In short
How the hosts build and use “AI stacks” (personal operating systems and an agency “creative strategy operating system”) to turn marketing/service work into predictable, compounding workflows; includes influencer marketing systems and AI ad production (video/static).
Guests
No named guests. Two hosts/partners discuss their own systems and reference external creators (Andre Karpathy, Liam Otley, Carl Weiss, Vin Vin).
Guest backgrounds (as described)
Operators building an e-commerce/marketing agency; one focuses on personal “second brain” using Obsidian + Claude; the other builds an operating system in VS Code and uses it for management, call/Slack transcription, and daily coaching/journaling. They also run internal platforms for client creative, reporting, and forecasting.
Key claims
Context compounds while models don’t; build data/context layers before automation; human feedback loops train reasoning traces; influencer winners replace one-off campaigns with seeding + affiliate conversion + retention systems.
Notable examples
Landing-page screenshot → instant critique/audit; daily Slack “VA” + journal prompts + imposter-syndrome patterning; AI management recaps from call transcripts; ad “framework database” that hypothesizes why top creatives worked; AI videographer constrained to ~40–45s scripts and 10–15 scenes; trigger-based auto-reporting when metrics like incremental reach or hit rate drop.
Written by AI. May contain mistakes. Listen to the episode to check what was said.
Chapters
Tap a time to open that second in VOReflecting on Team Development
0:45 to 2:00
Discussion about the team's recent social event and strategic direction.
“And outside of that, how we are going to, the service industry is very much going to flip from paying, you know, you get paid for your service to you're paying for an outcome.”
Shifting Trends in Service Industry
2:00 to 4:00
Exploring how service industries are evolving to focus on outcomes.
“it was a powerful ballad yeah but good night nonetheless is influencer marketing the most unpredictable line item in your budget Big month, dead month, big month, dead month.”
The Rise of Predictable Influencer Marketing
4:00 to 6:00
Insights on building a systematic approach to influencer marketing.
“So super excited to dive into some of that today.”
AI Integration in Daily Life
6:20 to 8:13
Exploring how AI is integrated into personal and business operations.
“So I think you've functionally created it in a slightly different way to myself.”
Building a Personal Operating System
8:13 to 10:40
Discussing the creation of a personal operating system using AI.
“Yeah, I think that Obsidian Vault is a really visual way of creating that database.”
Harnessing AI for Enhanced Contextual Understanding
10:40 to 12:20
Using AI to improve decision-making through contextual data.
“that will be able to drive much better outputs as a result of that.”
Advanced Management Techniques with AI
12:20 to 14:00
Leveraging AI for effective management and feedback systems.
“So again, it's that idea of avoiding context blow and self-updating each of these context buckets by providing it with the tools to drive those inputs.”
Leveraging AI for Effective Management
14:00 to 15:46
Learn how AI can enhance management practices and personal development.
Using AI for Client Growth
15:46 to 17:18
Discover how AI can optimize client interactions and improve service delivery.
“So yeah, I think the context for me has been very client-led, growth-led.”
Personal CRM and Networking
17:18 to 18:44
Explore how to build a personal CRM for better relationship management.
“I also am using it in a similar way to yourself, I think, from a personal level.”
Show all 25 chapters
Daily AI Assistance and Reflection
18:44 to 21:06
Understand how daily AI prompts can facilitate personal reflection and growth.
“And actually it's almost there as like a reinforcement loop of, for example, if you said something, um, did this recently, like I was, I felt, you know, imposter syndrome about this one thing.”
Feedback Loops in AI Utilization
21:06 to 24:15
Learn about the importance of feedback loops for personal and team development.
“using it using using ai as a feedback loop in in situations where you struggle to generate that I think founders or people who are in senior positions of a business get less feedback.”
Building a Strong Data Foundation
25:09 to 28:06
Learn how to create a robust data engine for better marketing strategies.
“I guess we've also gone through this process internally within the group.”
Refining AI-Driven Content Creation
28:06 to 29:00
Learn how AI improves content quality through iterative feedback and learning.
Building a Centralized Data System
29:00 to 30:27
Explore the architecture of a centralized data system for better insights.
“to improve the skill and that reinforced learning process over time to improve that output is really valuable.”
Automated Insights and Reporting
30:27 to 33:10
Understand how to automate insights and reporting using AI-driven frameworks.
“So writing great scripts, reporting, et cetera.”
Integrating Human Feedback in AI Systems
33:10 to 34:28
Discover the importance of human input in developing effective AI systems.
“They think, how do I automate the intelligence as much as possible?”
Forecasting and Strategic Insights
34:28 to 35:34
Learn how to enhance forecasting with strategic insights and reasoning.
“So the human is the teacher, not the user.”
Building Adaptive Creative Strategies
35:34 to 37:36
Explore how to adapt creative strategies using AI insights and adjustments.
“You can do that with Claude quite a bit as well, just if you're using Claude rather than this system, you can do that to the MD file, right?”
AI in Video Production: Current Practices
37:36 to 42:00
Gain insights into the practical applications of AI in video production.
“Yeah, it's a really smart way of working in terms of building that out.”
Optimizing AI Ad Creative Length and Scenes
42:00 to 43:09
Learn how to effectively set constraints on AI-generated ad scripts to enhance efficiency.
“I think the sweet spot really is for us like 40 to 45 seconds and under kind of 10 to 15 scenes if you can but yeah, more to come on that.”
AI Execution in Ad Production
43:10 to 44:29
Discover how AI is transforming ad production and reducing costs while enhancing creative diversity.
“Some funny cartoon ones floating around as well so we can share more as that comes to light.”
Scaling Static Ad Generation
44:30 to 46:16
Explore the process of mass generating static ads using AI frameworks and its impact on client output.
“Again, this is something that we're going to want to build into our control room, which is obviously going to house our growth compass, which they would have seen that episode down, right?”
Business Intelligence for E-commerce
46:17 to 47:50
Understand how a business intelligence platform can enhance client performance analysis and resource allocation.
“Click generate and it will just run this back-end process where it will generate those at scale.”
Leveraging Data for Strategic Insights
47:51 to 50:35
Learn how to identify triggers from data to drive strategic decisions and improve ad account performance.
“Another example is potentially hit rate.”
Transcript
Automatic transcript. May contain errors.0:00Olly:I've got it into such a good spot now where I can literally just take a screenshot of a landing page and just pop it in and it'll just give me like the best like critique and output for an audit.
0:15Olly:We've not done a joint episode in a while. Another little brief hiatus.
0:18Loukas:I feel like we say this every episode now. It's been a little bit hit and miss just with obviously the office move. We had 80 plus people in the office last week for the group social, which was a really great day. Working on our new EVP across the whole team. How did you find it all?
0:36Olly:Good, yeah. Good. It was good to be a bit more forward looking versus backwards and reflective this time. And just share a bit about the strategy and direction we're going in as a business. it's such an interesting place to be owning a service business right now in like trying to create a really strong strategy with so many like variables outside your control which I think we've done but it was interesting, yeah, great to share a bit of that with the team, I thought so yeah, it was a good day
1:11Loukas:I think yeah, it's interesting looking forward And I think the big bets that we're obviously, you know, talking to the team about and reverse engineering around, you know, how software and service, you know, the lines will become more and more blurred across both of those. And outside of that, how we are going to, the service industry is very much going to flip from paying, you know, you get paid for your service to you're paying for an outcome. so it was yeah good to get the team brought in on the direction and I guess some of the steps that we need to take to get to that point good karaoke for showing from yourself as well yeah broken strings James Morrison don't worry about it mate maybe that's why the voice is gone right now put too much heart and soul into that one yeah it was powerful it was a powerful ballad yeah but good night nonetheless is influencer marketing the most unpredictable line item in your budget Big month, dead month, big month, dead month.
2:13Loukas:You feel like you never know what's actually coming. The brands winning at this have stopped doing one-off influencer campaigns. They've built a system instead, seeding new influencers, converting the best ones to affiliates and retaining the top performers. This way, your pipeline never stops and revenue gets predictable. To build that system, you need a platform. That's Saril. Find influencers, reach out, manage every relationship all in one place without all of that chaos. 200 plus brands including Grunz, Soulwave, Space Goods and many more are running their influencer programs on Saral. If that sounds like what you've been looking for, go to getsaral.com slash demo and book a demo with their team.
2:54Loukas:Tell them we sent you and you'll get your$500 discount and white glove onboarding. Cool, today we are, we are, I mean, what's crazy? So we're doing an episode today, just a bit of a spitball into AI and like how we are just using it day to day, both in our personal lives, as well as in the business. I think we did this, what was this, about six months ago now?
3:20Olly:Yeah, maybe a little bit more, but yeah.
3:22Loukas:Six, seven months ago. And we were just talking about some of the tools that we were using back then and how we were kind of integrating it into our lives. I mean it was like a 25 minute episode that was supposed to just be we didn't really put much time or effort into it transparently but actually was a super popular episode because I think there were lots of actionable takeaways and ideas that people could apply to their own lives I think what's crazy is just if you then went back and looked at that episode it would feel like basic stuff now so it just kind of shows the compounding changes that we are going through So super excited to dive into some of that today.
4:05Olly:Yeah, it's interesting to... We're actually hosting an event at the office next week for women in e-commerce and helping them in collaboration with a couple of partners, helping them with their AI journey and supporting them with implementation into their business and into their role and education around that. and looking at like the content for that day it's interesting to see how like people go through like the adoption phases of someone even using AI themselves I think when we probably did that last episode we were still at that stage of like using chats and using tools but now we've really gone through a journey of creating like code, connecting data, rebuilding process and yeah just just creating like net new output and and really i just think that competency level internally is just elevated s like come so far since that previous i think i think what you'll
5:12Loukas:see like the difference is we were paying for a lot of softwares then in terms of almost not wrappers necessarily but tools that now you know by building your own operating system you can have far more control of and still do a lot of the things that that you know these individual sass products might have been able to do um so it shows you know that that kind of shift towards customization and building something bespoke around your own lives and your own you know desired outcomes so yeah let's start on the personal operator operating system piece um i know we've both been building some stuff here slightly differently to one another um and yeah how are you using this right now or what gives maybe some top level context of what that looks like for you?
6:00Olly:Yeah so Andre Carpathy I always get his name wrong I should know that by now Twitter OpenX OpenAI just like forefront of the of the industry put out a tweet probably about four or five months ago on like how to create like a second brain or personal operating system so So I think you've functionally created it in a slightly different way to myself. But for those watching, we can link that tweet down below. He shared a GitHub that walks you through this process. So if you're wanting to build what we're going to go through for yourself, I think that would be a great place to start. And he uses a tool called Obsidian, which is almost like a visualized neural network of context that you ingest into it.
6:52Olly:The reason it's so powerful is because it categorizes that data. You can set it up to connect to Claude code, so you just ingest via Claude. You can also set automated triggers that pushes into Claude and then it pushes it into the Obsidian Vault. His process has like a wiki attached to it that teaches the model how to use and how to ingest that data correctly. It links, it categorizes it by topic. So I've got like post-click, media buying, creative strategy, and all of these things. And then within that, there's multiple different pieces of context or data that I've ingested, whether that's resources, whether that's just call transcripts, whether that's QBR decks, things internally that have then started to link together.
7:42Olly:And it acts as this like retrieval database when you then prompt Claude. it references the knowledge base, will retrieve data around, through that network of information and then surface an answer. And the level of output that you get relative to just like normal prompting is just insanely valuable.
8:08Loukas:Yeah, it's that idea of like context compounds, but like the models themselves don't. So the more that you can invest in creating a system, that allows itself to compound in context that is related to whether that be domain context or business context, whatever that might be, the better that you can build that and then alongside that reasoning traces as an asset as well. So like, for example, if you make a strategic decision in the business saying what the decision was, why you made that decision and what you're going to be measuring that against, as an example, it will be able to actually store that context and as a result slowly start to understand the reasoning and why we made certain decisions and slowly be able to make better judgments and better outcomes as a result of that.
9:04Olly:Yeah, I think that Obsidian Vault is a really visual way of creating that database. It's crazy how sometimes I put a piece of our IP in there and it connected to so many things and it brought it all together into a super node which was really interesting it was like this new operating system we've built but yeah I've been just building that out over the course of like four to six weeks now and it gets better and better as I'm then using it for like the post-click part I've got it into such a good spot now where I can literally just take a screenshot of a landing page and just pop it in and it'll just give me like the best like critique and output for an audit that's so valuable and then i could just turn that into a brief without really having to do much like work because it's already ingested like the way i would go about that process myself um you can get context from so many different places i just went through a lot of thought leaders um in that sector and space and ingested a lot of their best content um that was really valuable yeah i think what this comes back to it's like you know if
10:10Loukas:you're quite early on in this process and i'm going to be doing a mastermind next week that goes goes i guess a step back but then showing you how you can get to this end state of like building an operating system um is really deeply understanding some of those basic fundamentals like how to create context documents yeah or build context that relate to why the domain or business um and then also things like skills as well so like once you've got skills and understand how to create skills in Claude, you can then quite easily build that on top of an operating system that will be able to drive much better outputs as a result of that.
10:49Loukas:So I've built a similar system slightly differently to you. So I used VS Code, which is basically just a source code editor that similarly connects and ingests with Claude Code. So again, what is useful about that is you can see the whole hierarchy of the operating system that you've built and the different elements that bring that together. I'll show a screenshot right now. And I actually want to shout out Liam Otley, who is an AI YouTuber. We both know him. We met him a few times. He's blown up in that space probably since we last saw him. But he's got a whole video that goes through like how you can build your own personal operating AI operating system, which is super super great and um it's a great video just to like get started um i can talk about maybe some of the use cases that i've done on top of that so obviously in terms of like the parts of context that it will require from from you within this this kind of hierarchy it's got it's got in-depth context and all of the brands um that that uh we we we run or that we own all the people that live with inside those brands, all my direct reports as well, and then domain context around certain specialisms within the business.
12:12Loukas:There's also a learnings context document, which again will update itself based on decisions that we've made in the business and why. So again, it's that idea of avoiding context blow and self-updating each of these context buckets by providing it with the tools to drive those inputs. So for example, I've got it connected to my AI notetaker. So every call that I have every single day, it will essentially transcribe every call. It will then take that and it will break it down into certain triggers or buckets that I've identified, such as this is an action item that needs that we need to add to your to-do list this is a strategic decision that was made in the business this is actually an update to people that we need so for example we're talking about someone leaving the business or someone getting promoted so then it will send me a recap at the end of every day saying these are all of the things we need to enrich your context documents with here's the category here's the update and then you essentially just say yes or no to each of them and it so every single day it's constantly updating according to your meeting transcripts, but then also your Slack.
13:31Loukas:So it is in your core Slack channels as well. So my favourite use case from a personal perspective that has been really useful so far is actually on the management side. So it has a deep, deep understanding of every single person that I manage within the business. It knows that we have a skill for how it can build quarterly reviews so it has a deep understanding of how we build quarterly reviews into notion what we score against um every single piece of feedback i've given that person it will obviously log that if it's on a call or in slack it will know to log that feedback and it will identify patterns over the period of time it also will cross reference against the last review that you did with that individual to see where they are improving maybe where the the points that they need to keep growing are and then I guess the domain context that I've also built into that is like my favorite um management books that I think are incredibly uh that create this kind of really useful layer of uh context before if I then need someone to to get some feedback on or or you know ask for some um advice on maybe a certain situation that's happening I've also it's also transcribed a of Leila Hormozy's videos and she's great with like leadership ops management so again what it will help me help me do is actually through its vectorized database it will be able to pull elements that are really relevant to what I'm working on and give me advice as if it was like Leila Hormozy or as if it was the person that had written one of these management books so it's a it's a great way to kind of continue to become better as a manager and a leader but also at the same time um you know ensure that we are identifying some of those patterns with those that we manage so we can become more intentional uh in in the way that we support them it's also trained on clifton strengths personality tests so it will adapt the way that you uh and tailor the way that it speaks to to each individual as a result or that give guidance on how best to to give them feedback um so that's i think that that piece is really valuable and that kind of self-learning system through decisions through context um has has massively helped me yeah i
15:54Olly:think that's a really intentional way of using a process like this for a specific outcome it's also like an area where it's like it's very easy to think of ways to use ai and like the execution but and the people process maybe i imagine a lot of people watching aren't aren't leveraging it to that level of intentions i think it's super impressive system i haven't really i think where Yeah, I've been on the other side of just mainly using it for execution and strategy. So yeah, I think the context for me has been very client-led, growth-led. Yeah. Starting with fundamentals. When I'm thinking about getting the most out of this process when using it for a client, you really do need to start with personas, with brand, even if you're doing work on landing page post-click.
16:44Olly:because I think when it has that context and that anchor to support with the output, it really produces just like far greater value. But I've gone through the similar process. Shout out Carl Weiss. I just ingested like 50 of his CRO videos and like documents that he's published online and it just built this amazing engine for offer optimization and post-click optimization, which isn't something we do like direct service, but it's massive lever when consulting with clients on unlocking further growth. Yeah. I also am using it in a similar way to yourself, I think, from a personal level. I've started to build out like a personal CRM through this process.
17:28Olly:So like a network, like visualizing your own individual network and using it as a way to better manage relationships, which I think is really interesting. apply a similar way to the similar view to like the management piece but on a personal level there's a few gaps in there like WhatsApp for example like not being able to fully connect every point of communication but I think they're very solvable over time
17:53Loukas:WhatsApp have API access oh no because I guess it's encrypted isn't it
17:57Olly:yeah it's a bit of a difficult one you can get it into OpenClaw but I'm not using OpenClaw so I've not actually I've not spent enough time going deep enough in to have to fix that problem yeah but um yeah just like better structure and management of that data um i know you've got a bit of a slack process from like a daily perspective as well
18:17Loukas:haven't you yeah it's quite funny actually because it's um created a uh a slack member called lenny um who's actually my my chihuahua or my fiance's chihuahua so it's just his little uh his little picture but he he's essentially my um he kind of almost like a va in many ways so i get a slack message daily with my to-do list for that day a breakdown of my calendar and where the gaps are and where i should be prioritizing my time as a result of that it gives a daily update on business figures as well across all of the businesses um and i think actually like my favorite part of it is i mean one is great because like obviously if you're just out you can just tell it to um you know if you remind me to do this tomorrow or next friday at 2 2 p.m it will just be it will just log that and then then add it to your reminders um but sec i think the the most valuable way i've been using that is at the end of every day it will it will give me some journal prompts so what it will ask me is like five questions every day how you feeling today what has impacted the way that you've felt and why um are there any kind of key learnings or insights that that you can take forward into tomorrow uh another thing that i've asked it to do is ask if i have felt um imposter syndrome that day because for me i've as the business has kind of scaled as quickly as it had I kind of fell into this trap of getting imposter syndrome more and more and I was very fascinated with identifying the root cause of that and when I feel it and why to then look to how I can improve in that area and as a result become more confident so what's great is like if I feel imposter syndrome that day I will log when I felt it and why and what is great is it's like like started to identify patterns in what, what, why I feel that.
20:22Loukas:And actually it's almost there as like a reinforcement loop of, for example, if you said something, um, did this recently, like I was, I felt, you know, imposter syndrome about this one thing. It pulled a quote that you said on a call that kind of, I guess, completely countered that, that me feeling like an imposter. And that's like, I think when you're, you know, it's that self-worth piece of you always mute the good things, but actually like you know you're very self-critical so it was a great kind of reminder of oh actually you know i am in my head here so yeah it's kind of like a bit of a bit of like an
20:58Olly:empowering mentor as well in some ways that's a really interesting i think that's a really interesting example of a concept that i've been talking to quite a few of the team about is like using it using using ai as a feedback loop in in situations where you struggle to generate that I think founders or people who are in senior positions of a business get less feedback. Yeah, exactly. And that can often, I think that can somewhat put like a ceiling on progression, your personal progression. Because I think it's, so things like in calls, when I'm doing sales conversations, built skills that take my the transcript and give me feedback based on like a framework of like best practice for that yeah um i think that's really powerful for for agencies because it's impossible to shadow every like team members call um really hard to give direct feedback across an ever increasing number of individuals consistently so i think that's that's a really good example of that on like a personal level that I think people can leverage yeah there's that um Vin
22:09Loukas:Vin uh gang I want to pronounce his name correctly but he's like a communication expert um sent you his podcast he's been on Dara of a CEO as well as a few others but you've got a great course uh all about presenting and communication I think another great example of what you just said there is um he's got loads of YouTube videos right so a great if you're struggling if you struggle as a presenter or communicator which is one of the most important parts of you know driving authority or ensuring that your uh your message is communicated in the correct way to the best possible outcome then you know you may want to work on your presenting or communication skills as you said if you're a leader in a business or you're someone that doesn't get that feedback loop as often as you would like.
22:58Loukas:Literally go to his YouTube channel, just transcribe like maybe 10, 15 of his top videos. You could easily build a really strong skill that reviews all of your transcripts from your calls, puts it through that skill to identify based on what he says is a great communicator. And he gives lots of actionable tips on how to improve in certain areas. It will give you direct feedback on your transcript and maybe where you've used filler words, how you could have said things slightly differently, where you were repeating yourself, where you were like mumbling or kind of exaggerating certain areas. And it will kind of, it's that constant feedback loop that will just make you better at that one thing by being truly intentional.
23:43Olly:Yeah, I think that's really powerful. It's definitely, I've definitely seen it drive positive change within specific members of our team, even like stripping out hedging language. I say it's like, the concept of like how you show up really matters in in in client relationships less so if you're in an econ brand i'd say but maybe if you actually progress up a business i think um how you show up is still very important um but yeah it's really difficult to generate that feedback yeah is
24:15Loukas:influencer marketing the most unpredictable line item in your budget big month dead month big month dead month. You feel like you never know what's actually coming. The brands winning at this have stopped doing one-off influencer campaigns. They've built a system instead, seeding new influencers, converting the best ones to affiliates and retaining the top performers. This way, your pipeline never stops and revenue gets predictable. To build that system, you need a platform. That's Saral. Find influencers, reach out, manage every relationship all in one place without all of that chaos. 200 plus brands, including Grunz, Soulwave, Space Goods, and many more are running their influencer programs on Saral.
24:56Loukas:If that sounds like what you've been looking for, go to getsaral.com slash demo and book a demo with their team. Tell them we sent you and you'll get your$500 discount and white glove onboarding.
25:09Olly:Talked a lot about personal there. I guess we've also gone through this process internally within the group. And we've got to a position now where we've just created a really strong foundational context and data engine that's making building on top of that layer more and increasingly easier. So yeah, I spent the last six months building this out, haven't we? Like ingesting platform data, creative data.
25:48Olly:client data, call data, refining our processes so they're easier for AI to pull through from the various places in which we store them. It's really exciting to see how that's coming together to create value for the team, both through easier processes like reporting, processes like new ways of doing QBRs and strategy decks and surfacing more value, speeding up execution, improving strategic decision making I still think we've got a way to go but yeah I definitely feel
26:24Loukas:like that value started to come through yeah 100 % I think most people jump to like function but really like it needs to start with data and context at a foundational level because I think once you and it's just about best practices right now like if you can if you can effectively set up a data warehouse or a data layer and context that feeds into domain and also, you know, brand, then you have the foundations to build anything on top of that. As long as you've got the idea of how you do that, then, you know, you're able to quite easily build on that. But it's really important that we are, you know, we're going back to those foundations before you're kind of jumping to, oh, I've got this cool idea that could do X.
27:09Loukas:because I think people just get a bit lost in style over substance without any kind of proper back-end architecture that will allow them to get the best possible outcomes.
27:19Olly:Yeah, it allows you to move a lot quicker once you've got that foundation in place as well. Building tooling becomes really much faster than when that foundation isn't set.
27:31Loukas:Yeah.
27:32Olly:We've built a really interesting process across our marketing data, so every time we do these episodes, we have created a process where we've built a context layer for the for the for our marketing output it ingests all of our lead magnets podcasts posts pulls them and transcribes them and then if it's a podcast it'll push it into a notion process where it'll create a draft of like here's four ways we can distribute this podcast episode through written output that draft is then taken by a team member who then refines the draft and then creates the actual post and then it compares the final version with the initial version and then um the difference between the two it creates feedback and then pushes it back into the the process so it's improving the quality of output itself over time it's like doing its own reinforced learning built into the process that we've set up which I think is really interesting I think you can apply that to other areas I know you're going to do a Claude sort of skills work, a masterclass but like script writing reviews comparing using a skill to compare feedback on like final version versus initial version using the gap between the two to generate a view of what was different and how that should be ingested as a way to improve the skill and that reinforced learning process over time to improve that output is really valuable.
Read the full transcript
29:07Olly:100%. Yeah, I've started building this for Club Neuro as a,
29:14Loukas:I guess there's like a test case before we tie this into the control room, which is essentially like the centralised place where all of our data lives across all of our client accounts. um now i think the the the key to like when i was building this because we've got the skills built out like for example script writing skills and reporting skills that can do all of these different things but i think what is super important is the compounding learnings over time and ensuring there's a proper process in place for that because that's where things will improve and get better as we know there are nuances to every brand and things are always changing in terms of what's working and not so the more that we can kind of automate that or get into good habits of how we optimize those learnings then the the better the system will become over time so essentially like what the architect so it's like an actual web app operating system uh it's built into like four kind of core layers it's function first which is the web app itself um we've got the dashboards that are built into that so that is pulling through from like klaviyo um shopify meta etc then the next level is data which is actually then it's storing all of that data in our warehouse on the back end through super base then the next layer is context So that's everything from like the brand DNA, having all of a breakdown of all of our macro, micro personas, obviously all of the skills that are associated from a domain point of view.
30:51Loukas:So writing great scripts, reporting, et cetera. And then the foundation is the Claude MD, which is essentially the operating system's brain and how it ties everything together. Now, if we look at some of the, firstly, because it will be easier to reverse engineer, but some of the features that are on this website, one is there's a best performing frameworks dashboard. board so every time one of our a new creative hits top spender it automatically runs a workflow where it will download itself from meta it will upload it into gemini it will review it will hypothesize why the ad worked and why it didn't cross-referencing to soft metric benchmarks and then it will also pull the whole framework transcript and then we'll upload it into our super base back in warehouse and then into the actual web app system.
31:48Loukas:So every creative will then live inside there. It'll be embedded within the database. But the great thing about that is every time a new top spender comes in, it's hypothesizing why it worked. Are there any new insights that maybe we haven't seen across other creatives? Even down to things like, oh, you know, we see now that there's blended NCPAs are 20 % lower when we're targeting this persona over this persona, as an example. It will be able to extract those insights and build the framework database, which again is its own data layer that feeds into the intelligence of the script writing skill.
32:27Loukas:Similarly with a hooks database, so anything that hits 30 % TSR or more, it will store it, hypothesize why it worked. And then the other parts of it is the reporting ones. There's a one-click reporting button where you can do a whole analysis of the ad account, a gap analysis from a valence zone point of view, persona. Similarly with Reddit, it does like its own weekly Reddit scrape of certain keywords. It will pull an analysis of if there are any new trends uncovered or potential gaps from a persona point of view. And the key is that this then is all feeding into that intelligence system. So it's updating its learnings as a result.
33:05Loukas:But I think the one thing that I want to say here is like what most people do when they are building these systems. They think, how do I automate the intelligence as much as possible? So it's self-learning, but I have to be hands off. I've actually realized through experience that you want to almost build a roadmap for the manual inputs over a period of time and cue seeing what the outcomes are. So an example of that, every one of these dashboards, so let's say the framework database if a framework comes in and suddenly there's a new um framework within that database it says this is here's the winner here's why it was a top spender here's our hypothesis here are some of the insights as a result of that i can within within the web app go and edit the edit something and it will pop up and say why did you make this change um you've gone from this to this why did you change that and i would essentially it will be the reasoning in which the or the knowledge that i have that the system didn't as to why i i made a certain decision or why i changed something and then it will store that accordingly those reasoning traces to to improve its output next time i think like doing that any other way than manually initially isn't going to get you the best output and it's not going to compound over time.
34:30Loukas:So the human is the teacher, not the user. So every correction that we make within the system is editable and it's training data and the UI is essentially designed to capture that reasoning at every part of the process. And then what that means is when it feeds all into the co-pilot, which essentially is like our own LLM for Club Neuro, the outputs are so much stronger because it is constantly updating its learnings as a result of that.
35:00Olly:Yeah, I think building that human feedback into the process is a really smart way to work. We've done that in the forecasting process as well. I think we've built within the control room, which we'll be rolling out over the next couple of weeks externally as well as internally is what I think is like a best-in-class forecasting approach for brands using data that we have available from all of the channels visualizing that in a sass but within any forecasting process you need to do some manual manipulation to the to the output that you get from the initial data view and it's it's making our strategists and those who are building those forecasts explain those amplifications why things have been changed doing it at the point at which it happens and then that that building a better model over over time with that internalized into
35:55Loukas:the process exactly and i think whatever you're building it for like the four layers that i would look at which is kind of what you've just described there the first is performance automated performance into hypothesis loops so let's say your example you know that's that's automated it's it's the performance is driving an automated hypothesis of why something worked then it's the hypothesis into correction loop so that's where we review the hypothesis and make changes if required that then adds it adds that reasoning to its own context moving forward then you then it's correction into rule loop so let's say that something is continuing to happen even if i'm in co-pilot i say you know let's say it's a total voice thing we could say oh no we won't ever use a word like that so save this to our rules.
36:48Olly:You can do that with Claude quite a bit as well, just if you're using Claude rather than this system, you can do that to the MD file, right? Exactly, I think
36:55Loukas:the only problem with it on Claude side that I've seen when you're actually using it inside the LLM is like, because of the size of the context window, it sometimes gets bloated and forgets but yeah, so correction into rule loop and then the final one is decision into outcome loop, so essentially what that means is another example is like let's say you've got a creative strategy dashboard where you've got funnel positions and an example of this with club neuro was we've got a strategic dashboard in the operating system so it's a pie chart that will show you know the splits from problem unaware all the way down to most aware we were over indexing on top of funnel and problem aware and we had an insight that basically suggested that we're getting a lot of ad comments that are either uh objecting on price or they're comparing us to competitors so that told us that we need to go we need to actually increase solution aware by 20 30 let's say so i made that change on the dashboard it asked for the reasoning and i told it that now that's something that AI would just not identify without the human in the loop aspect, but then it would identify that trigger
38:11Olly:later on down the line. Yeah, it's a really smart way of working in terms of building that out. Yeah, really impressive. I have no extra builds. I think, yeah, I think we're building that into the process in that way at the point of because we were thinking a lot about how do we capture those reasoning traces and it is a really difficult thing it's like trying to capture intuition yeah like because i do think i was chatting to sam the other day about this um that you can often go into an ad account when you've got a degree of like experience um and i do think a portion of media behind is intuition led um but it's like how do you codify that into a logic that then can be leveraged by ai but doing at the point of um through that feedback loop in the way you've described as a really really like organic way of capturing it yeah i think it's like it retrains
39:12Loukas:our brains doesn't it in many ways because i think as you say like a lot of these decisions we make subconsciously just from having you know real experience but it's like how do you what I've found interesting is actually tying it back to a certain trigger is that trigger data is that trigger yeah prior domain context or experience from another brand you worked with five years ago where you saw something worked and actually that that's what triggered you to do something here so it's like you know it's sometimes it's difficult to actually tie it back to that but it's interesting
39:48Olly:nice so moving on to like creative and what we've been doing internally I want to have a chat at the end around like where do we think video and image gen is going to be in like 6-12 months time again but we've obviously brought on an AI videographer we're doing a lot more AI static generation how are we approaching that currently? Yeah we're going to have a
40:13Loukas:meaty episode going into that end-to-end process we're going to share the whole sop for everyone because i think that's just going to become um a staple in every business that ai videographer but yeah we're working for it now look he's uh three four weeks in um and we're working up towards hitting a um a certain kind of kpi target in terms of output um we're kpi-ing against output but also against performance So we're obviously tagging it in client accounts, like AI creative or AI video, and seeing like how we can increase percentage of spend through those ads. In terms of like what we've done initially is we've created a framework database that the, our strategists have to use to deliver briefs to that AI videographer.
41:05Loukas:And the reason we did that is because, you know, with AI, like you can literally do anything. So unless you set constraints, it's going to be very hard to optimize a process. So we've given like kind of, you know, a few really clear standout examples of success we've seen through creative and, you know, created some limitations in terms of what we can do initially, which means that we can get up to a point of volume before then we kind of look to expand on that. essentially so yeah so far it's it's still very early in the process but we're we're just we kind of would rolling out first for one account and now from next week we're doing it for every other brand in the business and more to come in terms of what that kind of end-to-end process looks like we've got loads i want to do a whole episode going into it to be honest and show some of the top spenders that we're seeing coming through that process yeah but we've got some like really big ones starting to happen i think the one takeaway that i think is maybe more actionable for people right now that are looking to do more ai ads like the claymations the pixars that kind of stuff is what i have noticed is you know the difference in script length and volume of scenes can add days onto onto an out out into onto like outputting on a creative so for example we had one claymation that was about 90 seconds in length in terms of script and had 20 different scenes we've built a skill so we can reverse engineer from a brief how long it's going to take um based on you know what is required that was going to take three days to complete so i would set your strategist like some constraints initially in terms of length of scripts and also volume of scenes.
42:57Loukas:I think the sweet spot really is for us like 40 to 45 seconds and under kind of 10 to 15 scenes if you can but yeah, more to come on that.
43:09Olly:Yeah, seeing more and more of these AI executions. Some funny cartoon ones floating around as well so we can share more as that comes to light.
43:19Loukas:Yeah.
43:20Olly:I think it's going to save a meaningful amount of cost though from it. know from or increased capacity i think there's yeah there's a few and i what i will do in this
43:30Loukas:the next session when we go into this in more depth is i'll like share what kpis you know we're because again it's like what like anything like this anything ai related you need to know exactly what you're measuring success against and like what the ideal outcome is otherwise like you know there's no way of knowing whether it's working or not um so like i mean the because essentially this is like a type of high production uh that ai videographer is targeted to do 60 concepts a month that's unique concepts um if we translated those 60 concepts into high production so 60 high production concepts let me just do a bit of quick math um that would be the equivalent of 15 shoot days yeah mental so that's three weeks of back-to-back shoot days that one person can replaced that's three people that make up a production unit so again it's not like replacing production but it's again another lane in which will drive more creative diversity incremental reach but as a result actually you know reduce the strain that you're putting on a department yeah huge what about statics yeah so this is one that's dragged a little bit just because of other priorities that we've had over the last month but it's it's pretty much there and the outputs are very strong.
44:51Loukas:Again, this is something that we're going to want to build into our control room, which is obviously going to house our growth compass, which they would have seen that episode down, right? If you've not, we'll link it below. And also, obviously, the creative strategy operating system. Creative forecasting. Creative forecasting, exactly. Everything will be housed in the control room. But right now, whilst it's in MVP, it's within a separate web app. But essentially, we've created a process for mass generating static ads at scale. So we've got between 40 and 60 inbuilt static framework prompts that are prompts that will drive towards a great output.
45:41Loukas:So essentially, it's got all of the brand context and DNA built on the back end. So all you've got to do is pick the brand because it's already got, has all of the personas tied in already, has all of their brand assets, their brand guidelines, fonts, et cetera. So all you've got to do is pick the brand, pick the persona. So like P1.2, persona one, macro persona two within persona one. If there's any direction, you can give it. I don't know if I said pick the framework, but you pick the static framework. and then the amount of variations that you want. Click generate and it will just run this back-end process where it will generate those at scale.
46:25Loukas:And the outputs are getting really, really good. So I think it's super, super exciting.
46:32Olly:Yeah, really excited to see where that goes and just how much more volume it enables us to generate for clients and pace as well.
46:40Loukas:Exactly. I think where I see the use case initially is how we can be more reactive in the moment when we see something in an ad account and can just act on it straight away. So for example, a growth strategist or creative strategist who previously wouldn't have had the skills to go and execute on a design. If they see something that they think, oh, you know, this is something based on a certain hypothesis or insight that we could just test right now. It's very low barrier to entry for them to actually go and produce that asset.
47:13Olly:It's like convergence of role again. It's like enabling the team to drive more value. Exactly. And through these tools that we're building. Yeah. We've also done a lot of work on the back office of the business. I think we're talking a lot about like clients and e-com, but we've obviously built a completely, our own like business intelligence platform for the agency itself. Yeah. And around clients, utilization rates, profitability. I think that's been super valuable just connecting data sources and also getting a holistic view of our whole client base in one dashboard which is something that's difficult to generate previously and how they're comparing to expectation and their performance goals and allowing us to better allocate time really excited for the benchmarking that we're going to be able to start doing through this data and through this context that we've now got warehoused across like our full database but also vertical at different aovs at different sizes how how we how do we want to slice and dice the cohorts within our broader brand database yeah to extract useful insights for clients 100 i think a big one that
48:27Loukas:i'm looking at in q2 because i think we've identified this that like sometimes if people don't have the capacity they won't be able to proactively do analysis on certain elements to to drive a better result so for example you know i think you mentioned a good example earlier but there's let's say that you know there's there's we're under indexing on a certain persona within an ad account that's something that actually there's quite black and white triggers that will will tell you that um just by running one of these reports so like what one thing that i'm quite bullish on is like how you can create a system that identifies certain triggers that will lead to a certain report being executed on so for example if we are seeing a drop in incremental reach then there's a certain report that auto runs that will give the pod a breakdown of like what that looks like across the ad account, but also some potential insights that they can translate into strategic opportunities.
49:37Loukas:Another example is potentially hit rate. So let's say like, you know, there is a trigger that if hit rate declines month on month by more than 20%, let's say, then it runs a report to identify or hypothesize why that is and some potential solutions to fix that problem. because I think there's now, now that we've got like meta API access, you can run at any report essentially. Like you can do any type of analysis and not that that's ever going to necessarily be replaced by AI, but it's a good point of having the training in place so there's certain triggers that will, I guess, bring it to the surface.
50:19Olly:Yeah, it's surfacing the data and then enabling better strategic decision-making.
50:24Loukas:Yeah.
50:24Olly:So I think it's like prioritization and where to focus, which I think can be a challenge. I also think it can be a challenge when you're too close to sort of a problem and like having like a zoomed out view of an account or a business, which and being less in the weeds can help you sort of spot those opportunities. But I think AI can help us surface those.
50:49Loukas:Yeah, for sure. Well, I think we've gone through plenty there.
50:54Olly:yeah so um we've got a couple of masterclasses i'm going to be doing the um the creative forecasting one that i mentioned nice a couple of episodes ago we're just fine tuning like the model and the platform in which we're building that out in um i think most brands aren't spending enough time or thinking about creative in that way yeah i think if you as well like if there
51:17Loukas:are any specific masterminds that you want us to do comment down below and let us know because we read every comment and yeah we want to make sure that we are serving the DTC Diaries community but yeah make sure you subscribe please give it a like and we will see you on the next episode cheers
From the publisher
This podcast is proudly sponsored by SARAL.
Find influencers, reach out, manage every relationship, all in one place, without any chaos. 200+ brands including Grüns, Solawave, Spacegoods are running their influencer programs on SARAL.
If that sounds like what you've been looking for, go to https://tinyurl.com/3t4d4myx and book a demo with their team. Tell them we sent you to claim your $500 discount and whiteglove onboarding.
---
In this episode, Olly and Loukas go deep on the AI systems they've spent the last few months building inside Soar Group - the personal operating systems, the creative production infrastructure, and the internal tooling that's changing how the business runs.
This isn't a conversation about AI tools. It's about the architecture underneath them.They cover:
How Olly built a personal operating system based on Andrej Karpathy's LLM wiki pattern — using Obsidian connected to Claude Code. Call transcripts categorised automatically every day, Slack piped in, reasoning traces stored against every decision made in the business.
How Loukas built his own version in VS Code connected to Claude Code with in-depth context on every brand, every direct report, and every specialism, plus a reinforcement learning loop built into the content pipeline.
The Club Neuro AI operating system - a full web app built as a test case before rolling into the Control Room. Best performing frameworks dashboard, hooks database, one-click reporting, weekly Reddit scrape, and a UI designed to capture operator intuition at the moment it happens.
The four learning loops: performance into hypothesis, hypothesis into correction, correction into rule, decision into outcome.
How they hired an AI videographer targeting 60 unique concepts a month — the equivalent of 15 shoot days replaced by one person.
A mass static generation system with 40-60 framework prompts, brand DNA and personas pre-loaded, ready to generate at scale.
The principle that ties it all together: context compounds. Models don't.
---
00:00 Intro
02:05 Our sponsor: SARAL
03:00 Our personal operating systems we've built
21:00 AI as a feedback loop
24:14 Our sponsor: SARAL
25:10 Our foundational context and data engine
30:20 Club Neuro: creative intelligence operating system
35:00 Best in class forecasting process
39:50 Our internal creative processes
47:28 Business intelligence platform
Follow Olly: https://linktr.ee/ollyhudson
Follow Joe: https://linktr.ee/joe_marston
Follow Loukas: https://shorturl.at/TRI8B / https://x.com/LoukasHambi
Follow Liam: https://www.linkedin.com/in/liam-dean-a11a43152/
Follow D2C Diaries: https://www.tiktok.com/@d2c.diaries / https://www.instagram.com/d2c_diaries/
