worked on it for, if I'm not mistaken, like probably one or two years in their garage. I've been in many companies. It's so hard to stay focused. Like being able to just say, we're going to do this and we're going to work on it for a year and a half, two years, and then it's going to be awesome. And they did that. It changed the industry. And I'll be honest, I thought that was awesome, but I didn't anticipate that it would change an entire industry.

10:07Like now, like copilots, code generation is almost like, yeah, of course. So we started, it was, you write some comments and it generates the line of code or a few lines. And now it basically reviews your code, works in the terminals. It's crazy. We've seen the same evolution on the marketing side too. So it used to be, okay, write this ad copy for me. And it's very one-to-one interaction with ChatGPT. But to your point, it's completely evolving. You have videos that just are indistinguishable from reality. But to your point though, it still feels random. And I think the point you mentioned earlier was going from just having a single prompt box.

10:47And then the other thing too, I experienced this working with this tool and that tool And okay, I'm going to take, you know, what I've got from ChatGPT over into VO3 or whatnot. And then, okay, I want to create a, I like the style of the image in mid journey. So I'm going to do that and then feed this over here. And I think this is what y 'all are building in a sense at the brief is something more cohesive. So where do you see that evolving? And that could be what you have today at the brief or even into the future in terms of these like workflows and processes that are starting to be established.

11:21And I can show you in a second a quick demo, but you're absolutely right. If you watch any video on YouTube on kind of AI marketers where they show you how to use things, these folks have five tabs open or six or seven, 10 accounts on File.ai and FreePick and whatever and Figma. And then they're jumping between. They're like, okay, so I'm going to go and create a prompt. So let's upload that picture in chat GPT. And let's generate a prompt. Let's copy that. Now we're going to put that in that spreadsheet. Now I'm going to go to file.ai. We're going to generate a video. We're going to try six times.

12:02And you're like, if you are a nerd, you enjoy the craft of that, like a developer. But marketers, they are here to be productive, creative. but they don't want to prompt all that stuff. Find which model. Like, which model do you use? Oh, yeah. Oh, Kling is great for skin. Vio is great for product. So long story short, we believe at Debrief that we can have an environment where you can basically create things, integrate all the models, pick the ones that are best for specific use cases, and do this intelligence on top of AI where if all you're doing as a product is exposing a model in a prompt window, then Gemini can do that.

12:50You're not adding value. But what we do at Debrief is we, and I'm going to share my screen.

13:00Can you see this? Yes, we can. All right, we have this, for instance, this canvas that's multiplayer, where you can share that with your team. You can generate things here with this prompt. and you can basically do vibe marketing. So I can select that element, select some text. And for those that are like just listening right now, like we're looking at this, I would equate it to like a Miro style, to your point, canvas. And we have these different images and prompt boxes and things like that. And now you're generating something off of this prompt in relation to one of the images. So describe it for those that are audio only as well.

13:42Honestly, I can select anything on that infinite canvas. I can generate things. And as it's generating, I can show you what it's built. So here I have, for instance, I just asked to put these piece of clothes on these models. And I did that in... Look at the quality. I mean, if we were to describe this for the people that are listening, it looks like a professional photo shooting that happened in a few seconds. I can change anything. I can say, change the position of the body, make it add, have her wear some sunglasses. If I want to resize, which is the plague of any designer, hey, can you create, can you resize?

14:22It's not crap. We can just do that. So you can just resize. But then if you want to introduce new elements such as ingredients, I want to add a little doggy. I can just select that dog and say, add a dog to the scene. That's cool. And then you have that. And now the thing that we've built on top of this is what we think is really valuable is that if you want to turn this into a video, here's how you need to do it today. You're going to need to generate a prompt. So most people, and I don't think prompt is the end of the road, people will struggle to even, what is the prompt I'll use? to create a video.

15:06People have no clue. So let's imagine. Oh, so you need to go to JGPT and do that. You can just click here, remix. You can say image to video. And we will analyze your image, understand that it's December, understand that this is a photo shooting, and we will propose some effects so that you don't need to prompt. You'll just click on the effect. I'm going to say gentle, let's say motion, sway or floating. and I will generate that video and that's it. Yeah, because the video prompts are very nuanced, right? They're very nuanced. There's an approach to how you prompt it, right? Exactly. And the models, what we've seen is that they work better.

15:51The technology is not perfect yet. So if you go with a movement that's extremely strong, you're going to see hallucination. But what we do is we actually specifically ask for things that are more subtle so that it's more likely to succeed and look natural and be something that you would want to use for such a campaign. And if you wanted to generate more variation, instead of coming here, putting a prompt that says, hey, add some plants in the back or whatever and enhance your prompt, you can just click on the image, generate variation, say very subtle or strong, and then it will generate variations automatically.

16:37Right. And that's the cool thing too, because I feel like what's neat about this is like, you can almost use AI as another creative partner. Because I think a lot of people, they're in this mode of, okay, I have to tell AI exactly. I got to dictate exactly what I need to do, but sometimes you can almost let it go down its randomness, which will then like bring up another idea in your head, which will be like, oh, that I hadn't thought of that as like a campaign or a concept. Uh, I think that's an under just utilized way of leveraging AI and it's elevating it to this like status of a thought partner in a sense.

17:14Exactly. And so just one last example, like this is an example with one of our customer. Oh, I love these. Yeah. Are these chocolates, chocolate satas? Yeah, chocolate satas. Who doesn't like chocolate? We're so proud to have Lint as a customer. And they use AI and look at the quality of this, the scene. It's just insane. Text is now possible with image models. And you want to generate 20 variations of that. You can just click generate variation and it's done. And it took me 10 seconds to generate these. And so that is vibe marketing. That's the power of really integrating that technology into a tool that understands the marketer's workflow.

18:01It is a big point there too on the Lent side is a lot of the stuff you see out there on YouTube and whatnot, it's very cool. But if you're talking about a real brand like that, the consistency of their product in the ad is a non-starter if it's not consistent. And what you just showed, it was the exact same chocolate Santa, right? So there's no variation there when you don't want variation, which is critical for large brands. Now, I wanted to say there's a lot of cool demos with AI that are acceptable when you're an SMB and you don't have the same reach and brand, but for a Lint or a AstraZeneca or some other customers we have, they need precise fidelity.

18:48They say, no, the logo has to be eight pixel to the left here. This is the font that we use. So the image models are not there yet to be able to create something with the exact font, the exact placement. Maybe we'll get there. In the meantime, having this editor that we have where on the canvas, you can also open an editor, put your brand kit, put your logo things so that you can have access to that control is very important. Just like in cursor, you can decide to go agentic and have it generate stuff at any time. You can say, hold on, and you're going to start coding manually for two hours. Yeah, totally.

19:32I'm sure you've heard of this example recently, or maybe you haven't, but Coca-Cola, they put out an ad recently, it was a video ad and they talked about needing 70 ,000 prompts going into it. And a lot of people were using this as like validation that, oh, it's not there yet. The models aren't there yet. The quality, it's still AI slop and all that. And in my mind, and I'm sure you see them too, like the creators that have really learned this craft. Yes. They're doing insanely amazing things with much less in terms of prompts. I don't know what model they were using. That may have been a limiting factor as well.

20:09But I'm curious your thoughts on that, because I think there is, it's a gap in the approach you take, and then obviously the model under the hood that you're leveraging for the specific use case. Yeah. So let's be real, the video models, for instance, right? So VO 3.1 was a huge improvement, but back in the first version of VO and even VO 3, you had a failure rate that was, I can't remember, but basically it would take you still like a few times, a few like attempts to get the result that you want. So there's enough like an error rate that used to be higher. Now it's getting better and better.

20:46So there's that. The second thing is prompting and which tools. It takes a lot of work. What's interesting, and I'm curious your thoughts, I was thinking of, I think thinking of an analogy, I do think what this reminds me is I remember when I started back in 2006 to 2008, when the iPhone launched, there was a surge. Basically, the developers, the iPhone developers were the king of the world. If you knew Objective-C at that time, when the iPhone, when mobile became a thing, when people realized that they could have actually the web in their pocket. I don't know if you remember, but like mobile developers, mobile app developers, like it was huge.

21:35And if you had that skill, you had a huge advantage. And then later, Objective-C was replaced by Swift. But the people that learned that stuff back then had a huge advantage. And you can see a lot of these people that started and became sometime founders or started companies. or like they had some level of insight. They just knew how to create application and businesses on where people are going to be spending most of their time today. And I think the same thing is happening right now. If you know what the models, how they work, even if you use a lot of manual things, people will want tools that help them do things faster.

22:18But that way of thinking of how you construct things, how you use the models together, or how you use agents to supercharge your productivity. I do think the people that have learned that right now are going to have a huge edge in the next five to 10 years. Yeah, I totally agree with you. I think it's, I always go back to the Shopify CEO's leaked memo that he just put out there. And he says, AI, generative AI is just too unlike anything else. And the best way to get good at it is simply by using it a lot. Yeah, it is those people that are experimenting using a lot because they, they understand the nuance of working with the thing.

22:58And they're also the best people positioned to create these products and experiences to make things easier, but also adapt things as it evolves. I completely agree with you. And I think another interesting thing, I see this on my team where, especially in the marketing realm, people had very specific skillsets. This is my copywriter. This is my designer, videographer, whatever it is. But I'm seeing this blending of skillsets to where there's a lot of overlap, but it's in a good way because now my designer can like propose copy related to something and my copywriter can propose image ideas. There's this overlap that creates this new interaction experience between them.

23:40And A, they just move a lot faster, but they're able to, I think, communicate what they're thinking in a more nuanced fashion. Kind of curious, like if you've seen it. Yeah, yeah. You remind me of Adobe, the founders, the co-founders, Warnock and Geschke. They had this quote in the SF office where they say, good ideas come from everywhere. And it's awesome. It's a really beautiful principle because it is inclusive of people contributing. And like you said, historically, what prevented a copywriter to design something wasn't their creativity. It was just the fact that they didn't know the shortcuts in Photoshop or Figma.

24:22And so we prevented a lot of people that constitute humanity to contribute to conversation because they didn't have the skills. And so I think that's the thing that's the most awesome with AI. You have this dentist that's using Replit to create apps for, I can't remember the story, but I thought that's just awesome. Lovable has created a new category with people like, yeah, I can do this. And same thing in marketing teams, the copywriter might have some really cool ideas, but they were never able to express them and now they can. Yeah, totally. And I'm curious, one thing I think that's really unique and interesting about your product is you have one of those products that gets better as the models get better, which I think is a huge indicator.

25:09If your product gets completely overtaken, once the models get good, you're in a bad position. Right. But yours is uniquely suited. Nano banana pro coming out via 3.1 with SWORD 2. And the competition's just fierce. So it just, they keep getting markedly better. Where do you think we're going to be in like, say a year from now? Because I feel like each new iteration is just very impressive. Where do you see the biggest gaps that we're going to see solved in the next near term future? I think the, so I think on, on, on the, let's start with, So first, yeah, it's amazing to build and fly the plane as you're replacing Engie every two months.

25:50And so it's kind of amazing. And there was this interview from the CEO from, what was the name of this company? They were competitive to Cursor. They got acquired by the Devin. What was the name of? Windsurf. Thank you. The CEO of Windsurf was interviewed by Y Combinator. And he said something like, you have to be comfortable building a product that's 90%, 80 % done. and 20 % is still not working because the models are not there. So it's super interesting in the implications for product development. On where it's going, I think for image models, the thing that is still not perfect, but Nano Banana Pro got very close.

26:30And I think right now, Nano Banana Pro, from a marketing standpoint, I think overall is the best model, is text. So text rendering, for instance, for lint, very important. you cannot have the chocolate with not the right lint logo on the chocolate. They even have lint engraved in the chocolate chunk, the pieces. Nano Banana Pro is able to do that today. That's impressive. So I think the next version of Nano Banana Pro, I think will probably crack that and be perfect with text in all scenarios. I think they're getting really close. So I do think in six months, a year, text packaging and all that, cracked, it's working.