What Startups Are Using Now Will Define Enterprise
David Levy · January 30, 2026
venture capitalstartups
Welcome and Introduction
(0:05) Welcome to another edition of AI Insiders by AI Marketers Guild and Market Media. Great to have you all here. I'm here with an old friend. We're old friends now, aren't we, David?
(0:16) I think we're old friends.
(0:19) 10-15 years is plenty.
(0:22) We got to know each other when you were running
Social TV Startup and 360i
(0:32) that social TV startup. I was at 360i. I remember grabbing lunch around the Tribeca offices there. I can picture the exact time and place of that era and some of the weird but super cool things you were trying to do.
(0:56) It was fun. I know back when social was scalable but also weird.
Understanding Tech and Markets
(1:03) Not just eyeballs. That was fun. Being one of the people I've stayed in touch with better than others, I get to have an ongoing check-in to understand what's going on with tech, what's going on with markets. You know way too much about a lot of things. Since I've been able to learn so much from you, I figured
David Levy's Humorous Self-Description
(1:27) it's time the community gets to do the same. Welcome.
(1:30) Thank you, David. I consider myself great at cocktail parties. I can talk about many different topics, but only so deep, then I have to run away. Or a decathlete, which you've probably heard is someone not good at 10 things. Those are my dad jokes. Dad jokes over.
Entrepreneurial Journey
(1:51) The quick background is I started as a banker and investor in growth in tech for 10 years. I left to start two companies because I thought that would be more fun, more lucrative, more interesting. I was one out of three on that. The first company was a retail company.
(2:14) The second was the social TV company. We were white hot for a hot second. Of the early social TV companies, we had the highest dollar return for our investors in that I only lost a million dollars, while my competitors lost $10 to $20 million. Tada. After that, I did some
Corporate Venture and AWS
(2:40) EIO for a couple of corporate venture funds. One was Lauren Michaels Cooper Venture Fund. I never met him, but I spent time in his office and some interesting characters would walk by. I was the e at Comcast Ventures for a year, then fell into a spot at AWS where I co-built the startup engagement team
AWS, Stripe, and Rapid Growth
(3:04) starting in 2014. I was there for the better part of 10 years. I left halfway through that. I did the same thing at Stripe. I stayed there for a year and a day. It wasn't for me. I went back to AWS. Our group had gone from 14 people when I joined to 150 when I left the first time, and by the time I got back it was in the thousands, listening to a sales
Focusing on AI Infrastructure
(3:28) team. Many of us, overpaid ex-entrepreneur venture capitalists, overstayed our welcome. I left that world two or three years ago and have been focused on AI infrastructure, digital infrastructure, from chips to cloud infrastructure and other infrastructure, to optimization
Shifting Tech Adoption Signals
(3:58) tools on top of that, and playing around with some of the apps we're seeing now. David and I have always talked about how it used to be that when I was a Wall Street analyst, earlier stage companies would use whatever they used, and then large company CIOs, CTOs would dictate the
(4:19) enterprise software applications and infrastructure they would use. That's where you got your signals.
Startups Lead Cloud Adoption
(4:19) That flipped. From my perspective, it flipped right around when I was at AWS. The numbers I'll use are functionally 100% of startups were using cloud, but only 10%
(4:44) of enterprise IT spent on cloud. That changed dramatically.
Stripe's Early G Suite Adoption
(4:44) I went to Stripe. The weirdest thing in 2018 when I went to Stripe, and Stripe is a very odd place, was that they were using G Suite as their internal email system. This doesn't sound weird now, but then for a multi-billion dollar
(5:05) company, it was very strange.
Startups Predict Future Enterprise Tech
(5:05) What we started to realize, and what I've realized recently, is that you can tell what enterprises are going to use two, three, four, five years from now by looking at startups. I don't necessarily mean startups using startup products; sometimes it's startup products, but not always. The reason is not so
(5:34) much that startups are brilliant, which I do think they are, but it's out of necessity.
Startup Tech Advantages
(5:34) One, they have no technical debt, and two, they don't have options but to use platforms that get up and running quickly.
(5:50) What does that mean today? What's popping up on marketers' tech stacks now? What are startups using?
AI Models Before Cloud Infrastructure
(5:57) You tell me. I can tell you starting with tools, I'll call them infrastructure tools, which lends itself well to what's happening in the marketing world. A piece came out from Business Insider, my old group there. It was a leaked memo. I had nothing to do with it; I was long gone
Startups Prioritize AI Models
(6:24) point. It was a bunch of stuff, but the key point I saw was that for the first time, startups were choosing their AI models and tools before they were choosing their cloud infrastructure provider. Kind of a boring statement, but having spent the better part of 10 years at
(6:49) AWS, that was a monumental change.
New Era of AI Tool Adoption
(6:49) Because we had been so early, and in every company so early, we could influence how adoption of other things went. Now you're seeing something different. Companies are choosing whether they're using OpenAI's API,
(7:14) Cloud Code, or their own GPUs or GPU infrastructure. You're seeing that happen now before they start to choose other platforms. The reason I think that might be important for marketers is that the tools people are using today, they're likely choosing tools first that they
(7:39) wouldn't choose before. I'd love to hear what you guys are seeing.
Startup Automation Trends
(7:39) In general categories, it's customer service that first comes to mind, or maybe second. The first is cloud code, but that's less to do with marketing. Anything on customer service, anything on sales
(8:03) automation, that's what I'm seeing first. It's because startups can't hire 100 salespeople. They can't hire or outsource to a call center. Anything agency or agency-adjacent, writing copy, automating outreach.
(8:32) With things like sales automation, marketing automation, CRM, are there things like what you saw with the Google suite that are standing out as go-to options?
Shifting Preferences in Marketing Platforms
(8:32) It's a little hard to tell
(9:00) right away. What I don't hear as much as I used to is people jumping on to HubSpot so quickly, or Clavio. Presumably, there are platforms out there
AI for Marketing Effectiveness
(9:26) that are doing this much more effectively, or using artificial intelligence to make it easier. Have you seen any of those? I can tell you what comes up when I hear names, but I'm curious what you're seeing. A lot of it has tried to tap into the
Clay's Popularity in CRM
(9:50) AI tech from existing tools, seeing if that can get folks further along. On the CRM side, Clay pops up a lot.
(9:59) Yep. Clay is one for sure.
(10:02) On company platform types of things, Clay for sure, but that's been going on for a while. If you ask most enterprises what Clay is, they'll say, "Oh, my
(10:15) kid plays with it."
Modern HR Platforms for Startups
(10:15) Having no idea what this is. This isn't the marketing side, but every startup I know uses Rippling or something similar. Not ADT or some platform that has a Windows 3.1 interface. You're seeing the Salesforces of the world add all this functionality,
Salesforce Complexity for Startups
(10:44) that's super cool, but if you're a startup, installing it... At AWS, we had a department that managed our Salesforce instances.
(10:52) Okay.
(10:54) AWS had a team, not just a person, a team. As much as they'll adopt AI for their existing customers and enterprises, what's going to
(11:08) happen with these newer companies is anybody's guess.
Llama's Traction and Open Source Models
(11:08) I think we can make some good guesses by seeing what there is.
(11:13) I was seeing some questions about Llama. Yes, Llama is certainly getting a lot of traction. There are some limitations to it, but the open-source stuff is free. You'll see many inference players moving towards these open models because they can't run the closed models on their systems.
Security Concerns with Open LLMs
(11:30) And ditto regarding the Chinese open LLMs; there are very significant security concerns, and they persist even on closed systems.
(11:52) For things like Meta, you don't hear about it as much, and what's gone on with Llama. Are there certain kinds of businesses adopting it more, and what are they mainly using it for? Llama itself?
Ways to Build AI Models: Building from Scratch
(11:55) Yeah.
(11:56) Or others?
(11:58) Anyone building their model, there are three ways you can build your own model. I'm saying that in air quotes for a reason. One is you can build your own model. That is a lot of work. It requires a lot of expertise. Perhaps most importantly, it requires an enormous amount of compute. That's when
Ways to Build AI Models: Retrieval Augmented Generation (RAG)
(12:39) you're going to a CoreWeave or even AWS or one of the other neo-clouds. Very expensive. That's number one. On the other end of the spectrum, you have folks doing these RAG limitations, which is taking in OpenAI or Anthropic. I forgot what it stands for. It doesn't really matter, but it's putting a shell
(13:06) around it and being able to... It's reinforcement augmented generative.
(13:11) It's really... Yeah.
(13:13) I'm supposed to know that, but I don't, and I'm not going to Google it right now because
(13:19) Retrieval Augmented Generation.
(13:25) There you go. Thank you.
Ways to Build AI Models: Open Source
(13:28) Was that clearly telling you that was said?
(13:30) That was Google AI.
(13:33) Oh, there you go. Nice. In the middle, you've got folks using open-source models. You can use an open-source model and tweak it to any dynamic you want. You still get a head start on
(13:52) building a model from scratch. You'll still have compute overhead, etc. I don't want to call them more serious engineers, but folks who need more fine-tuning of a model, but don't want to build a model from scratch, will begin with an open-source model.
(14:10) Yes, openrouter.AI. Holy cow. I see
Openrouter.AI: An AI Model Tracker
(14:15) in the chat, somebody is talking about a good tracker for AI model usage. I have been looking for the "Built With" or "Intricately" for AI models for a long time. I went as far as to try to build something that looked into website code automatically to see if it could figure out who's using OpenAI calls or Anthropic calls. I couldn't find it.
(14:45) I spoke to friends at OpenAI and Anthropic. They said, "I have no idea." But this openrouter data is just amazing. I was trying to build something to see early adoption, then I looked at openrouter and realized they did that. All right, onward.
(15:05) That's a really good way to see what people are doing with models right now.
AI in Marketing: Voice, CRM, and Outreach
(15:08) Things pop up and down, and it's wild to watch. Back to the marketing side, the early things I heard about, and I don't know how they're playing out, are the obvious things: CRM, cold outreach, voice AI stuff.
(15:34) I don't know if anybody has received an AI phone call yet. I think I got one. It's wild; you still know, but it's getting close enough where it's harder to distinguish. I'm seeing a little more traction on that in international markets. What I've seen is direct outreach, SDR type stuff, showing up first.
Identifying AI in Phone Calls
(16:08) Yeah. Those calls, once you realize this is not a real person, and it's not one...
(16:17) The pause. Some of those...
(16:21) They're not even AI, right? It's just a timer.
(16:35) Yeah. Yeah.
Voice Interaction with Chatbots
(16:35) It's interesting though. Have any of you interacted with any models or chatbots with your voice? Have you used ChatGPT? Have you used Bard?
(16:47) Oh yeah. Yeah.
(16:50) Even in some other
(16:58) applications of it, I've...
(16:59) Rosebud.
(17:02) I've mentioned Rosebud here, and David's the guy who introduced me to it. I love...
(17:09) I haven't used Rosebud voice-wise yet, but I use it text-wise every single day, at least once, if not several times a day.
Normalizing Voice AI Engagement
(17:22) David, you can explain Rosebud: it's a therapy app,
(17:25) but it's more a journaling type.
(17:28) Anyone I've shared it with has come back and said,
(17:34) "Oh my god, this is borderline addictive." I think the word you used was addictive. Ironic given its context. I
(17:47) realized this is marketing. I realized it was becoming easier and normal to use voice AI to engage when I was trying to create a logo for my advisory group. I was using ChatGPT while walking through the airport. I thought, if I put it on chat mode and put the AirPod in my ear, nobody's going to know who I'm talking to, and it nailed it. I've used it to prepare for discussions – not this one, I probably should have. I've used it to prepare for interviews. It's a very good way to use it. As we become
(18:40) more accustomed and okay with using those for our own use, I think that'll drop the barrier to when you get an AI call, as long as it's not spam. I think we'd get more comfortable engaging with voice AI as we use it more on our own.
AI for Post-Meeting Productivity
(19:04) in-person meeting, and now I'm full of ideas. I've got some follow-up things, and as I'm walking out, I'll just start talking. I tell ChatGPT or Gemini, and I just start. "Remember this, remember that," and then have a five-minute conversation so that when I's back at my
(19:26) desk, I haven't forgotten the things I wanted to remember.
Agency-Adjacent AI Tools
(19:26) The other place I've seen a handful of companies, I don't know the traction yet, is what I call agency-adjacent.
(19:42) Sorry, I missed that. I'll get to you in one second. But these platforms will build your marketing
(19:53) campaign, build the copy, and get you 80% there. It's wild. I know 2048 invested in one that was in Tech Stars New York called, I wish I remembered the name, I'll find the name. With an R, I think. I assume there's a handful of those. I just haven't seen what's gained traction yet and what hasn't. And if folks are going to work with agencies or if they're
(20:18) going to replace agencies or what have you. Earl, hands up.
Custom GPTs for Linux Exploration
(20:22) Sure. Thanks so much, David. One thing those who know me in the space know is I push custom GPTs all the time. Going back to your questions of use cases, using them through talking to one of your custom GPTs, one recent use case I've had was I've actually been
(20:44) exploring Linux.
Hands-Free Installation with Custom GPT
(20:44) since all the news with Windows 11 and people making migration because of the limitations. Last weekend, I set up Winnix on one of my machines, and I used a custom GPT. I spoke to it as I was going through the installation and configuration process. That way, I could keep hands-free and navigate on my new machine, ensuring
(21:07) everything was set up. I could literally ask questions when I ran into issues or had questions regarding any issues during the installation, configuration, or migration process. It's a great use case. I even used it once to help me clean the bathroom and make sure I didn't miss a spot. You never know the ways you can come
(21:27) up with ways to speak to somebody knowledgeable. It's great.
AI Content Creation Tools
(21:27) I'm going to look for the AI creation. I couldn't find that one, but I did find a quick list of things like Jasper. Figma is an obvious one. Predus. I'm even curious with some things Jasper is a good example where a client needed some SEO-optimized long-form copy,
(21:56) and I started testing Jasper and Writer and a few of these tools. This was before Gemini 3 came out, 2.5, I can't remember all the numbering systems. But I was asking ChatGPT and Gemini, "Can you do stuff with these very specific parameters, the exact word density, and all these kinds of things that SEO folks tell you"
LLMs Overtaking Point Solutions
(22:28) need to be there. They had someone who said, "Here's what the content needs to look like. Whether it's right or wrong, I didn't care. I just needed to do it." I tried Jasper, I tried Writer, then I tried Gemini, and Gemini was way better at following instructions. I realized, wait a second, I don't even need to pay for this extra thing. The thing I'm
(22:52) already paying 20 bucks a month for is better than a lot of what's out there. I'm curious where some of those areas are going to just be obsolete. I stopped paying for Ideogram, which I loved for image generation. Gemini, and sometimes ChatGPT now, just do it better. I still enjoy riffing on Ideogram sometimes, but some
(23:16) of these things, you might think they have their niches, and now the major LLMs keep expanding their pool of what could be done.
Seeking Alternatives for "Agentic"
(23:16) I think that's a really good point. I'm trying to figure out a way to start this part of the discussion without using the word "agentic." Can we come up with a better
(23:38) word for agent or agentic?
Defining "Agentic" and "Digital Worker"
(23:38) I'll use it for now. For what it's worth, my personal definition of agentic is when models are prompting models. But I'm not going to use that right now.
(23:51) I sort of didn't get it at first. I got it like, somebody's going to go out and do stuff for you. I used Manis, and I was
(24:01) an early user that paid them a lot of money. They sent me a t-shirt, which is terrific. And then I got acquired by Fadersburg. But when I asked myself why I was using this when I could use ChatGPT or Gemini... Yeah, digital worker. I like
(24:24) it.
The Power of Digital Workers
(24:24) Digital worker, not agentic. A chatbot digital worker, there we go, if it's able to more effectively utilize different models, etc., to figure stuff out and create stuff for me, it's better. When I have more involved things to do, I go to Mattis, and it is a much more effective platform for me.
(24:54) I just put in the chat that the Techstars New York 2048 Ventures-backed company was called ripple.ai (getripple.ai). I don't know if you've seen that, but now it's describing itself as a marketing automation agent. If you have these digital-worker-esque
(25:17) tools that can use the best of different models and different capabilities, and their own, you can have a 1+1=3.
Specialized AI for Complex Tasks
(25:17) I don't disagree with you that models keep broadening out. There's a company I work with
(25:42) called Polymathic. It has nothing to do with marketing; it's AI for science, for lack of a better term. It's a bunch of astrophysicists and other PhDs that built a thermodynamics model. Can ChatGPT and OpenAI build those types of models? Yeah, but those cases are not nearly as
(26:07) good. Their "special sauce" (I know we hate that term, but it's retro, so I can use it) is they have their own foundation model, and their agent can go and pick the best things to do for a scientist to more effectively complete a very complex set of tasks. I think that's where
Value of Agentic/Wrapper Platforms
(26:28) these agentic or wrapper type platforms need to be. They need to be in a place where they're going to unlock more value than using a single platform.
(26:38) Selena's wondering if "digital worker" instead of "agentic."
(26:42) Yeah, but I think that's an interesting distinction. The digital worker is that thing that
(26:54) goes out there and does things for you.
Agentic AI and Infrastructure Impact
(26:54) When I think of "agentic," the real profound implication for infrastructure everything else is when humans,
(27:26) prompting a model, asking questions, asking it to do things, that type of activity is going to grow. It's obvious: more users are using ChatGPT and other platforms more often. But when the models themselves that you're prompting start to do things and prompt other models, they can do that
(27:52) all the time. They can do it much more frequently and much more quickly. All of a sudden, the number of inference calls and token consumption goes practically vertical. Is that when Skynet becomes alive? I think that's where I'll make the distinction between agentic AI
(28:18) infrastructure and a digital worker.
"Digital Worker" Nomenclature
(28:18) A2A digital workers are becoming the nomenclature in Silicon Valley. Terrific. That means they do dictate our vernacular right now.
(28:38) What are other folks in
(28:41) here using? I started to put together a list of AI-native marketing tools.
Audience Check-in: AI Marketing Tools
(28:41) Is anybody using one already? Something two years ago they did not use but now they are using liberally? Whether it's for another area, tell me the other area. If it's in
(29:16) customer outreach, branding, copywriting, and it's not one of the models, we all know the models, what are the other tools?
Happenstance for Business Development
(29:16) The one I probably use most, I want to hear from others here, is Happenstance. It's more biz dev than marketing, but I'm on there all the time.
(29:39) Talk more about that.
Happenstance: Smart Network Search
(29:39) Anyone is welcome to connect with me there too. I just put the link in there. Happenstance is basically a smarter way to search your network or the network of a group of people you're in. I'm in one of these marketing agency collectives, and now we've got a group on Happenstance where we can search each
(30:02) other's networks. It's also so useful for me to say, "I'm looking for brand marketers at companies between 1,000 and 5,000 people that are in the D2C e-commerce space." I'll give it some of these criteria, and it can search whether it's my network or for people who have opted in. I use this stuff all the time.
(30:34) Often, one of the biggest use cases I also find it helpful with is for investors because I never remember who I know in my network who invests in what. Even if I know you're an investor, are you pre-seed or seed or A or B? It's very easy if you're aligned with a VC firm and there's a clear thesis they put out there. But many of the
(30:59) folks I know who invest are not part of an institution. So there might be some info they've shared out there that's hidden on their LinkedIn profile, that along with running this company, they're a pre-seed investor in agritech, and I sure as hell won't know that.
(31:22) Sounds like a next-gen Hashable, not
(31:25) the New York.
Synthetic Audiences for Content Creation
(31:25) Oh, I miss Ash.
(31:26) Awesome. Flame out. Yeah, I miss it too.
(31:29) Eric, do you want to share anything about Go Marble?
(31:31) I use synthetic audiences a lot. I don't know if that's relevant to what you're asking.
(31:37) Yeah, sure. Synthetic audiences I find very
(31:41) useful for content creation, creating surveys, adding the results to build content. They're extremely useful tools.
Tools for AI-Powered Content Generation
(31:41) I use Ask Rally and I just started using Navara, from Jill who presented. That's a little
(32:14) more complicated; I like to understand it, but they have features for generating content. With Ask Rally, what I do is create surveys. I ask the audience, they take the survey, then I take the output, put it in Gemini for example, ask for an analysis, and then I can use that to generate my content.
(32:45) That is awesome.
(32:47) Yeah, and the other thing I do is, for example, research this other tool I use for researching prompts, what people are searching for, then use that and Ask Rally to generate surveys, then use all of that to create content.
(33:09) Sorry, did you say there's something that tells you what prompts people are
(33:12) using?
AI in Marketing: Limiting and Avoiding
(33:12) Yes, I'll tell you in a second what that is.
(33:14) Yeah. While you do that, I can bring up another angle I've seen AI in the marketing space, which is the flip side: not how to use AI, but how to limit or avoid it altogether. I know there are many companies that
(33:36) ask, "Is your brand being used properly, etc.?"
Voit: AI Imagery for Brand Protection
(33:36) There's a company I advised for a while. It's now called Voit. I'll put the URL in here. Forgive my Techstars plug again, but this is a Techstars LA company. They provide a workflow to make sure that AI imagery doesn't end up in companies' brands. Where they became very well known was in
(34:09) things like Magic the Gathering. They have a big contract with Wizard of the Coast and video games, etc., to make sure people aren't sneaking AI-generated content into their actual IP. In those fan bases and groups, there are very significant problems if it's not original content. I think brand management
(34:37) for sure is something I've seen.
Prompt Research Tools and Resources
(34:37) The company is called Gumshu AI.
(34:39) Oh, nice.
(34:40) Of course. Yeah.
(34:42) As we're talking about things for researching prompts, the two I've spent some time with are Otterly and Passion Fruit, and those are pretty good. What I'll also share in the
(35:05) chat is I updated my own collection of resources.
Vibe Coding on Base 44
(35:05) Not everything discussed here is on this, but some of them are. Feel free to check that out and recommend others that should be there.
(35:14) I was about to ask you if you have that, and I'm glad you do.
(35:18) I have it, and I realized that with my vibe coding obsession
(35:31) on Base 44, I checked a couple of weeks ago.
Experimenting with Vibe Coding
(35:31) I had so many more credits for this month than I realized. It's funny, often there's this fear of running out of them, and I was being too conservative with it. So, I just started vibe coding everything. Dave, I put something in the chat that I didn't even show you yet, that I asked,
(35:53) "Could I actually vibe code an alternative to Rosebud?" I created "Innervoice Me." It's not great, but it works.
(36:01) Yeah. Well, I think that's another really great thread. Whether it's vibe coding or things like Lovable, which David, you and I have talked about. I know you adore it.
Marketers Using Lovable
(36:20) "Adore," I guess, is a better word. You passed Lovable.
(36:23) How are marketers using platforms like that? Lovable seems more targeted at folks doing things that are more marketing-
(36:30) Yeah, it's really good.
(36:31) heavy. How would you line up Lovable, Replit, Vercel, and all
(36:42) these thousands of others?
Platform Technical Savvy and Entry Points
(36:42) For me, I think one of the biggest factors is how technically savvy you need to be to use them. Base 44 versus Lovable, for me, I still see those two as the first entry points. You can get something pretty competently done even
(37:12) with the free trial. But as you get into other things, even with Lovable, it helps once you start asking it to do something more complicated. It's like, "Register for Supabase, register for GitHub," and it'll tell you how to do these things. Many other platforms are free, especially if you're not launching a major consumer product.
Base 44's Integrated Development
(37:33) For me, Base 44 is the fewest number of other things I knew to use. I just realized, for instance, that they have backend functions and agent mode that you have to manually turn on. But once you do that, originally I was using Form Spree that both Base 44 and Lovable recommended as an email capture thing to sync with.
(37:59) It was easy and cheap. Then one day, B4 said, "Yeah, I can do that as part of this. You don't need to go somewhere else for this." I've tried Replit, I've tried Vercel, I've tried some of Google's own tools. I really want to learn how to use Cloud Code, and I have it downloaded. I wrote this last week in the AI brief of using
(38:21) Cloud Code and Ghosty as the terminal and Netlify. I've gone down all of this, but to me, Cloud Code is harder for me to understand than the Korean alphabet, which I'm learning on Duolingo right now.
Cloud Code ROI for Developers
(38:27) A data point I'll give you is a twice-exited founder CTO I know is using Cloud Code. Many developers
(38:53) pay $200 a month ($2,400 a year), and he's getting the equivalent of $2 million in annual developer work done, in a week. He knows these numbers because he's run large engineering organizations.
(39:18) So when I hear things like, "enterprise adoption is so slow with AI," that may be the case, but at some point, when you have these ROIs that are orders of magnitude, it's too hard to ignore.
Blurring Lines: Development, Agency, and AI
(39:20) David, do you see the line between, now that you have all these, what we used to call no-code, now it's not even coding? These are like,
(39:45) "Hey, what do you want to do?" Are you seeing the line between what is development, what is agency, and what is developer blurring? Yeah. How does that all fall out? Who's doing what?
Vibe Coding for Faster Site Management
(40:00) I have a freelance developer who I use for all my personal stuff on Upwork, and I took my main site for
(40:11) consulting back from him. I vibe-coded it. I gave it to him. He put it on WordPress. He did all the things any smart developer is doing. I trust him. But then I realized, "This is too slow for me to tell you what I want done," even if WordPress has better SEO functions or things like that. I literally was
(40:37) yesterday on Base 44, telling it to build sitemaps for me, then I was submitting them in Google Search Console.
Vibe Coding Speeds Up Creative Projects
(40:37) These things mean there's a lot of quick-hit stuff I don't need developers to do. For more involved things, I vibe-coded a page for a content project I'm working on with a client.
(41:07) They didn't ask for this. They didn't ask for anything like this. I gave a few things they didn't want in the process. But I was able to bring this to life for them. I thought, "Okay, great. Now you can put this as a page on your own site however you want. I just gave you a whole vision for it and this functioning version." This speed to
(41:30) develop... I wouldn't trust myself or a typical marketer for a really important interactive thing that has to work. But there's all this stuff coming from the agency side. There were all these things that even just for the agency's own marketing, let alone doing client work, where it would be a big resource discussion. Do you spend
(41:56) months in meetings trying to get something done just for some quick hit, some wacky idea you want to bring to life, so you could go and, in earlier, less crazy days, you'd put it out in a tweet, and see if it bites? Now you can just do that in a couple of hours or a couple of minutes. That kind of
(42:21) stuff, for a creative person, or even an account manager, to run something by a client and say, "What about this? Can we do something like this?"
Evolution of Content Optimization
(42:21) That's way better than creating 50 slides about it.
(42:26) Yeah. As long as you talk about client work, I feel we went from, in
(42:43) Web 1, it was SEO, then in social media times, they refused to use the term Web 2. It was, "How do you get your stuff in the socials?" Now, David, I think you told me they were calling it Gener GEO. Whatever it's called, there are so many of these.
(43:07) Yeah. I read something a year or two ago, and I said, "Let's call it Seal Mode." It was SC AI. It was a joke. LLM optimization. It was a joke, but it seems like that's a thing. And this is where we're going, Daniel. I see your question about compensation models for AI-powered creatives. What
(43:34) if publishers or content creators need to have their stuff show up in models because that's where everybody's looking for them now?
AI Content Compensation Models
(43:34) How are they getting compensated for that? How do you think that plays out? I have some ideas. I don't know. You can see what OpenAI is starting to do now. It's
(43:54) some revenue generation, but how do you see that working out? SEO or social media optimization is, "Oh, you can get more stuff in, you get more traffic." This is not traffic. This is content. This is content generated on your content.
(44:19) Although it depends what business you're in because the trends we've seen are fewer clicks, but those clicks become more valuable.
ChatGPT's Ad Business Model
(44:22) A reporter was just asking me about the ChatGPT ad business model, and was hearing something about a $60 CPM, which sounds absurd, especially when they should be able to charge potentially obscenely high
(44:56) CPCs compared to what people are paying if the intent coming from AI is higher.
(45:00) So this whole business... Given how AI has seemed poor for general branding, not that it doesn't influence brand, but branding... Are you going to get more
(45:23) traffic from that? Are you going to get more engagement?
OpenAI's Token-Based Business Model
(45:23) Yeah.
(45:24) Yeah. My opinion, right or wrong, is that OpenAI's business model is ultimately predicated on selling tokens, not ads. I could be a thousand percent wrong on that.
(45:50) The markup on a token is 2x, so you have a 50% margin. I think they said something like 40% of their revenue is now from API access. It was a quarter. I thought it was 75%. I was wrong, but I'm getting less wrong. It's DoneKazam, which is nice. It'll be
(46:19) interesting to see what their ad model looks like.
Future of Marketing Technology
(46:19) I heard people talking about it yesterday about this new product, and they thought, "We know what the product is."
(46:27) Is it new?
(46:29) What do you say, if 20-30 years ago somebody was using Oracle for their CRM,
(46:42) and you say it got replaced, then now within the last 10-20 years, it's Salesforce. Platforms like that, if we had to look forward 20 years from now, what do the big names in marketing technology look like?
(47:10) What gets replaced?
Viability of AI Point Solutions
(47:10) Curious what others here would place bets on and are thinking, just beyond the rich getting richer. Manis was fascinating because it kept popping up as a favorite of many marketers. I feel so many of these things
(47:39) that have gotten some degree of traction. If you look at Jasper Writer, or Beautiful AI, or Gamma, which both have probably had the most traction on the "what's the new PowerPoint" kind of thing, but if Gemini, Google, and Microsoft don't just flat-out acquire those, they'll build their
(48:09) own versions of them into it. I'd be skeptical that some of these point-solution things are viable for the long haul, especially when anything is competing with part of the Google or Microsoft Office suite. Just because Microsoft tends to neglect user experience in Office today doesn't mean they're not looking at every single one of these
(48:35) and are going to... Yeah, and won't place a bet for the rest of the decade. Yeah.
Native AI vs. AI Add-ons
(48:35) Yeah. I said this to an EVP at Microsoft five years ago. They were thinking about their startup engagement. He reported as SIO; he wasn't really concerned about startups, but I said,
(48:59) "Look, you don't have to make another nickel in Azure, but you need to know what startups are using." This is the thread we started with, because this is what your enterprises are going to be using five years from now. Google is not a tiny company by any stretch, but I don't know the market share numbers now, but
(49:19) I would expect that G Suite has a material market share in internal email systems and documents. What I think was very different when cloud happened: Microsoft said, "Sweet, let's put Word on the web." Google, I think they acquired into it, said, "It's not just put some word processor on and give you web
(49:43) access. What does a truly native cloud-based word processor document sharing system look like?" I think the questions being asked now are not, "How do you add AI to Salesforce?" or "How do you add AI to Excel or Google Sheets?" but, "What are the native AI? What does a native AI analysis
(50:11) look like? What does a native AI look like if we don't start with the tools we use now?"
Reconsidering Startup Tech Adoption
(50:11) Yeah. Jamie, I thought, "Holy, that's big market share."
(50:15) I know your question was, "What are the big names in marketing technology in the next 10 years?" but
(50:30) what I'd like to do is revisit something slightly adjacent to what you said at the very beginning when you took an interest in what people are using now and that adoption curve explodes over the next five years.
Fleet DM: A Model for Remote, Cloud-First Companies
(50:30) I do M&A, and I advise mostly on the sell side. I'm in outreach now for a company I'm selling based in Paris.
(50:54) I discovered a company, I'm going to drop it in the chat, that is refreshing and unique: Fleet DM. The reason I discovered them is their ethos is 100% remote, no offices. They have a post office box in San Francisco. The company I'm selling is the same way; it's in emulation and virtualization. It's very high-tech
(51:23) stuff, but these companies are operationally aligned, totally cloud-first, distributed delivery, distributed management. To decide whether or not to contact them, which I did, you find out their CFO is in venture, and she's amazing, but she's not on site. The guy they have running it, who actually is running corporate finance
(51:47) over there, is a specialist in building remote organizations. If you surf around their site, it's completely transparent. It's so refreshing to see a company that says, "We're doing it this way because it's transparent. You can trust it. We can move fast. We're not encumbered by anything that's legacy." I think that's kind of what you were hinting at at
(52:14) the beginning when you were saying, "What's going to move fastest with cloud?"
Wrapping Up and Community Engagement
(52:38) Yeah. Jim, I was hoping you'd chime in at some point. You never disappoint. David, you don't disappoint too often either. You shortened today. So I appreciate you coming. We've hit the hour mark. It would be great for others to stay in touch with you and have you back sharing more ideas about this. A really fun community conversation.
Community Shout-out and Next Steps
(53:03) A shout-out to Daniel and Jean for the technical recommendations. I like how there are many of us at different skill levels on the marketing side, on the tech side, on all kinds of sides of this business. This community is welcome to one and all. Porch Capital, learn more about David Levy, and stay in touch
(53:27) with him. His Substack is really good. I often need to not read it the second it comes out because I need to sit somewhere where I can process it.
(53:35) I write long-form.
(53:37) That's a compliment. Follow me on LinkedIn. All this stuff. Next week, we have Ad Age's chief technology
Next Week's Guest and Closing
(53:45) reporter, Garrett Sloan. Another great observer of the market. You'll be able to ask better questions than I could think of. Thanks everyone for coming. Hope it's a great rest of your week, and see you next week in Slack and everywhere else. Thanks everyone.
(0:05) Welcome to another edition of AI Insiders by AI Marketers Guild and Market Media. Great to have you all here. I'm here with an old friend. We're old friends now, aren't we, David?
(0:16) I think we're old friends.
(0:19) 10-15 years is plenty.
(0:22) We got to know each other when you were running
Social TV Startup and 360i
(0:32) that social TV startup. I was at 360i. I remember grabbing lunch around the Tribeca offices there. I can picture the exact time and place of that era and some of the weird but super cool things you were trying to do.
(0:56) It was fun. I know back when social was scalable but also weird.
Understanding Tech and Markets
(1:03) Not just eyeballs. That was fun. Being one of the people I've stayed in touch with better than others, I get to have an ongoing check-in to understand what's going on with tech, what's going on with markets. You know way too much about a lot of things. Since I've been able to learn so much from you, I figured
David Levy's Humorous Self-Description
(1:27) it's time the community gets to do the same. Welcome.
(1:30) Thank you, David. I consider myself great at cocktail parties. I can talk about many different topics, but only so deep, then I have to run away. Or a decathlete, which you've probably heard is someone not good at 10 things. Those are my dad jokes. Dad jokes over.
Entrepreneurial Journey
(1:51) The quick background is I started as a banker and investor in growth in tech for 10 years. I left to start two companies because I thought that would be more fun, more lucrative, more interesting. I was one out of three on that. The first company was a retail company.
(2:14) The second was the social TV company. We were white hot for a hot second. Of the early social TV companies, we had the highest dollar return for our investors in that I only lost a million dollars, while my competitors lost $10 to $20 million. Tada. After that, I did some
Corporate Venture and AWS
(2:40) EIO for a couple of corporate venture funds. One was Lauren Michaels Cooper Venture Fund. I never met him, but I spent time in his office and some interesting characters would walk by. I was the e at Comcast Ventures for a year, then fell into a spot at AWS where I co-built the startup engagement team
AWS, Stripe, and Rapid Growth
(3:04) starting in 2014. I was there for the better part of 10 years. I left halfway through that. I did the same thing at Stripe. I stayed there for a year and a day. It wasn't for me. I went back to AWS. Our group had gone from 14 people when I joined to 150 when I left the first time, and by the time I got back it was in the thousands, listening to a sales
Focusing on AI Infrastructure
(3:28) team. Many of us, overpaid ex-entrepreneur venture capitalists, overstayed our welcome. I left that world two or three years ago and have been focused on AI infrastructure, digital infrastructure, from chips to cloud infrastructure and other infrastructure, to optimization
Shifting Tech Adoption Signals
(3:58) tools on top of that, and playing around with some of the apps we're seeing now. David and I have always talked about how it used to be that when I was a Wall Street analyst, earlier stage companies would use whatever they used, and then large company CIOs, CTOs would dictate the
(4:19) enterprise software applications and infrastructure they would use. That's where you got your signals.
Startups Lead Cloud Adoption
(4:19) That flipped. From my perspective, it flipped right around when I was at AWS. The numbers I'll use are functionally 100% of startups were using cloud, but only 10%
(4:44) of enterprise IT spent on cloud. That changed dramatically.
Stripe's Early G Suite Adoption
(4:44) I went to Stripe. The weirdest thing in 2018 when I went to Stripe, and Stripe is a very odd place, was that they were using G Suite as their internal email system. This doesn't sound weird now, but then for a multi-billion dollar
(5:05) company, it was very strange.
Startups Predict Future Enterprise Tech
(5:05) What we started to realize, and what I've realized recently, is that you can tell what enterprises are going to use two, three, four, five years from now by looking at startups. I don't necessarily mean startups using startup products; sometimes it's startup products, but not always. The reason is not so
(5:34) much that startups are brilliant, which I do think they are, but it's out of necessity.
Startup Tech Advantages
(5:34) One, they have no technical debt, and two, they don't have options but to use platforms that get up and running quickly.
(5:50) What does that mean today? What's popping up on marketers' tech stacks now? What are startups using?
AI Models Before Cloud Infrastructure
(5:57) You tell me. I can tell you starting with tools, I'll call them infrastructure tools, which lends itself well to what's happening in the marketing world. A piece came out from Business Insider, my old group there. It was a leaked memo. I had nothing to do with it; I was long gone
Startups Prioritize AI Models
(6:24) point. It was a bunch of stuff, but the key point I saw was that for the first time, startups were choosing their AI models and tools before they were choosing their cloud infrastructure provider. Kind of a boring statement, but having spent the better part of 10 years at
(6:49) AWS, that was a monumental change.
New Era of AI Tool Adoption
(6:49) Because we had been so early, and in every company so early, we could influence how adoption of other things went. Now you're seeing something different. Companies are choosing whether they're using OpenAI's API,
(7:14) Cloud Code, or their own GPUs or GPU infrastructure. You're seeing that happen now before they start to choose other platforms. The reason I think that might be important for marketers is that the tools people are using today, they're likely choosing tools first that they
(7:39) wouldn't choose before. I'd love to hear what you guys are seeing.
Startup Automation Trends
(7:39) In general categories, it's customer service that first comes to mind, or maybe second. The first is cloud code, but that's less to do with marketing. Anything on customer service, anything on sales
(8:03) automation, that's what I'm seeing first. It's because startups can't hire 100 salespeople. They can't hire or outsource to a call center. Anything agency or agency-adjacent, writing copy, automating outreach.
(8:32) With things like sales automation, marketing automation, CRM, are there things like what you saw with the Google suite that are standing out as go-to options?
Shifting Preferences in Marketing Platforms
(8:32) It's a little hard to tell
(9:00) right away. What I don't hear as much as I used to is people jumping on to HubSpot so quickly, or Clavio. Presumably, there are platforms out there
AI for Marketing Effectiveness
(9:26) that are doing this much more effectively, or using artificial intelligence to make it easier. Have you seen any of those? I can tell you what comes up when I hear names, but I'm curious what you're seeing. A lot of it has tried to tap into the
Clay's Popularity in CRM
(9:50) AI tech from existing tools, seeing if that can get folks further along. On the CRM side, Clay pops up a lot.
(9:59) Yep. Clay is one for sure.
(10:02) On company platform types of things, Clay for sure, but that's been going on for a while. If you ask most enterprises what Clay is, they'll say, "Oh, my
(10:15) kid plays with it."
Modern HR Platforms for Startups
(10:15) Having no idea what this is. This isn't the marketing side, but every startup I know uses Rippling or something similar. Not ADT or some platform that has a Windows 3.1 interface. You're seeing the Salesforces of the world add all this functionality,
Salesforce Complexity for Startups
(10:44) that's super cool, but if you're a startup, installing it... At AWS, we had a department that managed our Salesforce instances.
(10:52) Okay.
(10:54) AWS had a team, not just a person, a team. As much as they'll adopt AI for their existing customers and enterprises, what's going to
(11:08) happen with these newer companies is anybody's guess.
Llama's Traction and Open Source Models
(11:08) I think we can make some good guesses by seeing what there is.
(11:13) I was seeing some questions about Llama. Yes, Llama is certainly getting a lot of traction. There are some limitations to it, but the open-source stuff is free. You'll see many inference players moving towards these open models because they can't run the closed models on their systems.
Security Concerns with Open LLMs
(11:30) And ditto regarding the Chinese open LLMs; there are very significant security concerns, and they persist even on closed systems.
(11:52) For things like Meta, you don't hear about it as much, and what's gone on with Llama. Are there certain kinds of businesses adopting it more, and what are they mainly using it for? Llama itself?
Ways to Build AI Models: Building from Scratch
(11:55) Yeah.
(11:56) Or others?
(11:58) Anyone building their model, there are three ways you can build your own model. I'm saying that in air quotes for a reason. One is you can build your own model. That is a lot of work. It requires a lot of expertise. Perhaps most importantly, it requires an enormous amount of compute. That's when
Ways to Build AI Models: Retrieval Augmented Generation (RAG)
(12:39) you're going to a CoreWeave or even AWS or one of the other neo-clouds. Very expensive. That's number one. On the other end of the spectrum, you have folks doing these RAG limitations, which is taking in OpenAI or Anthropic. I forgot what it stands for. It doesn't really matter, but it's putting a shell
(13:06) around it and being able to... It's reinforcement augmented generative.
(13:11) It's really... Yeah.
(13:13) I'm supposed to know that, but I don't, and I'm not going to Google it right now because
(13:19) Retrieval Augmented Generation.
(13:25) There you go. Thank you.
Ways to Build AI Models: Open Source
(13:28) Was that clearly telling you that was said?
(13:30) That was Google AI.
(13:33) Oh, there you go. Nice. In the middle, you've got folks using open-source models. You can use an open-source model and tweak it to any dynamic you want. You still get a head start on
(13:52) building a model from scratch. You'll still have compute overhead, etc. I don't want to call them more serious engineers, but folks who need more fine-tuning of a model, but don't want to build a model from scratch, will begin with an open-source model.
(14:10) Yes, openrouter.AI. Holy cow. I see
Openrouter.AI: An AI Model Tracker
(14:15) in the chat, somebody is talking about a good tracker for AI model usage. I have been looking for the "Built With" or "Intricately" for AI models for a long time. I went as far as to try to build something that looked into website code automatically to see if it could figure out who's using OpenAI calls or Anthropic calls. I couldn't find it.
(14:45) I spoke to friends at OpenAI and Anthropic. They said, "I have no idea." But this openrouter data is just amazing. I was trying to build something to see early adoption, then I looked at openrouter and realized they did that. All right, onward.
(15:05) That's a really good way to see what people are doing with models right now.
AI in Marketing: Voice, CRM, and Outreach
(15:08) Things pop up and down, and it's wild to watch. Back to the marketing side, the early things I heard about, and I don't know how they're playing out, are the obvious things: CRM, cold outreach, voice AI stuff.
(15:34) I don't know if anybody has received an AI phone call yet. I think I got one. It's wild; you still know, but it's getting close enough where it's harder to distinguish. I'm seeing a little more traction on that in international markets. What I've seen is direct outreach, SDR type stuff, showing up first.
Identifying AI in Phone Calls
(16:08) Yeah. Those calls, once you realize this is not a real person, and it's not one...
(16:17) The pause. Some of those...
(16:21) They're not even AI, right? It's just a timer.
(16:35) Yeah. Yeah.
Voice Interaction with Chatbots
(16:35) It's interesting though. Have any of you interacted with any models or chatbots with your voice? Have you used ChatGPT? Have you used Bard?
(16:47) Oh yeah. Yeah.
(16:50) Even in some other
(16:58) applications of it, I've...
(16:59) Rosebud.
(17:02) I've mentioned Rosebud here, and David's the guy who introduced me to it. I love...
(17:09) I haven't used Rosebud voice-wise yet, but I use it text-wise every single day, at least once, if not several times a day.
Normalizing Voice AI Engagement
(17:22) David, you can explain Rosebud: it's a therapy app,
(17:25) but it's more a journaling type.
(17:28) Anyone I've shared it with has come back and said,
(17:34) "Oh my god, this is borderline addictive." I think the word you used was addictive. Ironic given its context. I
(17:47) realized this is marketing. I realized it was becoming easier and normal to use voice AI to engage when I was trying to create a logo for my advisory group. I was using ChatGPT while walking through the airport. I thought, if I put it on chat mode and put the AirPod in my ear, nobody's going to know who I'm talking to, and it nailed it. I've used it to prepare for discussions – not this one, I probably should have. I've used it to prepare for interviews. It's a very good way to use it. As we become
(18:40) more accustomed and okay with using those for our own use, I think that'll drop the barrier to when you get an AI call, as long as it's not spam. I think we'd get more comfortable engaging with voice AI as we use it more on our own.
AI for Post-Meeting Productivity
(19:04) in-person meeting, and now I'm full of ideas. I've got some follow-up things, and as I'm walking out, I'll just start talking. I tell ChatGPT or Gemini, and I just start. "Remember this, remember that," and then have a five-minute conversation so that when I's back at my
(19:26) desk, I haven't forgotten the things I wanted to remember.
Agency-Adjacent AI Tools
(19:26) The other place I've seen a handful of companies, I don't know the traction yet, is what I call agency-adjacent.
(19:42) Sorry, I missed that. I'll get to you in one second. But these platforms will build your marketing
(19:53) campaign, build the copy, and get you 80% there. It's wild. I know 2048 invested in one that was in Tech Stars New York called, I wish I remembered the name, I'll find the name. With an R, I think. I assume there's a handful of those. I just haven't seen what's gained traction yet and what hasn't. And if folks are going to work with agencies or if they're
(20:18) going to replace agencies or what have you. Earl, hands up.
Custom GPTs for Linux Exploration
(20:22) Sure. Thanks so much, David. One thing those who know me in the space know is I push custom GPTs all the time. Going back to your questions of use cases, using them through talking to one of your custom GPTs, one recent use case I've had was I've actually been
(20:44) exploring Linux.
Hands-Free Installation with Custom GPT
(20:44) since all the news with Windows 11 and people making migration because of the limitations. Last weekend, I set up Winnix on one of my machines, and I used a custom GPT. I spoke to it as I was going through the installation and configuration process. That way, I could keep hands-free and navigate on my new machine, ensuring
(21:07) everything was set up. I could literally ask questions when I ran into issues or had questions regarding any issues during the installation, configuration, or migration process. It's a great use case. I even used it once to help me clean the bathroom and make sure I didn't miss a spot. You never know the ways you can come
(21:27) up with ways to speak to somebody knowledgeable. It's great.
AI Content Creation Tools
(21:27) I'm going to look for the AI creation. I couldn't find that one, but I did find a quick list of things like Jasper. Figma is an obvious one. Predus. I'm even curious with some things Jasper is a good example where a client needed some SEO-optimized long-form copy,
(21:56) and I started testing Jasper and Writer and a few of these tools. This was before Gemini 3 came out, 2.5, I can't remember all the numbering systems. But I was asking ChatGPT and Gemini, "Can you do stuff with these very specific parameters, the exact word density, and all these kinds of things that SEO folks tell you"
LLMs Overtaking Point Solutions
(22:28) need to be there. They had someone who said, "Here's what the content needs to look like. Whether it's right or wrong, I didn't care. I just needed to do it." I tried Jasper, I tried Writer, then I tried Gemini, and Gemini was way better at following instructions. I realized, wait a second, I don't even need to pay for this extra thing. The thing I'm
(22:52) already paying 20 bucks a month for is better than a lot of what's out there. I'm curious where some of those areas are going to just be obsolete. I stopped paying for Ideogram, which I loved for image generation. Gemini, and sometimes ChatGPT now, just do it better. I still enjoy riffing on Ideogram sometimes, but some
(23:16) of these things, you might think they have their niches, and now the major LLMs keep expanding their pool of what could be done.
Seeking Alternatives for "Agentic"
(23:16) I think that's a really good point. I'm trying to figure out a way to start this part of the discussion without using the word "agentic." Can we come up with a better
(23:38) word for agent or agentic?
Defining "Agentic" and "Digital Worker"
(23:38) I'll use it for now. For what it's worth, my personal definition of agentic is when models are prompting models. But I'm not going to use that right now.
(23:51) I sort of didn't get it at first. I got it like, somebody's going to go out and do stuff for you. I used Manis, and I was
(24:01) an early user that paid them a lot of money. They sent me a t-shirt, which is terrific. And then I got acquired by Fadersburg. But when I asked myself why I was using this when I could use ChatGPT or Gemini... Yeah, digital worker. I like
(24:24) it.
The Power of Digital Workers
(24:24) Digital worker, not agentic. A chatbot digital worker, there we go, if it's able to more effectively utilize different models, etc., to figure stuff out and create stuff for me, it's better. When I have more involved things to do, I go to Mattis, and it is a much more effective platform for me.
(24:54) I just put in the chat that the Techstars New York 2048 Ventures-backed company was called ripple.ai (getripple.ai). I don't know if you've seen that, but now it's describing itself as a marketing automation agent. If you have these digital-worker-esque
(25:17) tools that can use the best of different models and different capabilities, and their own, you can have a 1+1=3.
Specialized AI for Complex Tasks
(25:17) I don't disagree with you that models keep broadening out. There's a company I work with
(25:42) called Polymathic. It has nothing to do with marketing; it's AI for science, for lack of a better term. It's a bunch of astrophysicists and other PhDs that built a thermodynamics model. Can ChatGPT and OpenAI build those types of models? Yeah, but those cases are not nearly as
(26:07) good. Their "special sauce" (I know we hate that term, but it's retro, so I can use it) is they have their own foundation model, and their agent can go and pick the best things to do for a scientist to more effectively complete a very complex set of tasks. I think that's where
Value of Agentic/Wrapper Platforms
(26:28) these agentic or wrapper type platforms need to be. They need to be in a place where they're going to unlock more value than using a single platform.
(26:38) Selena's wondering if "digital worker" instead of "agentic."
(26:42) Yeah, but I think that's an interesting distinction. The digital worker is that thing that
(26:54) goes out there and does things for you.
Agentic AI and Infrastructure Impact
(26:54) When I think of "agentic," the real profound implication for infrastructure everything else is when humans,
(27:26) prompting a model, asking questions, asking it to do things, that type of activity is going to grow. It's obvious: more users are using ChatGPT and other platforms more often. But when the models themselves that you're prompting start to do things and prompt other models, they can do that
(27:52) all the time. They can do it much more frequently and much more quickly. All of a sudden, the number of inference calls and token consumption goes practically vertical. Is that when Skynet becomes alive? I think that's where I'll make the distinction between agentic AI
(28:18) infrastructure and a digital worker.
"Digital Worker" Nomenclature
(28:18) A2A digital workers are becoming the nomenclature in Silicon Valley. Terrific. That means they do dictate our vernacular right now.
(28:38) What are other folks in
(28:41) here using? I started to put together a list of AI-native marketing tools.
Audience Check-in: AI Marketing Tools
(28:41) Is anybody using one already? Something two years ago they did not use but now they are using liberally? Whether it's for another area, tell me the other area. If it's in
(29:16) customer outreach, branding, copywriting, and it's not one of the models, we all know the models, what are the other tools?
Happenstance for Business Development
(29:16) The one I probably use most, I want to hear from others here, is Happenstance. It's more biz dev than marketing, but I'm on there all the time.
(29:39) Talk more about that.
Happenstance: Smart Network Search
(29:39) Anyone is welcome to connect with me there too. I just put the link in there. Happenstance is basically a smarter way to search your network or the network of a group of people you're in. I'm in one of these marketing agency collectives, and now we've got a group on Happenstance where we can search each
(30:02) other's networks. It's also so useful for me to say, "I'm looking for brand marketers at companies between 1,000 and 5,000 people that are in the D2C e-commerce space." I'll give it some of these criteria, and it can search whether it's my network or for people who have opted in. I use this stuff all the time.
(30:34) Often, one of the biggest use cases I also find it helpful with is for investors because I never remember who I know in my network who invests in what. Even if I know you're an investor, are you pre-seed or seed or A or B? It's very easy if you're aligned with a VC firm and there's a clear thesis they put out there. But many of the
(30:59) folks I know who invest are not part of an institution. So there might be some info they've shared out there that's hidden on their LinkedIn profile, that along with running this company, they're a pre-seed investor in agritech, and I sure as hell won't know that.
(31:22) Sounds like a next-gen Hashable, not
(31:25) the New York.
Synthetic Audiences for Content Creation
(31:25) Oh, I miss Ash.
(31:26) Awesome. Flame out. Yeah, I miss it too.
(31:29) Eric, do you want to share anything about Go Marble?
(31:31) I use synthetic audiences a lot. I don't know if that's relevant to what you're asking.
(31:37) Yeah, sure. Synthetic audiences I find very
(31:41) useful for content creation, creating surveys, adding the results to build content. They're extremely useful tools.
Tools for AI-Powered Content Generation
(31:41) I use Ask Rally and I just started using Navara, from Jill who presented. That's a little
(32:14) more complicated; I like to understand it, but they have features for generating content. With Ask Rally, what I do is create surveys. I ask the audience, they take the survey, then I take the output, put it in Gemini for example, ask for an analysis, and then I can use that to generate my content.
(32:45) That is awesome.
(32:47) Yeah, and the other thing I do is, for example, research this other tool I use for researching prompts, what people are searching for, then use that and Ask Rally to generate surveys, then use all of that to create content.
(33:09) Sorry, did you say there's something that tells you what prompts people are
(33:12) using?
AI in Marketing: Limiting and Avoiding
(33:12) Yes, I'll tell you in a second what that is.
(33:14) Yeah. While you do that, I can bring up another angle I've seen AI in the marketing space, which is the flip side: not how to use AI, but how to limit or avoid it altogether. I know there are many companies that
(33:36) ask, "Is your brand being used properly, etc.?"
Voit: AI Imagery for Brand Protection
(33:36) There's a company I advised for a while. It's now called Voit. I'll put the URL in here. Forgive my Techstars plug again, but this is a Techstars LA company. They provide a workflow to make sure that AI imagery doesn't end up in companies' brands. Where they became very well known was in
(34:09) things like Magic the Gathering. They have a big contract with Wizard of the Coast and video games, etc., to make sure people aren't sneaking AI-generated content into their actual IP. In those fan bases and groups, there are very significant problems if it's not original content. I think brand management
(34:37) for sure is something I've seen.
Prompt Research Tools and Resources
(34:37) The company is called Gumshu AI.
(34:39) Oh, nice.
(34:40) Of course. Yeah.
(34:42) As we're talking about things for researching prompts, the two I've spent some time with are Otterly and Passion Fruit, and those are pretty good. What I'll also share in the
(35:05) chat is I updated my own collection of resources.
Vibe Coding on Base 44
(35:05) Not everything discussed here is on this, but some of them are. Feel free to check that out and recommend others that should be there.
(35:14) I was about to ask you if you have that, and I'm glad you do.
(35:18) I have it, and I realized that with my vibe coding obsession
(35:31) on Base 44, I checked a couple of weeks ago.
Experimenting with Vibe Coding
(35:31) I had so many more credits for this month than I realized. It's funny, often there's this fear of running out of them, and I was being too conservative with it. So, I just started vibe coding everything. Dave, I put something in the chat that I didn't even show you yet, that I asked,
(35:53) "Could I actually vibe code an alternative to Rosebud?" I created "Innervoice Me." It's not great, but it works.
(36:01) Yeah. Well, I think that's another really great thread. Whether it's vibe coding or things like Lovable, which David, you and I have talked about. I know you adore it.
Marketers Using Lovable
(36:20) "Adore," I guess, is a better word. You passed Lovable.
(36:23) How are marketers using platforms like that? Lovable seems more targeted at folks doing things that are more marketing-
(36:30) Yeah, it's really good.
(36:31) heavy. How would you line up Lovable, Replit, Vercel, and all
(36:42) these thousands of others?
Platform Technical Savvy and Entry Points
(36:42) For me, I think one of the biggest factors is how technically savvy you need to be to use them. Base 44 versus Lovable, for me, I still see those two as the first entry points. You can get something pretty competently done even
(37:12) with the free trial. But as you get into other things, even with Lovable, it helps once you start asking it to do something more complicated. It's like, "Register for Supabase, register for GitHub," and it'll tell you how to do these things. Many other platforms are free, especially if you're not launching a major consumer product.
Base 44's Integrated Development
(37:33) For me, Base 44 is the fewest number of other things I knew to use. I just realized, for instance, that they have backend functions and agent mode that you have to manually turn on. But once you do that, originally I was using Form Spree that both Base 44 and Lovable recommended as an email capture thing to sync with.
(37:59) It was easy and cheap. Then one day, B4 said, "Yeah, I can do that as part of this. You don't need to go somewhere else for this." I've tried Replit, I've tried Vercel, I've tried some of Google's own tools. I really want to learn how to use Cloud Code, and I have it downloaded. I wrote this last week in the AI brief of using
(38:21) Cloud Code and Ghosty as the terminal and Netlify. I've gone down all of this, but to me, Cloud Code is harder for me to understand than the Korean alphabet, which I'm learning on Duolingo right now.
Cloud Code ROI for Developers
(38:27) A data point I'll give you is a twice-exited founder CTO I know is using Cloud Code. Many developers
(38:53) pay $200 a month ($2,400 a year), and he's getting the equivalent of $2 million in annual developer work done, in a week. He knows these numbers because he's run large engineering organizations.
(39:18) So when I hear things like, "enterprise adoption is so slow with AI," that may be the case, but at some point, when you have these ROIs that are orders of magnitude, it's too hard to ignore.
Blurring Lines: Development, Agency, and AI
(39:20) David, do you see the line between, now that you have all these, what we used to call no-code, now it's not even coding? These are like,
(39:45) "Hey, what do you want to do?" Are you seeing the line between what is development, what is agency, and what is developer blurring? Yeah. How does that all fall out? Who's doing what?
Vibe Coding for Faster Site Management
(40:00) I have a freelance developer who I use for all my personal stuff on Upwork, and I took my main site for
(40:11) consulting back from him. I vibe-coded it. I gave it to him. He put it on WordPress. He did all the things any smart developer is doing. I trust him. But then I realized, "This is too slow for me to tell you what I want done," even if WordPress has better SEO functions or things like that. I literally was
(40:37) yesterday on Base 44, telling it to build sitemaps for me, then I was submitting them in Google Search Console.
Vibe Coding Speeds Up Creative Projects
(40:37) These things mean there's a lot of quick-hit stuff I don't need developers to do. For more involved things, I vibe-coded a page for a content project I'm working on with a client.
(41:07) They didn't ask for this. They didn't ask for anything like this. I gave a few things they didn't want in the process. But I was able to bring this to life for them. I thought, "Okay, great. Now you can put this as a page on your own site however you want. I just gave you a whole vision for it and this functioning version." This speed to
(41:30) develop... I wouldn't trust myself or a typical marketer for a really important interactive thing that has to work. But there's all this stuff coming from the agency side. There were all these things that even just for the agency's own marketing, let alone doing client work, where it would be a big resource discussion. Do you spend
(41:56) months in meetings trying to get something done just for some quick hit, some wacky idea you want to bring to life, so you could go and, in earlier, less crazy days, you'd put it out in a tweet, and see if it bites? Now you can just do that in a couple of hours or a couple of minutes. That kind of
(42:21) stuff, for a creative person, or even an account manager, to run something by a client and say, "What about this? Can we do something like this?"
Evolution of Content Optimization
(42:21) That's way better than creating 50 slides about it.
(42:26) Yeah. As long as you talk about client work, I feel we went from, in
(42:43) Web 1, it was SEO, then in social media times, they refused to use the term Web 2. It was, "How do you get your stuff in the socials?" Now, David, I think you told me they were calling it Gener GEO. Whatever it's called, there are so many of these.
(43:07) Yeah. I read something a year or two ago, and I said, "Let's call it Seal Mode." It was SC AI. It was a joke. LLM optimization. It was a joke, but it seems like that's a thing. And this is where we're going, Daniel. I see your question about compensation models for AI-powered creatives. What
(43:34) if publishers or content creators need to have their stuff show up in models because that's where everybody's looking for them now?
AI Content Compensation Models
(43:34) How are they getting compensated for that? How do you think that plays out? I have some ideas. I don't know. You can see what OpenAI is starting to do now. It's
(43:54) some revenue generation, but how do you see that working out? SEO or social media optimization is, "Oh, you can get more stuff in, you get more traffic." This is not traffic. This is content. This is content generated on your content.
(44:19) Although it depends what business you're in because the trends we've seen are fewer clicks, but those clicks become more valuable.
ChatGPT's Ad Business Model
(44:22) A reporter was just asking me about the ChatGPT ad business model, and was hearing something about a $60 CPM, which sounds absurd, especially when they should be able to charge potentially obscenely high
(44:56) CPCs compared to what people are paying if the intent coming from AI is higher.
(45:00) So this whole business... Given how AI has seemed poor for general branding, not that it doesn't influence brand, but branding... Are you going to get more
(45:23) traffic from that? Are you going to get more engagement?
OpenAI's Token-Based Business Model
(45:23) Yeah.
(45:24) Yeah. My opinion, right or wrong, is that OpenAI's business model is ultimately predicated on selling tokens, not ads. I could be a thousand percent wrong on that.
(45:50) The markup on a token is 2x, so you have a 50% margin. I think they said something like 40% of their revenue is now from API access. It was a quarter. I thought it was 75%. I was wrong, but I'm getting less wrong. It's DoneKazam, which is nice. It'll be
(46:19) interesting to see what their ad model looks like.
Future of Marketing Technology
(46:19) I heard people talking about it yesterday about this new product, and they thought, "We know what the product is."
(46:27) Is it new?
(46:29) What do you say, if 20-30 years ago somebody was using Oracle for their CRM,
(46:42) and you say it got replaced, then now within the last 10-20 years, it's Salesforce. Platforms like that, if we had to look forward 20 years from now, what do the big names in marketing technology look like?
(47:10) What gets replaced?
Viability of AI Point Solutions
(47:10) Curious what others here would place bets on and are thinking, just beyond the rich getting richer. Manis was fascinating because it kept popping up as a favorite of many marketers. I feel so many of these things
(47:39) that have gotten some degree of traction. If you look at Jasper Writer, or Beautiful AI, or Gamma, which both have probably had the most traction on the "what's the new PowerPoint" kind of thing, but if Gemini, Google, and Microsoft don't just flat-out acquire those, they'll build their
(48:09) own versions of them into it. I'd be skeptical that some of these point-solution things are viable for the long haul, especially when anything is competing with part of the Google or Microsoft Office suite. Just because Microsoft tends to neglect user experience in Office today doesn't mean they're not looking at every single one of these
(48:35) and are going to... Yeah, and won't place a bet for the rest of the decade. Yeah.
Native AI vs. AI Add-ons
(48:35) Yeah. I said this to an EVP at Microsoft five years ago. They were thinking about their startup engagement. He reported as SIO; he wasn't really concerned about startups, but I said,
(48:59) "Look, you don't have to make another nickel in Azure, but you need to know what startups are using." This is the thread we started with, because this is what your enterprises are going to be using five years from now. Google is not a tiny company by any stretch, but I don't know the market share numbers now, but
(49:19) I would expect that G Suite has a material market share in internal email systems and documents. What I think was very different when cloud happened: Microsoft said, "Sweet, let's put Word on the web." Google, I think they acquired into it, said, "It's not just put some word processor on and give you web
(49:43) access. What does a truly native cloud-based word processor document sharing system look like?" I think the questions being asked now are not, "How do you add AI to Salesforce?" or "How do you add AI to Excel or Google Sheets?" but, "What are the native AI? What does a native AI analysis
(50:11) look like? What does a native AI look like if we don't start with the tools we use now?"
Reconsidering Startup Tech Adoption
(50:11) Yeah. Jamie, I thought, "Holy, that's big market share."
(50:15) I know your question was, "What are the big names in marketing technology in the next 10 years?" but
(50:30) what I'd like to do is revisit something slightly adjacent to what you said at the very beginning when you took an interest in what people are using now and that adoption curve explodes over the next five years.
Fleet DM: A Model for Remote, Cloud-First Companies
(50:30) I do M&A, and I advise mostly on the sell side. I'm in outreach now for a company I'm selling based in Paris.
(50:54) I discovered a company, I'm going to drop it in the chat, that is refreshing and unique: Fleet DM. The reason I discovered them is their ethos is 100% remote, no offices. They have a post office box in San Francisco. The company I'm selling is the same way; it's in emulation and virtualization. It's very high-tech
(51:23) stuff, but these companies are operationally aligned, totally cloud-first, distributed delivery, distributed management. To decide whether or not to contact them, which I did, you find out their CFO is in venture, and she's amazing, but she's not on site. The guy they have running it, who actually is running corporate finance
(51:47) over there, is a specialist in building remote organizations. If you surf around their site, it's completely transparent. It's so refreshing to see a company that says, "We're doing it this way because it's transparent. You can trust it. We can move fast. We're not encumbered by anything that's legacy." I think that's kind of what you were hinting at at
(52:14) the beginning when you were saying, "What's going to move fastest with cloud?"
Wrapping Up and Community Engagement
(52:38) Yeah. Jim, I was hoping you'd chime in at some point. You never disappoint. David, you don't disappoint too often either. You shortened today. So I appreciate you coming. We've hit the hour mark. It would be great for others to stay in touch with you and have you back sharing more ideas about this. A really fun community conversation.
Community Shout-out and Next Steps
(53:03) A shout-out to Daniel and Jean for the technical recommendations. I like how there are many of us at different skill levels on the marketing side, on the tech side, on all kinds of sides of this business. This community is welcome to one and all. Porch Capital, learn more about David Levy, and stay in touch
(53:27) with him. His Substack is really good. I often need to not read it the second it comes out because I need to sit somewhere where I can process it.
(53:35) I write long-form.
(53:37) That's a compliment. Follow me on LinkedIn. All this stuff. Next week, we have Ad Age's chief technology
Next Week's Guest and Closing
(53:45) reporter, Garrett Sloan. Another great observer of the market. You'll be able to ask better questions than I could think of. Thanks everyone for coming. Hope it's a great rest of your week, and see you next week in Slack and everywhere else. Thanks everyone.