S2E2 | Claude Code as Daily Driver Strategy vs Execution Gap with Riccardo Vandra & Spencer Tahil

So guys, welcome back. So today we have my very good friend Ricardo Vandra. He's joining us for Behind the Growth, and we're going to be talking about kind of like the current state of go-to-market engineering or even AI automations in general. This is going to be a pretty interesting conversation. We're going to pick up a very private conversation that we had around a few weeks ago about what the current market landscape looks like. So Ricardo, um, feel free to introduce yourself and then we can just kick off.

Yeah, excited to be here, Spencer Tahil. So I'm Ricardo, founder of Uplifted Consulting. We focus on AI automation and more recently, AI orchestration. And so I feel like that is a term that has been thrown around a lot right now. And yeah, want to discuss more about with you about what does the future look like of AI in general and like all the different things that we can, that we can think of in this crazy space right now that is changing every 6 months, right?

Yeah, I mean, like, I feel like this is like very real. I mean, we sit on different sides of the same coin. I feel like I do a lot more like applied AI that like in marketing and sales. But dude, if there's like a technical question, I feel like I always ping you on LinkedIn.

I'm like, hey man, help me.

But like, for instance, like I feel like 6 months ago we didn't have like Claude MCP and I feel like now everybody knows like Claude MCP, Claude MCP, Claude Code Expert, like for you? Is that something that you're building out and what does that look like?

Yeah, so like that is something that really changed the, like changed the way that I work personally. And so like I generally try to stay away from all the hype and like say, oh, this is the new thing. Like this is awesome, this is amazing. But like actually generally speaking about that, Cloak Code is like that kind of ChatGPT model moment that I had during the last day, the last few months actually. Like I think I used to work on that probably 6 months ago, just getting started with that, not really in a structured way and then like got more back into it around November probably. And with that I actually started using that like on a daily basis, where right now I would say like 90% of my time is spent through CloudCodes as an interface. So yeah, that's that. And yeah, so the, I would say the beauty about the system itself is that on one side you can do anything you want. So generally the tool was born as a coding tool, right? And so A lot of the people get scared about the name itself, a cloud code. So you think that that is actually a coding tool and that those are the only use cases. But turns out that with the new models, with the Opus 4.5, you can do a lot of different things around non-coding realm. There are like, there are unthinkable, I would say like a couple months back, also now, especially with lots of new features, skills, subagents. They have their own stack of different features added into the tool itself. So that's crazy, man.

I recently converted my friend Jonathan into using Claude and he's like, oh, like, I don't know, I don't understand why you're like a max user on Claude and you like max everything out every other day. And I said, well, yeah, like you, it's like continuous context and just connects to everything on MCP. It's just, And then he, he got a subscription on, on Company and he was like, I understand now. It's so easy. It's so easy to use. And I said, yeah, I mean, I told you it's so much better. So I don't know. I feel like it's more of like an orchestrator now, especially like in the applications that we're using, like marketing, sales, and like go-to-market engineering. And this, like, I hate that word, the terminology, but it is what it is. Um, I really see Cloud Code as being able to bridge the gap between like button pushers and actual implementation, right? Like you can do execution. You have like these agencies that do the teams and, you know, competitors of mine do like strategy as well, but people will lap me in like 3 months if I'm not, if I'm not using Cloud Code, they just deploy so much faster.

I feel like, yeah, I think, yeah, that is, that is probably across every single dimension, probably. And so, like, just to give you an example, I recently released a video where I showed a full outbound process going from— those are the people that I need to target, and this is the specific offering. And the old system goes to finding the actual leads, creating an initial copy for them, and then pushing that into the platform in a matter of 7 minutes, 10 minutes or something along those lines, right? Obviously, like the process that I showed wasn't like completely optimized around personalization strategy, all these different things like that, but I feel like the possibilities are there and like with a simple kind of system that would've required probably, I don't know, like 5 hours to build, 10 hours to build with other kinds of tool, you can build the same very fast. And then on top of that, you can augment that tool with, for example, different email templates, different framework that you want to use, different ways of researching personalization. And so, yeah, the only limit that then becomes the actual strategy that goes behind. So it still requires a lot of expertise in thinking about what could be the levers that you want to pull, if you want to work on intent-based plays or different things like that, but yeah, that's the state of the game right now, I would say.

What I'm noticing, and we were speaking about this privately, is that, I mean, like, I'm shifting my business and I think you're shifting yours like a little bit as well. It's the core of the work is Actually kind of changing a little bit and it puts, I don't know, I feel, I don't feel threatened by people being more technical than me. Like you're more technical than me, but like if I have a question again, like I come to you and like, hey man, like, can you help me out? Um, but I do definitely recognize that there is going to be a strong push for fast-moving execution and fast execution, like through teams. What I do see like an opportunity in, like very heavily, are those high-level operators that are going to be able to translate an idea into strategy and strategy into technical execution and documentation. What I seem to see a lot of is people like in companies have these like giant ideas and they like, we get on discovery calls or networking calls like, oh, I know a guy that's building in Cloud Code or in like Cursor or something. And they say, we don't like, you know, like we, we, we just paid for Clay or we paid for this thing. And I'm like, what are your 5 use cases for that tool? Why are you paying that much money? And I feel like the gap right now in the next 6 months is going to be people like us where we can actually translate the vision into strategy and strategy into execution. I don't know if you feel the same way on that though.

Yeah, I, I feel like, um, on that topic, the, um, yeah, the problem is, uh, probably very similar to the problem that was when we actually got started in working with AI, where like everyone talks about AI, but no one has a very like a clear understanding of a clear set of use cases where you can think about how to use that. Because if you think about that in general, you could say AI is useful for everything, right? But at the end of the day, if you think about that in the business setting, that needs to be translated based on your business and based on the specific context that you have about your business. Because you can think about using AI to write a LinkedIn post if you wanted to, right? But if your post then sounds like all the other posts in the space, it doesn't even make sense to actually write that. And so it all becomes a game of how do you actually get the best possible context that you can use for your own strategy, for your own team, for your own business, I would say.

Do you follow Jacob Deitel?

Yeah, yeah, yeah, yeah.

So this is somebody that I follow. I mean, I, we talk, me and you talk often enough for me to understand kind of like what's going on in the technical development world side of things. But with Jacob, I always see his posts and I say things like, Oh damn, like context engineering, right? This like idea of like engineering context and contextualization over a data layer. This is— yeah, this is so cool to me. And I think this is, you know, I was just on the phone with a startup and they're connecting the backend ERPs, they're connecting HubSpot, they're connecting the market intelligence layers, and they're not doing it through Clay. They're doing it through custom software and then they're duplicating that software across multiple instances for their clients, right? So depending on whatever the integration points are, there's a contextualization layer that's sitting in between all those datasets. And for me, I think that's very strong. I'm very bullish on this idea of like contextualization and why does it matter? Like understanding why does it matter to me is pretty much the future. Like anybody can, anybody can create a what, anybody could create an email. Just because you create an email doesn't mean it's going to perform.

Yeah, yeah. I will, I'll tell you two examples on that, which I think are really useful. And so one example is for me right now, recently what I did is I launched a school group about AI orchestration, right? It's like a free group. And obviously when I wanted to launch something like that, I don't have a full context of what the group is gonna be about, what are all the people are gonna be interested in. And things like that. So within something like a contact system, what I can do is I can turn every single conversation that I have and every single call that I have into intelligence that would help me refine the positioning of the group or the positioning about future offers that I might have, the positioning about my consulting and different aspects like that. And so that is one example. Another example, which I think is really useful, if you think about applying data in a go-to-market environment for a go-to-market agency, for example. So I have this client where we are building this out, and essentially what you can do is you can build a context system for every single client that you have. So within that specific client, you would collect specific things such as their main messaging, main positioning, different things like that. But on top of that, what you can do is you can also collect the different outreach campaigns that they launch. And so you can connect the two information, which is what you wrote within the outreach campaigns on one side, what was your hypothesis about that, what was the angle that you use about the campaign itself, and then you can collect that directly to the metric that you got within the campaign, right? And so in this way, what you can do over time is you can build a general understanding of what is the market, what the market is thinking about. And this also helps you refine the different campaigns that you are building for that single client, but also as an agency itself, you can think about how you can take the context that you have throughout all your clients and use that for like refining all your campaigns at the same time. So yeah, that's kind of a nerdy kind of topic, but yeah, that's, that's where I see this.

This is a, this is an interesting example because I've been recently brought under agreement to actually build something very similar. I wouldn't call it like ContentOS and I wouldn't call it every, like, I wouldn't say that everything is operating inside of Cloud Code, um, on the backend. We're connected to a data warehouse and, and what's happening right now is the data warehouse is sitting siloed from the CRM and the CRM is sitting siloed from call intelligence and the call intelligence is sitting completely siloed from market intelligence and the market landscape.

Yeah.

And it's my job to put together the strategy sitting between those 4 siloed data sources. And so that one contextualization layer can scan across all of those intelligence and across all that intelligence. Layers and all those data layers, but be queryable via like Slack or like a Slack bot. So like we're going to enable the sales organization to be able to query across all of their different data warehouses, all of their different, you know, intelligence stacks, but with one single input. So this is, this is where I see it, like in the, you know, $5 to $50 mil ARR band. This is what a lot of companies are trying to do. It's just very interesting to me that it's just consistent, like a consistent desire to contextualize everything across like unify the context across every data warehouse.

Yeah, I feel like we didn't actually have that possibility in the past. And one thing is that right now with, like, in the past, what you needed to do, like, for example, I come from a CRO background. So what you do within within there is you go within Google Analytics, you go within the product analytics tool, and then you analyze the tool, and then you analyze the metrics, right? And out of the metrics, you need to get some kind of insight out of that, out of the website. You need to like merge the two different things, merge the qualitative aspect to the quantitative aspect. And instead, right now with AI, I feel like any kind of thing that could be either transcribed or where you can have a text version of that. And right now, actually, with Gemini 3, you can even have video understanding. So like you could possibly even map out like expressions or things of that nature possibly. So yeah.

I think it's very interesting what we can do now. Uh, I mean, 12 months ago we didn't have this. We didn't even have MCP available en masse for things like call. I, not to my, not maybe like you were using it, but I didn't even start using MCPs until maybe 3 or 4 months after. I knew about them. I knew how they worked, but yeah, I don't know. I feel I was just talking to, um, a girl named Anna. She's, uh, in the, in the, um, go-to-market community and AI automation community, and she, she was showing me like what you could do with Claude Code and N8N. And, and I just said like, so who, who can build it faster, you or Claude Code? Like how to, to build a N8, like a, you know, a multi-agent chain in N8N. Yeah. She's like, uh, most of the time I just iterate in, in Claude Code and then I, I build it by hand and it'll be much faster for me to build it by hand than it would be for another system to be able to replicate the level of detail that I, as an expert, can give the system. And I was like thinking about it. I was like, interesting, but not, not far-fetched from what I believe in actually.

So is your idea that the expertise that you have, like there's still a layer between the actual expertise execution and the, um, yeah, and the, capability of executing of AI models that we have right now?

I think that there's a gap, realistically. I think that there's a gap between understanding the why, and once, you know, like, why— once an agent, you know, like when you build like a, you know, an AI agent, or you build an AI workflow step, I like— a lot of people just understand user prompt, or a lot of people, a lot of people that don't even like build agents don't even understand like system and user prompts and tools. From a UI point of view, from a consumer, all I'm going to do is give it a user-level prompt. I'm going to give it instructions as a user here, go and do this. When you start to build it in like workflows, you can set the system prompt for like, you are, and then you give it all the specifications of how to behave, right? You don't really give it the why, you kind of give it the how. And depending if you're few-shotting, if you're zero-shot, like, you know, stuff like these, like specific prompt techniques, then it'll understand more context. You still have to give it a user prompt to tell it exactly like why and how to behave and stuff like this. But with the experts that are building these, they already have the vision and the strategy of why that agent or that workflow is going to be designed. And I think like that part is being skipped right now. And a lot of like implementation plans, like people just want to like throw shit at the wall and hope it sticks. And this is like, you see this everywhere on LinkedIn. Like, I will be very open about it. Like people's, people have stopped, but for the most part it still exists. It's like, I made this N8N template and like, here, like download it. 752 fucking comments. And I'm like, yeah, guys, it's not going to work for your scenario because it's, it's built for a different why. It's built for a different reason. Like it's, it's just a template. It's like a recipe in a recipe book. It's just a guideline. It's not, it's not something that you're supposed to take N8N. If you like 2 cloves of garlic, put 2, but I like 8. So I just put the whole garlic bulb in. Right. It's like, it's used as a guide. It's a different scenario. And I think there's a miseducation between what experts are capable of doing versus what people are demanding from so-called experts. Right. Like, and I don't, I don't, I think I mix match your question of like, where do I see the gap? But I, there is definitely a gap between what experts are capable of doing right now versus what they're showing off for that top of funnel, middle of funnel content that's being picked up on people that need it.

Yeah, I feel like there are two teams here. One team is clarity of communication, which is like something that I started to think more and more about as I'm working with the AI. And so the thing is, if you say to an AI something like build me an NLP workflow, like the context is so broad, the idea is so broad that it doesn't have full context. And the more you are able to add the details into that, the more you are able to actually create something that is a one-shot kind of prompt where from one single document or from one single element, you can create a full automation. And so this is why, for example, for every single automation that we started to do with clients, we start with specs and we spend a lot of time on the specs so that once we spend 1, 2, 3 hours on the spec, the next step is almost immediate, right? And so yeah, that is one topic. And the second topic related to the workflow kind of craze, those kind of things like that. I would say that in like, I'm thinking about myself in the YouTube environment, like when I'm creating different videos and things like that. And the thing that a lot of people need to understand for that specific thing is you will never be able to balance out the creation of a proper production-scale AI agent within a YouTube video with the time it takes you to actually create the content. Because like, if I spend all my— if I actually spend the time creating a video where I do a production-scale AI system for a client, like it would take me like, I don't know, one month or something along those lines, right? And so that is one thing. And like, but when it comes to the templates is, yeah, you feel like it's, some people see the shiny templates and think that is the actual solution, but in the meantime, they don't know how that can even be applied to their business and they don't have context about that. So yeah, I feel like that's the problem.

I think what would be, what I've, what I've noticed at least, I mean, I've actually like unfollowed people because all they do is spam like NNN templates, which is like whatever. I mean, NNN is a great tool, but like the whole community there is like really funny to me. It's just hilarious. But the people that are commenting, right, like on this, uh, that are like asking for these templates, what I've noticed is that they are not usually the ones that are doing the implementation. They're the ones that are either in charge of the people that are doing the implementation. Or they're the ones that are trying to learn how to do it better. And I've— and this is like very interesting to me. I mean, you could scroll like 100 comments and just be like, oh yeah, you know, Claude Code, Claude Code, whatever template. And it's just very interesting to me how the perception of the industry is being perceived in my head as a, you know, as a fractional CMO, COO here. That supports product teams, it looks as if there's a lot of solutions in the market that exist that are templatized already that can be applied to like the exact business, you know, problem.

Yeah.

But in reality, when I actually go in and like build with these product teams or with these marketing teams, they are hyper-specific iteration loops. They're hyper-specific multi-agent systems that we have to build for their scenario, and those have to be scoped out in discovery. Discovery, and that takes a lot of time. And this is, I feel like the people that are trying to purchase agents or trying to, you know, take that shortcut, they're just going to come out on the bottom. What I've noticed that, you know, I know I'm rambling here, but I, what I've noticed is that the people where I actually do the proper discovery with, they already have an end desired state in mind. And then we can work backwards from their desired state to their current state. And it's my job to bridge that strategy. And I think that's what a lot of these technical people are not able to do. When you have like CTOs, they say like, we want to implement AI, we have an AI mandate, XYZ, XYZ. Okay, but how are you going to do it? What is like, what is the strategy? Why is it hard? Like, this is a good question for you. Why do you think it's hard for these very deeply technical people to translate their ideas into strategy? That is then translatable into agent architecture.

Yeah. Like, I feel like there's a, like a related to this question. There's a perfect example of the basketball players that then try to do the coaches and they cannot actually do the coaches because they are great at the game, but they are not great at teaching others how to be great at the game. And so I feel like that is the actual, the same thing actually, where you have people that are expert within their specific fields, but maybe they are not expert at extracting their insights that they are getting on what makes them actually successful so that that knowledge can be reused. And Personally, I think that this is a very, very important skill, especially going forward as agents will have more and more capabilities going forward. Because if you are able to think about what, like, for example, I'm thinking about myself when working with Danny Dan, when working like building automation fields like that, what are the patterns that I was using? What are some design strategies that I choose to apply over others? And so all these topics are things that you don't think about when you are actually building. You just go there and build, right? Meanwhile, if you need to work in the agentic way, you need to spend a little bit more time at the beginning to extract this information either out of people's head or out of your head and that so that the AI can learn more about that and that can become actually useful at the end of the day, I would say.

And do you feel like there's a context overload or an context underload when people start to design their first agent?

Um, I would say underload, like, uh, the, if you think about, uh, when you, like, if you think about creating a post, like, um, a successful person creating a post, uh, if the agent would have the same context about that specific person that is creating viral posts all the time, the agent would be perfectly capable of actually creating the same style of post with the same kind of aspect like that. So a lot of that, like that, at least to me, is a problem around the context. And so, yeah, maybe the agent isn't always perfect in giving you the actual solution that you want, but with enough layers of information that you add. And like, if we are speaking about content creation, that could be what is your positioning statement, what is the way that you write, what is example of previous posts that work for you, what are examples of short for short post, long post, and all different things like that. That helped more and more in creating the system. And like, I would say that also it really depends on what are you trying to build. Are you trying to build something that is like something simpler or just to get started and try to come up with that? Or are you trying to build a production system that is more robust and grows over time?

And okay, so realistically, when we talk about building like the, like a robust AI system, right? What is your definition of a robust agentic system right now?

Yeah. So I feel like there are a couple of things. First off is human in the loop. As you start, that is, you always need to have a component of human in the loop, especially at the beginning of building a system, and especially when you need to face either like business critical activities or client-facing activities. So when thinking about any kind of thing where like, I don't know, processing orders or thinking about creating outbound campaigns, I always want to control of the system before going full spectrum and go full agentic and let the system run itself. I always want to see what is the output looking like prior to going forward at different steps. Maybe I want to see, have control at the beginning with step 1 of a process, which like in the example that I mentioned you at the beginning was outbound system that searches, creates outbound campaigns and then pushes those campaigns. Maybe I want to have control on what is the kind of way that it searches, what are the number of leads that the system actually gets from me so that I can save credits on the API and things of the nature. So yeah, human in the loop, first thing. The second thing is self-learning and iteration. So as you build up systems, what you will recognize is you will recognize that agents inevitably fail like at some specific things. Which could be minor things or which could be like major things, maybe like get a research completely wrong or write an email that like doesn't really make sense. Your job, especially as the orchestrator of the system, is to always review the different, yeah, the different systems that are running at least once every 2 weeks. Or something like that. And then understand, is this running the way I want to? Is there additional things that needs to be fixed? Does an agentic system require more context? Does it have context that it shouldn't have? And all these things like that might create what is the actual robustness. I would say it is like a mostly about iteration because like all these things are about like, if you think about that, human in the loop is about like stopping the agent and letting the human verify before going forward. And then the actual like self-learning, self-improvement aspect is another thing like that.

Do you, okay, so this is something that I want to kind of like pull on here because we were pulling on the thread of contextualization, right? And it, it's like, um, how do we translate the obvious? So I, I know like this is a lot of teams and a lot of, you know, mid-cap companies, SMB, any business, they should reasonably, if they're making good money, I feel like they're gonna have some sort of processes documented. Yeah. They're gonna have sales playbooks, they're gonna have marketing playbooks, they're gonna have tone of voice, they're gonna have brand guidelines. Like all of that, those documents are pretty standard in companies that I work with at least. Um, I have to continue to do user interviews. I have to go and interview the sales team. I have to interview the AEs, the, you know, the BDRs, the SDRs, the VPs, the directors, the actual marketing operations people, the CRO, right? I always tell the people, I was like, I need you guys in a room, in a recorded session with me, and I need you guys to tell me the painfully obvious pain points that you guys are having. Um, because contextually, I'm in the dark here. I see the documentation, but I'm never going to get that level of granular feedback contextually to be able to build a better system. I don't, I don't know if it's something that you and like you struggle with, with your clients, like during that onboarding process and that discovery of like actually collecting obvious knowledge or obvious context. But I feel like this is something that a lot of things, a lot of people are like missing. Before they start to build a system. They're like only taking it off of like documentation. They're not taking enough of like added context.

Yeah, I feel like I, like at the beginning of my journey, I actually did this mistake where maybe you have a process, you have a project and you want to get started with that project as soon as possible. And so like, okay, you pitch an amount that you didn't, you don't even know Are you pulling that number up or something like that? Just because you want the business and you want to work on it. Meanwhile, right now I think it makes more and more sense, especially for relevant projects and projects that actually make sense to have a road mapping process at the beginning. A road mapping process could mean a lot of different things, but In our specific case, it could be interviewing the key people that will be involved with that specific process or will be involved with the automation of that specific process and getting more information out of that about the process. Because we don't know all the skeletons in the closet, right? We can only see what we see right in front of us. Especially with those businesses. Obviously with ours, it's easier because like you can see everything, you can know all the ways in which you are bad, but for other businesses, for client businesses, yeah, you definitely need this kind of level of clarity from the client and then like also trust from the client that you will build a system based on the full understanding that you're gonna get from there. So yeah, I feel like this part is become more and more important for every single project. And if a client doesn't want to do a project because it says, no, I don't want to do a roadmap because it doesn't make sense for me. I just want the price or whatever. Like it doesn't actually make sense because I'm walking in the dark with a project that I don't know nothing about, or maybe I know like, I don't know, 50% or 60%, but that, that other percentage, the unknown, could be way more than, than anticipated in terms of time, in terms of resources spent. And yeah, those things like that.

So failure points of like, Agentic builds, AI agents, workflows, where do you see it happening the most? Because on my side, I would say that 80 to 90% of the time when I build a system, the only reason why that system would fail is because of a lack of context. I don't think that there's anything that's gonna throw off an output as much of like in, in a single instance missing information. I feel like if it's incomplete datasets, incomplete instructions, that's missing context and that throws the whole thing. Do you have any other typical failure points that you've noticed over time of like messed up implementations where you've walked into the dark?

Yeah.

And so on and so forth.

I would say that another big problem for me is making sure that people understand how to execute the system because like not all the time we are building stuff that needs to be completely hands-off and just work magically like that is, right? So people actually need to operate the system and need to get started, like use the system and understand that. And so, and like another aspect, another component of the system would be something which I think could be called UX design, if you think about it, where the importance is how does the person that needs to operate the system uses AI or like understands the system or things like that. Is the person technical? Is the person not technical? What are the things that they want out of that? Do they have context about the things that they want out of that? And so like sometimes it's also like sometimes what happens is maybe someone has an idea, but they don't have full context about what they want. So they leave, yeah, they want you to actually figure out their own strategy instead of actually thinking about the implementation itself, which is like your expertise, I would say, like in some instances, right?

Yeah. Yeah, I agree. I think like people come to the table with an idea, but they don't think, I mean, they need an expert to connect the dots between an idea and execution through strategy. And I think like that is not going out of business. This this idea of executing is going to get faster and faster. Like, Claude code is crazy. It's, I mean, even with Claude MCP, I can do like crazy stuff or just Claude in general, but yeah. Um, I, I don't think that we're going to have an issue developing tools or developing software or developing code in the next immediate future. I realistically don't, I don't think people are going to have a tough— have trouble executing. I think, I think teams are going to lose their competitive edge and they're going to have to find new strategies and they're going to have to come up with fresh strategies to stay ahead of their competition because everybody's going to eventually have access to the same agencies, the same teams, the same vibe coding tools. It's just a matter of time. Widespread. I mean, it's already having like AI mandates, but I think strategy is moat. Right now and not execution. I think we're already seeing the flip.

Yeah. Yeah. Like knowing the what to build is more important and the why, like, and the does it even make sense to do that are, I feel like a very key question that a good strategist should always ask when like going into an implementation and going into like a person wanted to build AI systems. So yeah, those are the tricky questions, I would say.

What I always ask are, what are the edge cases that you guys have experienced before?

Okay.

Or like, that's like one of my discovery questions actually when I talk to businesses. Like, what have you guys done previously and what has failed previously? Why did it fail? And what are the expected edge cases as far as output or implementation? Like, you know, points that you're already foreseeing so that we can build around that and put in safeguards into the prompting, into the agents, into the whole system thinking diagram, stuff like that.

Yeah. Yeah. Yeah. Makes sense. It's, uh, it's something that, uh, I also started to do with, um, with the agents themselves when like building, uh, starting to build out like spec documents and things like that. Always map out the edge cases and always mapped out, uh, What are the key questions that needs to be asked to the team in order to determine more and understand additional edge cases that might come up? So yeah, that's key topic.

Cool. What is one piece of advice that you would give to either a person or a team of people that are looking to get started with implementing AI agents?

Um, start small. Iterate from there. Simple as that.

Any hot takes to close us off?

ChatGPT is going to fail to Google and Claude will become the business model for, the model for business. That's an outtake, man.

Okay. I won't disagree to it because, uh, I, I don't like GPT. It's, uh, but this is just me. That's interesting. I do. Why do you think that Claude is gonna come out on top though? Is it the same reason that I have? Because, uh, I think that they have, they're just, I think that they're, they're only gonna come out on top because they're first to market. Widespread adoption across multiple APIs?

No, I don't. I think it's a matter of focus. I feel like that, actually, I made a video on that. And yeah, if you want, like, you can, you can have a look at that as well. But the idea essentially for me was the following. Claude is very, very focused on their own use cases, which are mainly study with coding and they're all focused on coding is very deep. Like they don't, they don't try to mess around thinking about adding different tools or building different things. Obviously they have like Cowork, they have Cloud Code, but like their main, their main focus is around the one specific use case. Meanwhile, I feel like ChatGPT is becoming more and more like a mass adoption tool. And if you are working within mass adoption, when you have a giant that has unlimited resources, unlimited money, their own GPUs, it becomes quite difficult to compete with that, I would say. And so I see Google becoming more and more powerful in terms, and like also the latest models show that, like Gemini 3 is like kind of a very, very good model, like across all boards. And yeah, and that would mean that, yeah, Google is essentially on a big market that is focused on everything. I don't think you can out, you can beat Google at their play. They, I mean, it's kind of a, yeah, it's kind of a big thing, but Yeah, that's my own thought at least. And one thing which I would suggest you, which it's maybe interesting, is there's a podcast from Acquired that talks about Google as the AI company. And that is for me what made me think about the shift and thinking about, okay, Google is actually like might be on the cutting edge, might be winning the mainstream. Race.

I, I mean, I did a benchmark, a very, very small context-based benchmark on one scenario of writing outbound emails in, in, uh, one-shotting. And I tested Gemini, I think it was like Gemini 2.5 at the time, uh, GPT and Claude, and I put them head to head.

Yeah.

Just using user prompts, just like their base models. Gemini came out as the cheapest. Claude came out as the best written and GPT came out dead last. And that was like, uh, 8 or 9 months ago at this point. And, uh, and you know, now I have a Gemini subscription. Like Gemini 3 is just crazy. And then you have Nanobanana, which is really cool. I, I need to fool around with that more. And then, I mean, Claude is my daily driver.

Yeah.

Yeah.

So you agree with me?

I agree. I don't disagree with you. Okay. Uh, yeah, I, I would, I would say I'm, I'm team Claude all the way and then I'm team, I'm team Gemini, but of of course, like, when I go into— of course, like, when I go into, like, client meetings, I'm like, yeah, I don't— I, you know, like, whatever is gonna float the boat for you guys the best. But I'm always gonna be more, um— I'm always gonna tell them, like, what my use— what is the use case? What is, you know, what model is better for, you know, image generation, XYZ? What is the best model for this? I think this is where I still struggle, is like understanding— I mean, Claude has, what, 4 models right now, or 3 models, main flagship models? Like Opus, Sonnet, uh, and then in Gemini there's Flash 3. Okay. Uh, there's Flash, Flash 2.5, Flash Pro, whatever. And then GPT, it's like 5, uh, 5, 4. Okay. What is the difference? Like contextually, what are they good at? What are they worse at? And I, I've only seen one tool and it's called OpenRouter. Yeah. That like auto, that automatically switches between like the request, like the contextual request, and then it categorizes the request and that routes it to the right, uh, prompt. Or the right system. I think this is super cool. Because like, even for me, like I can't keep up and I'm like sitting right there next to you at the top of the curve, but there's just so much stuff flying around that I just don't know. Yeah.

Yeah. I feel like right now, like having a, trying to do a comparison, it becomes more and more difficult for these models because like, what happens is they are all at the like 90 percentile or whatever kind of percentile. And so like if you show the, if you see the benchmark, you don't understand anything as a like end user that needs to use the model. And so at the end of the day, because like the most important thing is if you use them day to day and you can get an understanding of like what is the vibe, even though like there is no technical term, you can get an understanding at least of like what you feel like is working for you and what you feel like it doesn't work for you. So yeah, at least for me is mostly about like testing out and like, to be honest, I don't, I'm not the guy that tests out a model and then after like 5 minutes to say, okay, yeah, this is like the newest model. So whatever the case might be. Like I'm the person that said when ChatGPT-5 was released, I don't like this strategy. I feel like this strategy is geared toward reducing cost instead of actually providing a better model. And so yeah, that is why I'm also team Claude and Gemini.

Got it. Awesome. Okay, so I think we've been chatting for quite a while. So I have one, only one question left for you. Where can people find you? I know you're on YouTube and I know you're on LinkedIn. Is there any place else that people can find you?

So yeah, people can find me LinkedIn, YouTube. Right now I'm building a free group, which is called AI Workspace Builders. Essentially what we do is we teach people how to build AI workspaces. So how to get, get context, how to execute. Task directly within a cloud code interface and make it that your daily driver for whatever kind of business you are in, especially service-based business. So yeah, that's another cool place where I'm focusing a lot of attention on.

Cool. Well, Riccardo, awesome to have you on. Awesome to chat with you again. I feel like we don't do it much anymore since we're so busy, but appreciate you coming on the podcast and Yeah, catch you on the flip side.

Yeah, me too, man. Bye. See ya.

Ciao.