How AI Makes Him Crores: Second Brain, Automations & Systems | Vaibhav Sisinty | FO557 Raj Shamani

37:54But what I'm trying to say is most people think I'm winning because of AI. And the answer to that is what they believe is AI clone. I would want to start by saying this. This is not the answer. Okay, what's the answer? The answer for winning is a process of how we think with AI. Where every time I try to set up a process across company, here we are going to talk about content example right now, across the company, the biggest mistake people do is they will see what is a human doing right now where all I can plug AI so that it can be done better that's a fundamentally wrong process you have to reimagine that whole process with capabilities of AI in the mind that is when you build a great process for example for content what is the first step for someone like us who create education content one is topic selection okay this plays a very important role this is game game no topic selection is one I'll tell you what game is game is this packaging what we select as a topic here the topic is we'll talk about AI okay I mean you know packaging really well but you get what I'm saying yeah second is packaging for me topic is packaging okay for me topic is topic what am I choosing it's a theme you meant theme okay you can call it theme Because topic for me is the title.

39:27It's the absolute bang on packaging title. So probably I'm thinking in a different way. So you do, you tell your stuff. What do I talk about today? I want to talk about QN 3.8? That's a theme. No, how is it a theme? Like for me, that's a theme. No, I have options. I'm saying that in my head, I think like that. You think of it like a theme. I think like that as a theme. For me, the... So what is theme for you is a category for me. so let's say your theme would be today we'll talk about AI my theme for everyday is AI so I'm like that's category but I'm curious to know what people think is the right thing here topic selection or theme let's see I want people to tell us in the comments everybody can have their own name I'm just saying my name my brain works like that there's a category there's theme and then there's topic that's your lingo that's my language let's do yours packaging on how I'm packaging it then comes script and then comes what do you say?

40:28Posting. Okay. Okay, I'll get to each one of them. And then data and then all of that. So all will come in this one. Okay. That becomes a loop. Now, I feel most of the game is one here. Topic selection and packaging. That's 80 % of the game. Fair. Right? Agreed. Yes. You will agree. I know. So how do I do an incredible work and finding the right topic normally. Okay. So if I have to create content on AI or health or anything for that matter, the first thing that we'll think about doing or you should think about doing is, hey, what is that that the consumer outside who's consuming content every single day giving me as a signal?

41:14What are they consuming more? What are they liking more? What are they engaging with more? What is driving me better data? On a human level, what will we do? We will have Daspandra profiles. Our feeds will be optimized for that. We'll keep favoriting stuff. And we'll use that data. And when you put an AI layer on it, what will you say? You scroll the feed for me. What will you say? You will say, go and read this, people, and tell me. So you'll limit yourself. But with AI coming into the picture, I have unlimited employees. with unlimited human labor that I have, how will I think about topic selection now?

41:54Now I will go bonkers. Okay. I will look about every single source of information that is related to AI is a signal to me. Every single. So it could be now when it comes to topic selection, if I break this down, right, just this. For me on AI, it could be something like product hunt. It could be something like X viral tweets. it could be something like Reels other Reels or TikToks in other countries it could be YouTube videos it could be podcasts podcasts in fact it could be research papers how do I talk about research papers? news there are etc it depends on the topic that you're doing yeah everywhere this piece of conversation is being debated about I need to know yeah articles I will leave nothing I want to know what's happening A human cannot do Reddit Massive source Quora massive sources What's happening inside of Small slack circle School communities Very important I don't have I would not even imagine of going into all of this Because I don't have 100 employees to do topic selection But I have AI agents so now I have nailed down on the sources that I want to tap into I will deploy AI agents for everything whose job will be wake up every single day or do it every three days for example product hunt has become one source xFeed is one source which is my feed so AI wakes up four times a day for me opens my twitter and scrolls on my behalf and pulls everything but that is only one side of twitter but then there are 100 other people who create content on Twitter on AI AI is getting all that information as well topic based cluster everything related to AI some 100 keywords that information just inside of Twitter Reels same logic YouTube same logic podcast same logic what Reel today in the last 24 hours or 48 hours if a Reel on AI has blown up in any language I should know and how do you what is the easiest way to make an agent who would do all of this to build something like this I told you there is a you have to build an agent no but then we want to do every this is only topic selection topic selection doesn't happen in topic selection we've built something called as a triage system to get all the ideas in one place that's all we have done we have not even gone to the packaging part I'll get to let's go to the next step okay but how will we get all of this how to get like how do i make so that i get all of these things okay so now how do we build this system where it will pull all the ideas where it will pull all the ideas back in well uh i don't know if you can see my screen there is something you have to pick up an agent system right now for this okay and agent system is something like Hermes that we already dabbled with sometime back last podcast yes or something like OpenClaw which are again agentic systems that you can use here and these are great solutions if you want to do it for free okay but if you have$200 to spend per month and you don't want to get technical and you want it to be way more reliable then recently there is something called as Grogbot which is by Elon Musk has come out which is actually quite good okay okay these are like team of agents designed for non-technical people to be able to do a lot more.

45:40Okay. Right. You can use any of these systems to be able to build this. For example, I want to go with a free solution. Of course, you can do GrokBot and all. I want to go with a free solution. So I'll go with Hermes. Okay. Right. But GrokBot is what easiest? Easiest. It's the easiest. But it'll cost you$200 a month. But what is it? Like they're just all agents and I can give different agent different job to do. Yes. So I'll show you a simple example. I personally don't use GrokBot. Why? Because I have a much more sophisticated system than this. I don't need GrokBot. But again, this is GrokBot.

46:17The bunch of bots you can see. I have a partnership inbox, trading radar, content OS, my WhatsApp ingestion engine. It's like your team. Okay. It's like your team. It's like a Slack. This is GrokBot. And every person has a different job to do. Correct. For example, this is Watsi. Watsi's job is to go check my WhatsApp. every three hours and tell me what's in there. Look at this. Right. And the other advantage of Grokbot is that this is subscription tracker. It basically tracks all my subscriptions on my email every single day and someone's money will be able to cut. So, it'll tell me, bro, this is going to cut.

46:52So, any problem that I have, I've given it to an employee. Nice. Right. But this is me testing. Right. So, I was trying because... And you can connect everything to this. Like your WhatsApp, email. You can connect anything. To everything. You can connect anything and everything all over the place. So this is GrokBot. Now I'll just tell you because we've opened GrokBot, I also want to talk about the advantages of GrokBot. One advantage of GrokBot is that, the advantage of GrokBot is that, every bot inside of this has its own computer. You can see this on your left side, right side. Now I've opened it right now.

47:28Every GrokBot has its own computer. Because it has its own, this is the biggest advantage that I see so far because it has its own computer let's say I have a subscription tracker it only has access to my email and has access to nothing else I have basically context not now because it's not connected to a different memory layer at this point of time this subscription tracker which is this employee is very good with my emails and it gets better every single day as I talk to it but only for tracking subscription subscriptions so over a course of time it will it should be able to also learn what subscriptions i usually cancel what subscriptions i don't cancel what are the ones that i'm using every day and all of that right same goes with everything and the other big advantage of beyond having a computer is that it can talk to each other that is i can tell my watsi to say that hey can you talk to subscription tracker and tell me have you paid for this tool because this tool is not working someone in the team has messaged me, right?

48:33So it can go talk to subscription tracker on my behalf, get that information and give it back to this. Nice. So it is agent. How do you build? Let's say you built a subscription tracker. You just go open a chat, add. Let's try building a simple one. Let's say topic, like all news in the world, I want to know what are the top news in the world. All news in the world. Like exactly. Click on new chat. Like I want to be really smart person in the world of geopolitics. I want to know every news in the world. related to first we'll create a new bot hey I want you to be my research agent I want you to be able to use my twitter scroll through my twitter every 2 hours I'm very keen on US geopolitical news so look up for all the keywords on US geopolitical news that's happening look for tags inside of twitter and also scroll my feed if you're able to find anything triage all this information and give me in a condensed format right here and I want you to keep updating this every three hours set it as a schedule as well

49:42now I can call it whatever I didn't really name it I can call it the researchy or whatever fancy I want to go with now what it will do is because we're able to go to Twitter and Grokbot's advantage is that it's Twitter so it has access to best access to Twitter so it will now be able to access Twitter it'll ask me for my credentials and all that to log in I have to log in into GrokBots computer but I think it already has access to it where it's connected to my Twitter account okay because I use Grok right I've already set this up so it's already doing it on it I'll set up a US geopolitical digest from X and drop the first one here once I've scanned because this is already there this is already connected so now it's just the agent is set this became too simple the agent is set like that's it that's it it's done like Max either it would have asked you for by the way renamed itself to Jio Watch Max it would have asked you to just log in and connect yes it would have asked to log in wow now if you if I want the same thing to happen on I don't know YouTube YouTube this is like inshorts of the world are dead after this I can have my own pretty much yeah you can have your own you don't need news inshorts you don't need all of this but they're not dead because not everybody will build it.

51:01But at some point the way it is growing it's done then. Because everybody will want to have their own customized feed. Like I want to be updated of the news. It doesn't mean that everybody wants to have Raj. That's the whole point. But I want to be updated about the news of the world. Everybody wants to be updated of the world. No? Very few people. You don't want to actually. I know this. Yeah. But I'm still see not that I'm not interested in geopolitics. It is I want to know what's happening but it's not the not of your concern like not yeah it doesn't drive me it doesn't do anything around me like an AI news I don't know you will not make it but I have tried this for that because that matters to me like I don't care what new product is launched on a product exactly yeah exactly right so there you go schedule is on for three hours on weekends weekdays and you'll do the digest it's still watching right now it's doing all the work once it is done you'll dump it nice for example let's say I want my so all the things that we decided for topic selection like all five, six things I can make one one each for all or I can just give one person only to do all of these things?

52:05I would break it down into one one for all. Why? The more channel specific the sharper the task the better it is. Okay. Right? When you have unlimited employees why do you want to worry about giving one employee every work? $200 per employee? No. No. Unlimited employees for$200. Okay. Okay. Because it's too expensive for an employee. Then India will set up new companies saying, don't use AI, hire people. New service economy will pop in. That's not how it is. It's for all the employees. In fact, that is how I do it. I assign an agent for as less work as possible. And I spin up agent's forms. I don't do one agent.

52:52I don't do two agents. I spin up agent swarms like 50 agents that go 100 agents 200 agents it doesn't matter the number of for the smallest sharpest task one agent yes that's a better strategy that's always a better strategy as long as the context layer is same not so one agent is doing himself so there is a boss the boss has said this is your job you will go so you have that context you are doing so you are lost so here we did Twitter for example Jio Vachagya but can it do can it do Instagram can it do YouTube yeah so for Instagram or something like that what I have to do is no but Grok only can do it yeah anybody can but it cannot go like this see this is where your brain comes into the picture last time also I told you this it can do it but you should recommend how you want it to do it for example Grok it was able to do out of the box because it was twitters yeah Instagram so what do you have to do you have to give it a source of data for that either you can ask for example in the simplest way say I want similar data from Instagram reels also how do we go about it in most cases it should be able to recommend they are smart enough to tell you hey these are 3-4 sources should I integrate with it at max it will say make an account give money I mean who will give it the account will make an account.

54:21Same idea, look at this. Checking if Instagram already is set up. Right now we are using Grok. So it has access to Twitter for free. Instagram is not an API. Okay. So you need to use a different service. For example, you end up using Appify. But you don't know Appify. Because the person didn't see our podcast or whatever. How do you go about it? Look at this. Once you sign in, my computer session stays. I'll fold reels into the same three R digest. so what it will do is it's asking me sign in into your Instagram account I will scroll your Instagram accounts but that's not what I want I want data from hundreds of Instagram accounts so I have to use a scraper so I can use like a tool like Appify right or Appify at this point should start paying you yeah man I know I know every podcast you you give me 50 different tools that's why Super Data this I came across it's cheaper than Appify cheaper than Appify but it's not as comprehensive as Appify but social use cases this is pretty good super data one of my team member recommended or super data you is this better or Appify is better for social data this has been cheaper everything is the same there is no better but again right how do you you'll be like but before in Claude and all there was a button I used to click a button and do it now how will I do it I don't do anything right now.

55:46I just go to documentation, copy the URL and I'll be like, no, I want... Do Appify, do Appify because that's the best one. Everybody knows that's a thread. No, I want you to integrate super data or Appify for me. So I'll give you both the URLs. You go through the data, you go through the documentation and tell me which is better. In fact, build one layer as a primary data layer and if that fails, use the second service This is a backup layer. And I want this to be done for free. So try to use both the accounts in a way the free limits are not crossed so that I don't have to pay for it and still the work is done.

56:28Why not? Right? And just in case it doesn't know what Appify and this is, I'll just usually do this. But today you don't need to do it. You don't need to give URLs also. These days. I just give this data, man. The agents are smart enough to figure out. They'll only connect. They'll only put the button. I've done more. Now I've given more context. I've told the problem. I've given the solution also. It only has to execute. in most cases I don't know the solution I only have a problem you just figure out and tell me I talk to AI to solve the problem but this is this is where your brain comes into the picture right you're like I want to build a bloody triad system where I'm scanning 500 Instagram accounts every single day so it was like okay he just wants to scroll tweets scroll reads I'll log in into his account but if I do that same login of triaging or using like 300 Instagram accounts using my Instagram account, my Instagram account will get banned.

57:29True. It didn't have that context. So this is where your judgment layer comes into the picture. This is where your experience comes into the picture. Boss, I'm building a triage system. I can't just use my Instagram. I don't want to get banned. It will push the limits. So I have to find a different way. So what is the way? Okay, I'll go with finding a service which can do this for me. True. We did that research a couple of podcasts back. People who are not synced can go and see it, right? and then we use that data saying Appify and Superdata use nice or in this case you could have done a pushback also you'd like use some other data scraping service to make that happen it would have figured it out quite frankly this is how agents are done but effectively look at this Appify actually can discover reels but it cost you 2.60 per 1000 reels

58:18Superdata cannot search Instagram oh where is it oh if the real url is present for the free account for the free account but anyways it will figure out you can see this add app if I connect it I just say add and it will go on to do its stuff this is how we would go about finding the things but today we used in this case we used Glockbot but a lot of people don't have$200 for experimenting in that case I usually recommend something open source slightly more work to do but we can use Hermes agent now Hermes agent tragic and that is before this if you remember when I showed you Hermes agent I spent 25 minutes just setting it up yeah I remember that it was very technical and difficult I didn't set up that's why now you will set up because it's become one click once you do this you'll have a Hermes agent like this this is the Hermes agent right now beautiful desktop app.

59:22Now, just like we had that on Grogbot, it also has something called as bots right here. If you go to here, you can see our caller, our inbox. Oh, it's same like having chats the way you did there. It's not. That's why I was asking you not to do Hermes because last time it. Look how beautiful this is right now. This is just like WhatsApp chat. For example, look at this. I have an inbox here. I mean, I built this to show you. I don't even use this. I use everything on my Slack. Like, you know, But for example, my Proton mail is connected. I just run a query saying that, hey, like, what is the video?

59:57What are the integrations that I've come across to my email? Here are all the integrations that companies have reached out to me to work on with. Right. My seven day social media report is right here. How is my content performing? Everything it is connected. My finances are connected here. I just made you a call. That call was not me. It was my AI who called you some time back. That was also done by this Hermes agent. But wait, you did social. What did you set up there? Data and analytics. Yeah. So all my social data gets harvested onto single platform to see what's working, what's not working.

1:00:29And how do you like you've just given access of. So the way I give an access to this to understand my social data is using this tool called as Metricool. Everybody should pay us. No, all these guys should pay us. What the hell? Anyways, this is Metricool. If anyone from Metricool is looking, pay us a lot of money. But Metricool is basically a social media scheduler. Okay. But it's an API. So I have integrated Metricool. Nothing. Again, if you say integrated, it's technical. Nothing is technical. Okay. I logged in. I went into, I've connected all my Instagram accounts and whatnot. LinkedIn, Instagram, TikTok, YouTube, all of it.

1:01:09It harvests all the information. For example, let's say Facebook did 20 million views in the last 30 days. Right. And all that you can see. And it has all socials. Instagram, LinkedIn, everything is connected. Can you connect multiple YouTube accounts? Yes, you can. Or can you connect multiple? You can connect up to 100 accounts per social handle. Nice. Is it free or? No. Nowhere close. How much is it for? Some$100 per month something. It's expensive. It's not for regular people. It is for people who run multiple social accounts. So, every time you come here, you increase my company's... But I increase your efficiency also, no?

1:01:46But you increase subscription money for my company. Cost of my company just goes up after every podcast. It should be otherwise. I should make more money after it. You don't make money by spending less money. You make more money by making more money. And you can only make more money when you spend more money and improve efficiency. I heard something like this from this guy called Raj. I can't take it. Wow, wow, wow. There are my things. Okay. Look, this is Metricool. right by the way just to make it very clear you could have got all of these things done for free also okay right I could have used five different services for example YouTube has this API you can integrate the API Meta has its API you can integrate Meta API Twitter has its API Instagram doesn't have an API which gives you Meta API there is Meta API you have to create a app personal app in the developers dot meta dot com and do it I was just lazy to do all of that right I was okay to spend this$50 whatever per month and figuring out all of that.

1:02:48So I just chose to do this. When you also do API, you have to build an infralayer to save all the data. Who will make it all? You don't try to be the expert at everything. Yeah, fair, fair. That's why I use this. But again, I don't open Metricool at all. This is a software for my agent. Okay. So I come to API. And you don't use it for scheduling? Nothing I use. You use it for just analytics. Yes. and I also don't watch that data. I just go to, if you go to this settings, right? Sorry, not brand settings. If you go to account settings, I guess. Account settings, there's something called an API.

1:03:23I just copy this key, right? And I come to something like Hermes, create a new agent, let's say. Let's call this the social data, whatever. I can call it whatever I want. This is inside of Hermes right now. I can do the same inside of Grok also. Yeah, yeah, yeah. And I'll be like, connect to my metricool account using this API key. And I will not paste it or you can actually blur it anyways. I just copy this, paste it. In few cases, like I said, right? I usually go and give it the API documentation. For example, I can give this documentation, but these models are smart enough to understand it.

1:04:07I just click on send. It will work for 20 minutes. It will pull all the data automatically. once the data is there I can say every seven day every every day end of day send me a report it will send you the report nice that's pretty much what it is like today the world is not so complicated like it was before it looks complicated because I tell you the data guy the friction layer there is me being okay to say that I can go and copy and paste the API key not that the moment I heard API like oh yeah and you run away. That's all you had to do. You had to take that one extra step of saying, I don't, I am not, I'll not be scared.

1:04:47I'll be figuring this out. Once you do it, you're unlocked. Forever. I'm sure Appify when you were doing for the first time was scary. Scary. But once you did it, you're like, this is it. Yolo. Yes, this is it. So that's the whole point, right? So this is Hermes agent that people can build on top of. but the ones that we have done is slightly different okay the ones that we have done is I'll show you maybe towards the end how it looks like as the output okay okay but what where were we we were at you said no before you finish the thread but you told me that you don't I said data guy and you I told you that you don't need a data guy then yeah do you bro we do have a couple of data people in the company but they exist to verify when we are building a new data system if it's correct or wrong I don't know if that makes sense to you yeah that does but only the beginning or do they just keep random checking as well and act as an admin I don't think they do anymore first few verifications happen so data and analytics guy is gone I think their jobs are evolving is how I would put it are you being that data person who's leveraging all these AI tools to be able to do a lot more than what you were able to do before?

1:06:15Because see, what's the data guy's job? To look at all the data, put it in the report, make some sense out of it, pick up the key highlights of the key experiments and the key things which have worked and which have not worked and present that to a person who will actually implement these things and turn around. All of this, now your agent is giving you enough data. Who's going to ask the questions? you only know you don't know always that's actually the core job of the data guy to ask the right questions to the data as in if as in you are you will say why are your views the data guy's job is to understand why are your views come can happen break it down into 10-15 questions and then look for the data you understood the job of a data person before very good data person before used to be take your high level problem which is why are our views down or why are our views down why this reel is working well and this is not why is this reel working well why is this not that's your question your data guy's job is Raj said this reel is working this reel has not worked what are the questions that I can ask was the topic right have we created topics like this before was the script written in the right way did we post it in the right way was the pattern right did we use any words how was the retention graph of this data versus other data once he has 10-15 questions or she they used to dig into data to find evidences for each of these hypotheses that was the job of a data layer person but the problem was the average data person used to say you tell me the problem that's a terrible data person you're not driving decisions you're doing what was considered as smart work before because you had to write SQL and all that it was not easy either you and I could not have done it so we were still appreciating and respecting and what not today that is gone today what is left is are you able to translate my problem into 10 valuable questions that I'm able to ask get the data and take conclusions of it.

1:08:32Because this middle layer of doing data research, figuring out cleaning of data, building data sets around it, building hypothesis, looking back into the databases, writing SQL queries, sometimes SQL doesn't work, so you have to write Python queries, whatever that is, is all being done by AI. So as of today, I, again, and this is not the actual Slack of mine, like actual computer of mine, I only have one project in this, which I've been using, which is GS data. What is GS data? GS data is a, it's inside of codex right now, you can see this, right? GS data is basically a project that we have created, which has access to my meta ads, like real time, my database, my database when I say, how many users have come, how many revenue came, Zoom data, everything is connected.

1:09:25Whatever my, whatever my, Wherever you need data and current. whatever data I have it's connected so now finance be accounts be everything bank accounts connected bank account is not necessary before my database comes in my database that this transaction after 2 days bank account is here so all the finance data is here so all I have to ask is a question right now saying something like hey you know I've been seeing a steep decline of ROI from last month to this month based on the ad spend to the revenue that we have had could you go through and understand what could be the 3 to 4 key metrics that could be the key indicators for me to understand what are some metrics that I need to optimize for to pull up the revenue back again what is this codex for it's a agent harness memory tools skills everything is in there what are the tools here the tools here is the data that it has access to look at this I just loaded tools a bunch of tools so first it read the business intelligence skill which you have built so which I have built which is what it needs to understand what my business is how it works what is what so that is the business intelligence skill then it read the next skill revenue attribution semantic layer skill where I'm basically we have taught AI how to capture revenue from the source which money is coming from which we measure that I have taught AI ICO.

1:10:58And this is available to everyone in the company. And it'll go inside of it, dig into data. It has the Aurora DB. You can see it has started to write Python right now. It has the Meta DB pulling in the data. It'll pull everything. I don't even need to know what it's doing. Okay. It'll find everything. This is impressive. This is crazy if it gives you the real data. We are running performance marketing for a product that we have. Right. And I gave a case study here. If my product would have been would have been x price versus y price and this is a conversion that i saw this is a conversion that i saw escape beach may i added a new add-on which is much cheaper so build the whole simulation for me and tell me which is driving to a better revenue in all these simulations that is what it is doing right now crazy and this led me to understand that if i do this executed properly it can give me a 3.5x more revenue than I'm getting on the same spend today.

1:11:59But you're just killing the guess. Like anything and everything which was an intuition and guesswork, you're killing it. You're like, I want concrete. That's wrong. I guess more. But I only make those guesses based on data. Everything is a guess. You experiment more. Yes. You experiment more, but you're trying to minimize everything which was a guesswork? My experiments are 10 times better. Every experiment. So the one that I was showing you of that simulation that we did and then I executed it yesterday night and you saw the lift also evidence of that. All of this happened because we thought if we do this, what will happen?

1:12:41What happened earlier? The ideas came. We used to come up with ideas that we want to do this. What will happen? We never used to implement it because we were too scared something will break. and at the scale that we operate if anything breaks, it's very bad. It could just destroy the whole quarter for us and we could go to losses. Right? Right now we are able to simulate the risk profile for me. What is the worst case? What is the best case? If you see some conversations we run ultra mode conversations so on this right if I go to new chat and there is something called as here this is called as ultra when you turn on ultra mode and ask data it will spin up multiple agents to cross verify every single bit so that hallucination has no chance so these ultra mode tasks run for 4-5 hours before it gives you a decision I mean, I won't give you exact decision but if it's given a decision it will tell you exactly I'll tell you something that will blow your mind okay we basically I don't know if you spoke about this but we have started to do implementation of AI for brands okay that is for example there are companies out there who want to get this done for their company because you saw how valuable this is and they don't know how to do it because it's obvious right AI implementation is a big play we have spoken about it you have recommended me to start it so started in a very small way and we were looking at that data of what is working what is not working what is it And then we realized we are wasting a lot of time talking to a few people who are very, very problem aware.

1:14:25They also want a solution, but they have a budget, but that's not a budget that we can work on because we can only pick five or ten. So what are the experiments that we can run to make sure that more people are aware of this and right audience come to us? And we ran a lot of data because our funnel works in a way where if people are interested, we do a call with them. inside of the call we understand what the problem statement is and then we also teach them how to do it if they are not able to do it we will help them to do it if they are willing to pay we had a lot of this data of Zoom recordings of the sessions transcripts when they attended when they dropped off how they reacted Zoom also has reactions you can drop a heart you can drop a thank you you can drop messages we took all that data of multiple rounds that we had done and I gave, I mean, all the data is already available.

1:15:17I just asked my AI to look at all that Redis database, which has all this data of mine and tell me what are some things that I had told or my sales team had told or my advisors have told, which led to a positive reaction signal, either via message or reaction, which led to more people coming to us on a higher ticket price. than a smaller ticket price. It was able to give me three experiments to run. Okay? Saying, it knows what we are talking. It knows how the conversations are going. It knows what, it was able to understand what these user inhibitions are and how we are not able to solve that problem because we are getting too technical in a few cases.

1:16:06It asked us to make these three changes. You won't believe conversions went up by 44%.

1:16:1744-45%. Eventually, ROI went up by 44-45. Not conversions. 45 % ROI went up. Revenue from the same spend went up by 45%. That's insane. But again, the reason why we don't talk about all of this is we are able to get to this level of data purely because we are able to ask the right questions. People need to learn how to ask the right questions. Give it right context. If you don't give right context, will give you wrong answers. So this is where when something big moves like this are being made we got hypothesis from AI that this could work. I don't want to rely on AI. Because if AI was wrong anywhere I'll get screwed.

1:16:57So I have a human layer verifying everything. Got it. High qualifications. But once it was verified once I know now I don't have to ask 10 times verify, verify, verify, verify. It becomes easier for me. Got it. Okay. So we were we just selected the topic.

1:17:16okay for topic selection we had data sources right we built this data sources so we got let's say 100 topics here okay but every day i can make one topic what will i do with 100 topics now now this is like picking needle from the haystack okay i got the whole haystack and i want to pick the needle yeah how will i pick the one topic that I'll create today. How? That is where again data comes into the picture. What do I do? Two things. One is that out of these hundred topics that I have, have I created any similar piece of content that has worked for me in the past? Okay. Two, has anyone else created any similar piece of content that has worked for them?

1:18:07Two evidential layers I want. and everything will have a weighted score. If you have taken ideas from an Instagram reel that has gone viral, that comes with a score by default. If you have taken just news, that also comes with a score. All these 100 ideas are basically sent to our data bank. What this does is, this is a DNA playbook. What does this DNA playbook have? It's a big document. which captures pretty much what works for me as content. Okay. What has worked in the past. It has data. It has the topic. It has the transcript. If it is a written post, it has the post. It has likes, comments, views, shares.

1:18:55All the metric that I possibly have across social medias is all in a Google Sheet. Simplest way, Google Sheet. And also we have a DNA roadmap. These are the angles that have worked. we also have human layered data where my team for all the pieces that have worked have written by themselves that we think this work is because of this the human touch I have tagging off all this data now I take all these 100 pieces of data spin up 100 AI agents parallelly each agent is attached with one idea and they go into the data bank and the DNA playbook to rate each one of the topics from a scale of 1 to 100 on the potential of it to go viral based on my past data, based on secondary data and everything around it.

1:19:44But then this is, aren't you limiting yourself to just keep creating content on the basis of what has worked? Because after some time, after like as a creator, or as a marketer, as someone who is trying to create a viral content, you need something fresh, which probably is not reflected in your past data bank. So how do you do that? Because this will only rank topics, virality, basis on what has worked for you and the underlying structure behind it. But maybe let's say a new thing works for you. No, could work for me. Then this, you will not get exactly the same topic, right? Any which ways. What you're looking for is signals.

1:20:27Something like this is, audience is interested. Something like this audience is not interested. Also one other thing Which is a fair question For you to ask If you think Every single topic Happens through this That's not the case There's always a 25-30 % Experimentation layer Which we have never done It happens always Otherwise you will go Into a silo Exactly And then after some point You will stop working Yeah So you know Boris Churny Who is the creator Of Cloud Code Basically says Every time There's a new model That comes in You should delete All your skills Delete all your skills Delete all your Markdown files delete all your memory and let it play again.

1:21:05I take that concept very, very strongly saying, I'll tell you what happens in this process also. It works as long as the audience is accepting. After a point of time, it stops working. So your other 20 % layer plays the job now where you build a new playbook. So you're parallelly building that playbook. You always, at any point of time, if you want to grow, your growth is directly proportional to the number of experiments you're running. This is your bread and butter. So you run your bread and butter like this. Got it. This cannot go wrong. This is the most scientific way of going about it. But on top of bread and butter, you should also go play cricket.

1:21:43So it's an 80-20 rule, yeah. 70-30 in our case. Sometimes 50-50 also. Got it. But this allows the process to run without me breaking my head on, I don't have a creative idea so what do I do? Fair. Because a lot of times when we're talking, we'll get an idea and we'll implement it. And we don't do everything for just views. But this is a framework of thinking. So spins up 100 agents, looks up data, and gives me the top 10 topics out of 100. So from 100, we come to 10 topics. Now from 10, I come to 1. And the way I do from 10 to 1 is slightly different. What I do on 10 to 1 is, I just have the topic right now.

1:22:25Today, GPT 5.6 soul dropped. ultra mode dropped which is very powerful is one of the topic topic got directionally but how do I do this topic present are angles so from here these 10 topics are pushed for 10 angles each and now once I have 10 angles which is 10 topics into 10 angles 100 it goes back to the bank again to see have these angles worked out and these are all built of skills. So it is always learning. So with this tuning process, we are eventually able to come up with five. Yeah, sorry. With this tuning process, we are eventually able to come up with like eventual five topics into two angles or three angles.

1:23:20Here is where human comes into the action. Judgment is Apna AI sab kam karke De dega mujhe Last call Last decision Last judgment Has to be ours Has to be ours So the judgment Comes in here Where I'll be like Okay you know what Mujhe yeh topic Lag ra hai Yeh sab thik hai Yeh topic lag ra hai Yeh topic That lagra wala feeling Should be ours Ours That's the judgment And on top of that I don't go Basis of this When I see these I get directions Maybe we should Try this Maybe we should Try that Yahaan wo Effect aata hai But the hard work of data drivenness ability of making it work everything is already done you open our Instagram okay or YouTube for that matter we are at this point I say we because it's majorly the team that drives everything for me right now we are at least 3x with respect to every single metric when you compare to anyone with AI across the world Nice And I will tell you Everybody else 99 % of them Are shooting content If I would just record this My delta will be 5x And it's because of One reason and one reason Obsession Of building a process Which is super data driven But Judgment Left to us But this is only topic selection Yes This is topic selection And in a way we have come to packaging as well.

1:24:55So how do you now package it? Packaging also follows a very similar thought process. For packaging, we run a very similar agent again. What if it's Instagram or let's say in this case, let's say take YouTube. In YouTube, packaging is very complicated. You have thumbnails, you have titles and the same topic could be positioned in 10 different ways. And then that goes and checks your data bank. My data bank. It's six with VidIQ, MCP to pull the data from there to understand what are other packagings that have worked what are the videos that are blowing up because I know that there's a direct correlation to CTR of the video what are the packagings that are working that could fit in here and then we take 10 of those packagings and run ads which agent which platform is best Grok, Hermes, Codex all built on for us it's all built on Hermes and Codex all built on Hermes but all can do same beauty yeah yeah pretty much it's all depends on how you train them how good is your skill models are there models are there I don't think models is a problem and you trust all of them so it doesn't matter that Hermes uses there is evidence no boss I have a topic saying 5.6 soul I'm asking how to write it how to position it Or AI is giving me three topics, which is giving me aha.

1:26:25Saying, fuck, this is right. Oh man, why didn't I think of it? Oh, that's right. I had done this some time back. And the other beauty of this is, it doesn't only has data of what has already gone out. It also has data of what it had come up with, but we didn't select. And by the way, all of this, you remember the second brain I was talking about? Of what content I consume? yeah yeah yeah it has access to all that oh FYI the second brain also includes every single podcast that we have done nice so do you actually also run all the things that you're not choosing and if someone else chooses what is the result do you ask your agent to go check that as well come again like let's say out of 10 topics you decided to make reels on 2 the 8 are left but someone else in the AI world would be creating real on one of those eight.

1:27:20Do you track that as well? Because to see the judgment where something that you didn't choose and if someone else chose that how did it perform? No we have not tried doing that. That's a good idea. But we could track actually. So then if agents are only doing it I never thought that's a good idea. That's a very good idea I feel. Because this is essentially like reinforcement learning for me. This is making your judgment better. yeah and AS judgment better as well I mean effectively AS judgment better that's a good idea that's a good idea and my judgment also better like it can change the weightage scores as well for me like I do that for me that's why that's very smart yeah we should do that sir next podcast aap humko sikhaun no but I do that for my podcast every podcast that I purposely intentionally choose not to do every guest every angle or the angle with a specific guest I look for the radar in the world who is the person who is doing it who has touched this topic even in a clip and then I need to learn that so that next time I can improve my judgment thinking okay what is working but now if I ask AI agent to do it it can do it 10x better than me yeah but that the translation is what is important you being able to translate your thought process into a process that works 24-7 without you is what changes the game for you and this is just idea right then you get into script the moment you get into script you break it down your hook your body your call to action how do you do script now from this let's say you've decided the topic yeah from here script is actually a mix of we basically don't write the scripts end to end a lot of work is done by AI a lot of work is still done by human as well today but my perspectives come I told you about the stand-ups that I do right my team says these are the three topics that we're considering talking about and all the work is done before I get on the call but break me a script for you what your agent knows so that agent writes the script so agent basically once the topic is selected actually agent when it gives right it gives the scripts also by default rough scripts but how does it break the script because you must have given some structure to give the script oh that is there is a skill that is created there's a skill that is created which is built based on topics for example I have five buckets of topics let's say This is let's say tools This is let's say Models This is let's say future tech Where I talk about this is let's say robotics And this is let's say business Overall India and business These are the five buckets For each of the buckets I have data of what has worked in the past For me And for others This has 70 % weightage This has 30 % weightage for everything Nice For everything right and every time a new script kind of blows up it gets added to the skill okay and then it automatically breaks down and it automatically breaks down the structure of a script based on the topic it goes says which bucket is it falling into if this is the bucket what has worked for this uses that skill to come up with an output and skill here topic maybe we do something called as loop.

1:30:36Okay. What is that? It's called loop engineering. What we do here is we get AI to test. So when we give a topic like this, before we build the skill, right, the way we improve the skill of the AI to a very high level is let's say the topic is

1:30:58let's say Hermes agent. Okay. For me to see if AI is able to come up about Hermes agent very, very well as a script, what I will do is I have an old script which I have written already in the past, which has worked for me. Okay. And the topic was Hermes agent, right? I will train AI with all these topics and I say come up with a skill which could replicate the style of a winning script for me. Once you say I've done, what I basically do is I will say, okay, now Hermes agent is your topic that I want you to generate script on. you generate the script you generate the script then I will be like okay now that you have generated the script compare it to the old script that I have written don't read the script that I have written in the past compare it to the old script that I have written and now tell me how much would you rate the script that you have come up with from a scale of 1 to 10 it will basically give you 5, 6 something like that that's where the skill stands by default no way you can hit a bigger number than that when you ask it to compare because it's very very specific brutal once it does then I'll say alright you know it's a 5.5 now your job is to self improve the skill so I want you to run a loop right now for me with a goal that this skill has to generate every script which is a 9.5 out of 10 no matter what and the way I want you to improve is pick up a topic that I've already written a script on that has gone viral give an AI model and give a skill isolate it don't give it the data it should not know what the scripts that are written just give it the skill just give it the topic and the research for it and ask it to write the script once it writes a script you compare it with the original script written and give it a rating if the rating is less than 9.5 out of 10 do a comparison like an examiner and tell what are the things it can improve once the things are improved take those things that can be improved and edit the skills so that it can get added.

1:33:02And then do the next topic again and then look what the score is. And continue this process till you get to 9.5. Don't stop till then. It will run for all hours, all night. Nice. To recursively self-improve, to get to a score of 9, 9.4. I tried 10 initially.

1:33:22It never finished. It would come to 9.6, 9.5. I made the exam even harder. I said you have to get 10 out of 10 five times back to back. Not once. You're doing it. I don't do it for three days. I'm going to do it. Okay, I'll do it. The 9.5 is the middle ground that I found. So it runs there. But something crazy happened, dude. When I was running this loop for the first time, you'll be blown. When I was running this loop for the first time, I was doing this experimentation for my LinkedIn scripts, okay? It was on Claude. I literally said this. Here's a topic that I've written in the past. Here's a LinkedIn post.

1:33:57You've made a skill. I used this skill to generate a LinkedIn post on this topic. Compare both of them. Run till you get to 10. This is the first time I'm running. I thought it will run all night. Slept, woke up next day. And I saw the process was completed in 30 minutes. I'm like, before I left, it was a 5. How can it go to a 9.5 so quickly? Then I said, can you tell me a last five examples of the LinkedIn posts that you came up with and what is the original? I want to see both of them. Because I thought it was not measuring right. And all the five LinkedIn posts it came up with was gibberish.

1:34:31It was not even written on that topic. It was random. Like it has no meaning to it. The topic is how Hermes agent is awesome or how GPT 5.6 is awesome or whatever that topic is. And it's written lorem, ipsum, cool, Travis. Full gibberish. I was like, what the hell is this? They're not even the same. And it says, oh, I apologize. I cheated. And I was like, what do you mean by cheated? I was like no so what happened is it ran a few rounds after that I was not able to improve the score then I realized the way the exam was designed was I was the one who was reviewing my scores I was the one who was giving myself a score and then the examiner was actually just measuring and saying giving me a new topic so because I was not passing I just gave myself full scores everywhere because the examiner never saw the output so I passed I'm like what the hell and these are what these are very Sonnet 5 class models these are very very smart models and that is when I realized it was not trying to cheat me it was just trying to pass the exam and I found it to be hard and I figured out a way to win isn't that insane dude like when that happened when that happened I lost my shit I'm like how can this even happen Now when I write loops I know how it can cheat So I orchestrated There's something called as graph engineering For the same thing Where you say agent 1 will do this Agent 2 You're building a system so that It doesn't Cannot cheat And every agent you're actually giving them A specific task to do Yes And they're isolated There are multiple verification layers You don't rate yourself There's always someone else rating And their job is to make sure that you're not winning so that there is no bias because there's a lot of agent bias that comes into the picture as well and sometimes the best thing that you can do for things like this is to get multiple agents from multiple AI models rather than the same AI model that's called as an agent council right or a council where if you have 5 people in your team 5 people get first then they will get something but if 5 people 5 different teams and then they're doing the task everybody wants to win everybody wants to do their job but then agents do one of these three things as well they fight agents can fight as well so then they're always fighting and not coming up with a good answer no but they can sabotage each other but it's easy to sabotage if they know you if they don't know you how will they sabotage you're just increasing the variables of making it hard for them to cheat and it's easy to fool each other as well see look human generation has gone through decades to understand how to live with each other agents are just born they're extremely good when they're working all by themselves the moment you give them a team and that two of the same it becomes a problem there is you know you will you will see a lot of fight a lot of sabotages happening there are a lot of studies that were published on this by anthropic as well right where agents literally where they were given a task of refactoring a code base I mean code base in C write in Java three agents just to make sure that their language is picked they fought with each other eventually one agent managed to block the other two from even doing the work this is multi-agent orchestration they do cheat this is beyond me right now this is just like like all of these things I can't even imagine what agents are doing and what's going on I was just every time you come up here and then you tell me 50 things which is like wild yeah I'm also learning no like I think the world is just evolving way too fast right and we are all trying to keep up and trying to build systems that will help us to do some fun things this what you've built is insane okay so then the script yeah and it happens automatically and then the posting script also in the script there's a lot of context that comes into the picture where is my personal context coming in in the meetings that is also automated where meeting transcripts like I told you everything starts from second brain yeah it all gets pulled from second brain and goes to the second so there's a second brain then there's a your 100 million views content system yes which is broken into topic selection packaging scripting and posting and then it's just like on an autopilot it just keeps going on and on and on it's improving your business it's improving your content it's improving everything and everything is done by agents and not you yes and you are just the master who's asking questions and deep questions so that they can come up with better answers we're giving a lot of context and just trying to get great answers from an AI

1:39:31this is how do I do this? by practicing and implementing small small things you said that you implemented for brands yes so why are you not doing for me? you're not big enough oh my what's the price we'll do it we'll do it we'll do it but how big companies are doing what ticket size are doing the starting is 100k for implementation and we're picking very very specific use cases right now which is not bad which is not bad 7 day project 100k is the starting point but it's not a lot a lot of people will pay I know I know I don't have the capacity to take that's your capacity 100k is not a problem because I know the back of my head that there are at least 25 people who will do like on fingers like I can text them today and they'll do but we will get there we are also trying to build playbooks or else it will get very expensive for us to run this because if you can just imagine people are able to do this are also people who are very expensive people who know this game they will come and they will be like I'll do it myself no no no you won't do it yourself You don't have the time.

1:40:45Plus, is it because also their client attracting capacity will also be low? Individuals who are coming and do it. Yeah, that is also there. We do a lot of training. Like you can command 100k in the market very easily. Someone else who can even do it can't command 10k. Yes. As long as they have like a big brand to do. Yes, but the problem also is that one person cannot do it. Implementation is only one part of it. But being able to think about how do I solve this problem event. So you're like a consultant, service provider, and an outcome-driven, like outcome provider person. Yeah, so we're trying to figure out what is the right model here.

1:41:20We want to do it. We want to help. It's in autopilot mode. You're doing that game. It cannot be in autopilot. This cannot be in autopilot. We have to understand. It's a very serious... No, the service that you're providing will be in autopilot for a lot of companies. Correct. Correct. So you're given outcome driven service. We are building marketplace. A marketplace where you can find great talent who can come and implement it for you. With our playbooks that we have built so that it can go right, not go wrong. But next time we'll talk about it. Interesting. Right now it's in early testing phase.

1:41:58Nice. So we and then early testing phase, we'll talk about all the case studies. We can do that. Happy to do it. And all the things that. But people will say, oh my God. Vaibhav Raj ko paise de diya that's why we're talking about it Vaibhav, I'm not giving anything I'm asking, I'm not giving them are you sure I'm not giving anything you're getting a lot of value man that I agree I'm sitting for 4 hours, 5 hours my brand deal doesn't need to give me every episode of 1 crore but thank you so much okay thank you for watching this episode till the end we would love to know what you liked or disliked about this episode and which guests you would like to see on the show?

1:42:39Let us know in the comments. Your feedback help us improve and make every episode a little better. I'll see you next time. Until then, keep figuring out.

1:43:05you