The AI Arms Race: Real Tools, Real Use Cases & What's Actually Working Right Now

Speaker A: Foreign.

Speaker B: Hey, everybody, and welcome to another episode of the Ad Project podcast. Um, today I've got a super special guest joining. Uh, Mr. Ryan Craver. Ryan, it's awesome to have you on the podcast.

Speaker A: Joe, fantastic to see you. I just wish I was in Duluth with you, man.

Speaker B: You'll get up here. You'll get up here. Yeah. Last time. So Ryan was trying to get to Duluth, and you guys had a foot of snow in New York.

Speaker A: Foot of snow. I think I was moved on Delta flight seven different times, three different airports, and, uh, never made it out, which was pretty bumming.

Speaker B: Yeah. That's awesome. So, um, so Ryan is going to be one of the new co hosts coming up for the new Power Craver Charge podcast. Um, and so wanted to introduce you. I mean, this is a good time to just introduce you to the audience as a who. So maybe give us a quick background and then how you and I are connected now.

Speaker A: Quick background. So I was in retail for a, uh, number of years. I was at Accenture Consulting. Left to then build a, um, agency. Started in New York. Paid, uh, the first payroll, actually, with Bitcoin, which I wish I would have held, but, uh, that. That allowed me to.

Speaker B: That.

Speaker A: That was 2016.

Speaker B: Okay.

Speaker A: Right. So just don't. Don't even do the math, because it'll just make.

Speaker B: I'm doing the math right now.

Speaker A: Be depressed. Um, but I left to then, uh, found what was called Commerce Canal, ran that for eight years or so, and then merged, uh, in. Acquired by Mountain Gate Capital in Padian. Uh, and now fast forward to today, and I'm running some products, some AI and our consulting teams as the Chief Commercial Officer.

Speaker B: Yeah. Yeah. And so now we get to work together since AD advance was also brought into the Padian umbrella. So, yeah. Ryan is one of the folks within the team that I've been super excited to work with.

Speaker A: Yes. Amazing.

Speaker B: Don't make that sound bad. One of the many, many people.

Speaker A: No, I knew from the first time we met in Vegas that this was actually. It was in New York City. We met in this office. Um, that this was. This was a good mix. Good mix of. Of brains, power, understanding, and knowledge.

Speaker B: Yeah. Yeah. So, I mean, what we want to talk about today, like, anytime Ryan and I get together, we're talking about all the new updates that we're seeing within AI. Um, it's just changing so fast. Um, and so it just. It kind of. It feels like this battle of constantly trying to keep up, but at the same time, we have all these new tools that help us Keep up. And so, um, I, I don't know, Rya, what's, what's your overall perspective of the industry? Like maybe you start super high level, like what do you see in. And then we can take it more down to the weeds and key use cases and key applications that we've seen.

Speaker A: Yeah, yeah. We can certainly geek out on this. I think what's so alarming to me is the pace with which it's updating. Right. So we all have to Remember that in November 2022 is when OpenAI released ChatGPT and it felt like we saw iterative change. Right. We saw at launch it was a blitz, everyone wasn't excited. Then if you remember, Google stock went way down because they were worried that Gemini was going to be nothing. At the time it was called, um, started with a B. Can't even remember Baird. Was it Baird or something like that?

Speaker B: Bard. Bard, yep. With Microsoft's, uh. Bard.

Speaker A: Right. And so then they iterated. They got.

Speaker B: That was Google's Bard compared to Bing. Yeah, right, right, yep.

Speaker A: And then they got in line and then we started to see anthropic start to pick up. We saw some mix of perplexity and. Fast forward to today and we've got a. Basically a flip flop that's happened over the last couple weeks. Right. So Claude sun at 4.6 came out and that led to OpenClaw, which then led to Perplexity Computer. And it's just, it's been a wild time. So the iterative change and speed with which it's happening has, has completely accelerated. Um, and we've gone through some peaks and valleys as to whether we think that this is a game changer or it's evolutionary or it's a compliment to everything that we're doing. And I just think that, uh, the pace probably seems like it can't get any faster, but it just may.

Speaker B: Yeah, I, I totally agree. And it just, it's wild, the timescales when we're talking about AI too. Like for instance, like one thing that was really eye opening for us as the team, um, we had Amazon's Creative X team that came into the Office and we were talking through just the updates in both video and image generation. Um, and they were talking through, you know, Yep, the last update, like May of 2025, like, you know, super. Like so many, so many things have changed. And we Talked about like 2025, May of 2025, like it was like the Dark Ages, you know, and it's like just for perspective, like, this is like Eight months or ten months ago, and just the wild changes over that time, just with that one single product. Um, it is just crazy, the pace that things are updating right now.

Speaker A: Yes, it's unbelievable. I mean, let's just take an example of, uh, Anthropic releasing dispatch two days ago. Right? Two days ago they launched it and you and I are like, have you tried it yet? Have you tried it yet? And we have this fear of missing out that we haven't even tried it. Um, it's incredible, but it's so fun. So fun.

Speaker B: Yeah. So. What's some of the key ones that you're going to right now? Maybe we'll start there and then we can kind of take bigger picture perspective on how this is going to impact industry marketing, uh, agency life as a whole. But like, what's some of the key use cases? If you look at your current AI stack, what's that look like?

Speaker A: Yeah, I think what we saw was, uh, ChatGPT became kind of the go to for a lot of us. It even took over the verb. Right. We no longer say we're going to Google it, we're going to chatgpt it. And they, they took the consumer route. Um, it took the consumer route. So I was using ChatGPT for considerable period of time. We were also deploying it internally within the agency as kind of the go to. Um, and then we saw Gemini come along and Gemini had stuff like Nano Banana too, so we were using it for creative and um, anything tied to videos, new assets on PDPs, stuff like that. And then we started to see first through browsers with Perplexities, comet and, and OpenAI's, uh, Atlas, the ability to take over your computer, which then morphed into Claude Cowork, which truly took over your computer via an app. And it wasn't browser based. And I think that that's what's really stuck more recently. Right. So the ability to create these tasks, to run everyday market research, to run new competitive intelligence. It's so seamless, simple and automated through Claude Coworks, I'm using that pretty considerably. Um, I have tested OpenClaw about five, six weeks ago when it first launched. It actually led, I think I told you this story to my wife that Saturday receiving a bizarre email saying, I'd like to extend my network on intelligence.com with a connection. And she was like, is this you? And I said, I mean, it is, but it isn't. And then the subsequent emails from clients the next week about it. Um, so I think what's interesting now is we've seen this evolution of openclaw just in six weeks to bring up things like Perplexity Computer which is cloud based, a little bit more simple for the less technical user, uh, to build. Um, so I've got some of those tasks running again around market intelligence, competitive intelligence, uh, daily recaps, those types of things.

Speaker B: Yeah, yeah. So I, I'm currently, I have a home server because, because I'm cool and so I've got a, I've got a virtual machine running on my home server that's running Open Claw right now too. Um, for those who aren't familiar, there's been so much talk about openclaw like give a, give a high level breakdown on what it is,

Speaker A: the orchestration of agents. So replace agent with automated tasks and people just going out using various LLMs to complete something that you'd like them to do. And it's the orchestration of those agents and typically done through Telegram so quickly just texting to openclaw the setup is what I find uh, probably the scariest for the non technical users.

Speaker B: Sure.

Speaker A: So generally if I talk to someone about openclaw I'll say hey, maybe try Perplexity Computer Even though it's $200 a month or try something just around Claude cowork to get going. But the orchestration of agents, is that a simple enough definition?

Speaker B: I think so, yeah. And so I mean the big thing, like I see a lot of people coming in and they assume once they install OpenClaw that it just runs itself, but you need a uh, model that's tied to it and so you can buy like the little Apple Minis. Like the benefit of that is the RAM that's in there that can also be used for like the graphics memory which is used to run these different local models. And so that's why you see this big push for everybody buying these little Apple Minis. You can run models locally on it. Um, but the easiest path to actually get going on it is to just install an API key and run it through one of the core base models. So when I initially set up Open Cloud or Claw, I was running it through Claude. And so there are still like LLM costs that go, there's API costs that go along with that. Um, and it's not fully local. So I think that's one piece where I've seen a lot of people get tripped up where they think they can just uh, install Open Claw and that's like every everything now you've got to connect it still to a model. Um, and then there's all the connectors that are in there where it can interact with your email or your calendar, different things like that. And then you tie it to a text, like usually like an app like Telegram. I have it tied to WhatsApp. Um, and from there from your phone you can control the agent, but in terms of like general controls and keeping it from going off the rails, like that's kind of the bigger concern. Um, other pieces is like, as you continually build up more and more memory within it, um, context window can get really large within the models and if you're paying for that, then costs can skyrocket. If you're trying to run it locally, those models might not be able to handle it, which opens up other security concerns. So there's a lot of potential issues there. Um, tell me about perplexicomputer. So, so this was one that I hadn't heard of before.

Speaker A: So the interesting thing about Perplexity is the fact that they've gone to the market to say we are the wrapper for AI, meaning we will tie into any model. We are going to have Kimi 2.5 from um, Alibaba, we will have anything related to Claude. So Sonnet 4. 546-ChatGPT 5. 2, et cetera. And when you're processing a prompt or a query within that platform, they make the decision to have the most relevant model they believe will answer that query.

Speaker B: Mhm.

Speaker A: Originally when they launched they were more real time because they were creating their own LLM to have more real time ingested. Because if you remember the days of ChatGPT early, they would only update the data through a certain date and then they'd say ok, you always get the

Speaker B: prompts where it's like, I've only been trained through May of 2022. And then it's like, okay, sure, right.

Speaker A: So that's evolved. And so now what they've released is Perplexity Computer. And their promise on um, Perplexity Computer is somewhat similar to Claude in that it can do deep research across, I think 19 different models LLMs to provide you with, with the best possible output. What I found in the use of it is it heavily relies off of Sunnet 4.6, which is Claude.

Speaker B: Sure.

Speaker A: So, um, what I found is it's a great way to test as opposed to pay the token price of Sonnet 4.6 because as you mentioned, it gets pricey quick and you can't really determine on your query what the cost of that's going to be. And if it's a long standing query that you have running or task running every single day, it adds up pretty quickly. Um, so that's Perplexity Computer. It's all within your browser, it's all tied to your mobile app. So very similar thinking to Telegram, WhatsApp with OpenClaw and then now Dispatch with Anthropics. Claude.

Speaker B: Sure, sure, yeah. So, so those are kind of like a couple different iterations and then on top of that, which I would say is a little bit more polished or maybe uh, maybe more like roll out corporate friendly. Um, yes, we've been, we've been investing a lot into Claude lately and especially like the desktop and the Mac app that we have, which ties in chat, cowork and code all in a nice to use graphical app. Um, we've been having a lot of fun here where we've enabled Claude code for some of our team and just being able to tap into that. It's been amazing to see what some of the team has done or tasks where it's like we're getting multiple of these a week where it's like, hey, this would have taken me over an hour and now it takes five minutes and I can set this up on a recurring schedule. Um, and so just being able to give access to the right data in a controlled environment, um, it's been simplifying so many things down that we have in agency life. And what's really nice with the Claude app is now you can build out plugins that we can roll out to the team. Um, and within these plugins there's two different options. So there's connectors, so what data can you get connections to? So we can control that and roll that all out kind of in a standardized plugin, which is really nice from a team perspective. Uh, we can control the permissions and everything and make sure that um, it's read only in most cases. Um, and then there's also the skills piece, um, which has been really fun too. So you can. One of the core issues we've had with the models up to point, especially from like a business perspective, is just being able to provide that additional context that you have that we've learned over time. Or hey, we want to approach the issue this way. Or hey, we've already filled out so many of these proposals. Like there's we should build up a knowledge base on what that is and so uh, layering in skills that allows us to really tap into a, uh, lot of these cool items where now we can create this more repeatable process. So when somebody is looking at like, hey, I want to pull up these metrics here. Um, it can tap into a skill that will highlight what that looks like. Um, and so like with Cowork, you're kind of getting the agentic perspective where, like now with Dispatch, that kind of acts like Open Claw, where I can use the cloud app and I can actually interact with my computer locally. It just has to be running. I, um, can give it access to certain files. So if I pull an Excel file or download a report, it can pull that in. Um, and then with Skills and then the connectors. And then the other cool feature is you can set up recurring tasks. So every day I can have it scroll through and. All right, check out all my Slack messages and emails and let me know what I should focus on. Uh, also my calendar. You know, there's so many different applications that get me, uh, really excited. And I feel like we're at this tipping point with AI where now it's gone from like, more of a glorified search engine to like a true assistant or helper.

Speaker A: Yeah, I totally agree. The thing that I'd love to hear your perspective on is I know within the agency world, the way in which we're using AI today, we're ahead of the curve. But little things like if you go back, do you remember the EM M dash controversy, the crazy EM dash craziness where so, uh, ChatGPT had over time developed this EM M dash where it was very clear that someone was using double dash.

Speaker B: Yeah, right. Every. Every prompt, I would have to say don't use the double dash because it looks like a.

Speaker A: Right, exactly. So do you believe that our analytics reports and our PowerPoints that clearly look like Sonnet 4.6, eventually other agencies will catch up as well on AI and will have similar use cases to what we're using. And it seems as though Anthropic is winning the race right now, at least for Enterprise.

Speaker B: Sure.

Speaker A: That we're just going to dumb down our output and it's just going to be so clearly visible that we're using AI. Where's your head on that?

Speaker B: I think what I've seen right now is you can customize it enough where. If I can feed in skills that are brand guidelines, anytime it creates a document, anytime it creates a presentation, I can have it customize it to look the way that we want to. And what's so cool and where you can really accelerate this is like, I can just feed it a lot of our documents and then it has a Skill builder built into it on how to build the best skills, and then we can just keep iterating as we go. And so that's another key piece that we're looking at is like, okay, um, we want to roll out the scaffolding to the team and we're going to have the smart team that's going to think of all these use cases that we could have never even predicted. Um, you don't want, like, me sitting there trying to build all these skills for the agency. We want to empower everybody. Um, and so I think what's going to be cool is the quicker that we can continue to drive that education and adoption. We're just going to keep iterating. The models are getting better and then I think skills is going to become another one of those, like, natural pieces that other people also take into account in the future. And so one thing I am worried about is, like, we're building around Claude right now. I don't want there to be any future lock in. Um, so longer term picture, I think we use an app that's going to give us access to multiple models, but then we should be able to reuse all the skills. So, uh, with AI, I just don't see us getting pigeonholed into one area because the second something else comes out, we'll just say, okay, I translate these skills into whatever format you need next. Awesome model. Uh, and we can make the jump there too.

Speaker A: Uh, that's a fantastic question. I was actually thinking about that just the other day, because as we continue to lean into Claude, I mean, is it too early to say, Would it be unwise to say that Anthropic has won the race? Right.

Speaker B: Like, totally unwise to say.

Speaker A: Right. So if you just, if you think I try to liken it to Microsoft, back in the day, took the enterprise route, Right. Embed our software, embed windows into the enterprise, it will then flourish to the consumer. Apple took the route of, clearly, they started ipod, then they did iPhone, then iPad. They took the route of the consumer. So is. Is Microsoft anthropic and Apple is OpenAI ChatGPT. They took the verb, you know, like, I don't know, portability amongst tools. It's going to be interesting.

Speaker B: Yeah. And I think that's the key thing. It's like, you don't want to get locked into something. So in the future, I think we're going to want to give multiple different options and just even like the scale with what Padilla is working on, like, different models are going to make a Lot more sense on the creative side versus the technical media analytics side, um, versus supply chain and all those items. There's going to be different models that are going to be better. And so in the future we've definitely got to plan for that and give people optionality. We don't want to force one option. Right now this seems to be the best, most m contained option that we can roll out as quickly as possible and then from there see what people build and then as new models come out like hopefully we can get a nice container application that gives us multiple optionality where we can reuse skills across the um, but then be able to pivot as these new models come out, not just be locked into one ecosystem. That's my hope.

Speaker A: Do you ever anticipate, here's, here's an interesting question. Do you ever anticipate Anthropic opening up to other models?

Speaker B: Maybe, uh, I don't know. It's this weird. You get into this weird area because we're going to quickly get to the point where the models are so good that models are going to become more of a commodity.

Speaker A: Right. And then be token trading. Token trading.

Speaker B: I think the key dollars that are going to be made are the people who uh, are selling the picks and shovels versus the model creators. And so uh, then it turns into more of like a cloud hosting type play where all right, if I can just run through their systems and it's all in this nice contained system and it's got the right controls around so I feel like it's not going to get off the rails for our team. I potentially, I, I don't know. I mean that would probably be a pretty big jump for Anthropic to take a leap and start offering other models. But eventually these models are uh, going to be so good where it's not going to be. I'm going to pick the best model all the time. So I'm going to pick the most cost efficient or it's going to fit this application best. There's going to be routing overall and it's not just going to be one company's model. It's going to be all models. Which one should I use for this task?

Speaker A: Um, kind of the perplexity model. Right?

Speaker B: For sure, for sure.

Speaker A: It's just I think about user interfaces that we start to get our teams associated to and we're going down the Claude anthropic route and you know, so much of it is build your own platform like Purvey, right. To utilize these various models. So we can more independently switch. I mean, we made some switches on Purvey, specifically where Sonnet. Sonnet was just too expensive. At one point we jumped to Kimi and then we went back to Sonnet. So, uh, it's just going to be fascinating to see it all play out within these walled gardens that eventually do become commodities. So.

Speaker B: Yeah, yeah, yeah, it's, it's a wild, wild world every time we talk about it too. I just, I, I spent so many times this week where I've been up way too late because it's like up. Uh, just got to get the model going again for this one and we're building out something new. That's cool. Uh, it's, it's just changing so quickly. Yeah, it's a, it's an exciting time to be in the industry.

Speaker A: 100%. 100%. Want to do anything else?

Speaker B: Well, I think this is a good preview on what kind of the next iteration of the podcast is going to look like. Um, so in future episodes when you join in, it's not going to be the Ad project anymore. It's going to be the Power Craver charge. Um, and so make sure you tune in. There's going to be a lot of really cool things. But I mean, Ryan, give us a quick overview. What's your take? Like, how's the scope change from Ad Project, which is much more like retail media, specific to the new podcast?

Speaker A: Yeah, so we're going to do three main topics every single, uh, episode. So we want to make sure that we're not just speaking specific to retail media because we're a much bigger organization that's got global services. So we'll, we'll have probably a tech lens, um, where we'll talk about the ways in which we're using tech within our industry. We'll probably have a marketing or advertising piece that will talk about what we're doing within that space. And then we wanted to keep it fun. So we'll probably have something around lifestyle and, and how we're, we're evolving as an agency and, and leadership, uh, within an agency. So it should be entertaining. We'll snip it up so it's quick, um, bite sized elements and we'll, we'll broadcast it anywhere that we'll take it. So looking forward to it. And Joe, we're having you on for sure.

Speaker B: Sounds good. Yep, I'm looking forward to it. So make, uh, sure you tune in. I'm excited for the, the next, uh, journey of the podcast. And uh, yeah, it's going to be a lot of fun. So, Ryan, as always, great talking to you. We could. We could have continued this for another couple hours, but we'll save that for upcoming episodes.

Speaker A: Awesome. Uh, thank you, Joe. Great seeing you.

Speaker B: Yeah, thank you. And for everybody who's always listening to the Ad Project podcast, really appreciate you listening, and we'll see you on the next episode.