in this video you're going to learn Wait hold on I feel like I'm too loud because I guess this is a general problem in my life Okay you didn't come here for this In this video we're going to learn how to take any system that you have and turn it into an AI system And what we're going to do is as I'm going to develop a prompt live that will take all of my learnings everything I've ever learned everything I ever known and will basically allow you to take any system and help you think through how an AI system should be designed for that problem And so it'll be a mega prompt that will control the model It'll teach it how to talk to you It'll ask you the right questions and then it will kind of develop the system So let's dive in First can you tell anything my health from this picture something I might not know Something I should be concerned about Look at this My tongue is healthy Pink white coating central groove fissured tongue pattern Get out of here No I don't No you mild Look mild redness It's like yo you're you're spending too much time on the screen It's like That's true That's true Like small sun freckles I do have too much UV exposure Nothing alarming Josh modifiable risk you only see a sun exposure Daily sunscreen plays compounding dividends I do need to be doing that My bad Um anyways I like this I'm not an a clinician like tire All right So that's a little embarrassing about me All right I forgot for a second what I was doing All right All right I have a general belief about an AI system that an AI system needs to have context about a problem It needs to have context about the sender and it needs to have context about the receiver Um it also needs to uh doublech checkck itself It needs to identify the role that it is to play It needs to have really complex inputs It needs to query the user And so what I'm trying to do is I'm actually trying to build a single prompt that will guide a large language model to chat with the user to AIify whatever problem they have So in the true spirit of this prompt what I want you to do is ask me questions about my goal to understand it before we try to solve it And I want you to keep asking me questions relentlessly until you understand what I'm trying to do And then once you know what I'm trying to do we will create a prompt that will basically help people chat with a large language model uh and guide them intelligently to create an AI system structure process these prompts just for them Now ask me all the questions in the world to help me understand So we're doing something a little meta here We are talking to the model to have it help us basically diagnose Now the single biggest problem with with any I give it a bunch of context here but the single biggest problem is that there's context in your head that doesn't exist on the page yet The model doesn't know it and the model needs to know it So as part of this process we are going to help the model get the best out of you which is sort of the biggest thing Yeah Yeah Yeah Yeah Yeah Yeah That's right All right So AIing basically the my my kind of general thought is that it identifies what problems AI is good at what problems it's not good at like it's very good at research it's very good at narrow problems it's really good at this thing I called I my kind of envisioning principles creative constraint with context it is very good at being creative but only if you constrain the model and you give it context so it needs to have those things so part of this needs to be what kind of context is really important and that might be uh information from a CRM it might be uh information from a user's head it might be from gong calls it might be from anywhere the context could be from anywhere uh the creativity I think we have to let the model decide but constraint is really important which is the model has to kind of know what is the what are the subtasks that it can do uh that are really well constrained where the model can be successful and another agent another program uh another person can pick up from there So really we need to figure out how do we divide things into small context pieces and also sorry provide enough context but but the context should be relative to the constraint and it needs to provide constraint uh for the to be able to accomplish system We also need to figure out reversing backwards from the goal Um let's just focus on go to market systems That's sort of what I do I am going to um give you all of my um like my training information so you know sort of how I think um this is just for go to market stuff I I I want people to to have some flexibility because I think other things could apply but really we should probably stick it to go to market stick Um so uh and this is just my training so you don't have to overly on this but this is you just get a sense of how I think Um yeah so two concrete examples uh uh just to give you an idea like one example is I have a bunch of sales reps and how or I'm trying to AI my go to market Well one example that you might need to figure out is like okay well what research matters what are discrete units that AI can do it can research accounts It can identify good prospects It can help write messages What AI can't do is it can't know it can't talk It can't really basically go provide value to the users independent It can't take a sales call Um but it can think about the people in the process and remove them Another example here is like let's talk about it from my friend Kevin Dorsy here he thinks about evaluating the reps via Gong right i just want to know what reps are doing what well with Gong data and Gong doesn't really support this So an example would be okay well the AI agent can basically figure out like are they asking the right questions and that could be an AI agent and that's like and the AI agent should basically help sort and prioritize those things So maybe that's not the very most important thing Maybe the most important thing is are they asking enough questions are the right questions so it kind of has to be able to uh ask questions of the user first There's going to be a huge interrogation period to really understand what problem uh the person is trying to solve Um yeah basically it's just asking questions until it has the context Um and yeah this is like five W So you're going to have to like basically envision a Socratic method You're going to have to break down problems into specific steps Um and the question should go deep because you really can't understand a system if you don't know how the user thinks about it And you are going to have to sort of reorient the things because right now jobs are a poor container for problems So you're going to have to say okay this is how the user should now interact and this is exactly the input uh that we need that for the AI model will do really really well at because this is a discrete task Um but then the user needs to do things like have a conversation with the prospect the AI can't do that or the uh they're going to have to synthesize uh some of the prompts and or like the prompts will come back and the user gonna have to synthesize Um uh yeah I think people generally struggle because they're they're trying to bolt AI on top of their existing system So I'm going to talk to you about how I sort of um this is my website so you can sort of get a sense of how I talk about it Um uh that they just try to throw AI on top of existing roles and that's not what I'm looking to do I basically want to be able to dissect a human role into where AI can be successful and redefine what that human role looks like Um basically the output I think is going to be here's what AI can do well Here's what your people can do well Here's the tools that you should need to use Here is exactly how they should interact here is um like uh AI agents that you can use um and it might have to be multiple parts because I know well actually I think these models can now export a lot of tokens and I'm going to give you an example of like an AI agent So possibly as part of this you're actually going to uh define an AI agent Um and this this is text from some of my AI agent building Claude Project Uh and so hopefully you have enough now to ask me more questions to understand exactly what I'm talking about Yeah interrogate me All right You know sometimes I'm I don't know This video is like a little weird because I'm like pausing I feel like I'm large language model just like stalling up Oh so what I So you haven't understand what just happened I have an AI agent building process You should go buy Agent 7 if you haven't It's $299 It is the world's best AI agent building course I've sold $80,000 of the course since October 1st People are clamoring about it They're very happy I've had to refund one person I don't do refunds She was a single mom She didn't know what she was buying She didn't use clay I think she got very excited about buying an AI agent Uh oh wow It's so big Uh it's uh the prompt is all right Run that Um it it was too big Uh yeah So I I said "Look I don't need to." But I did say it was funny when I first posted a seven I said "If you're unhappy consider changing your life circumstances." I really try to dissuade people from buying my courses unless you're like really sure that you resonate my content etc Um I'm a bad salesperson that way I'm the opposite of a used car salesman I'm like don't come to the lot Don't buy anything Go home and ride your bike It's actually it's a I'm very proud of it It's a very good course and I do think you'll enjoy it Um okay Let's go talk back to AI now that it's done Oh I'm intrigued by your intriguedness Thank you Yes Yes And really it's going to be a single sort of uh artifact that in Claude Yes it is a it is a messy human frame problem Yeah we need to scale our outreach and this is why you need to get to the heart of it Um yeah so for example an a uh an S str container for a uh you know in the AI world you might think of an SDR's role is like talking to customers and researching for example these are two tasks well researching can be done by AI but what type of research right AI can't just do a really good job at like generic research is like well what do you look for and why and why does it matter and so AI can then split that into tasks where it's like okay can I and then AI can evaluate can I do a reasonable job at this can I reasonably understand this problem and so it really needs to interrogate the user break down things into uh semantic parts and then figure out what are what are humans good at and what is AI good at And AI I think is really good at like doing these they don't have to be they're discreet task which is things that you actually can find online So I want to make sure that and you can just assume that the AI agent has like online research capability and eventually we'll be able to talk to tools And so the first thing you probably need to do is once you understand the problem actually the second thing once you understand the problem then you need to remove out the human element What are humans going to do really really well and that's like for example discernment Humans are going to be able to look at some piece of that puzzle add their feedback to improve the quality And humans are also really good at talking to other humans AI is like really bad at that So those are some things that you can uh uh evaluate around Yeah Like if I want to AI AI my sales team that doesn't mean anything People don't really understand So really the question you might ask is like well what does your sales team do where do they spend most of their time so like you should really hard be against questions like AI if my sales team that doesn't really make any sense Um there is an art the critical pieces of context are what are your what are your team spending the most time on and what do they do today and that's like and then you should take Charlie Mer's inversion principle which is like if you had unlimited amounts of time what would you do so one of it is the art of the possible and then the other is the sort of what's taking up the most time so and I think the blind spots are they just again they try to like bolt a to their existing systems This really takes a reimagining of the org people's roles and it it it's going to take uh strategic frameworks It's going to take like a lattice work of intelligence You might look at some of the stuff that Shane Parish has talked about right mental models Uh and so you're going to need to be able to like bake in some mental models here to help the humans think because we have a very poor understanding about AI You might go online and get a little bit of understanding about how other people have thought about doing the types of things that I'm talking about Even in this conversation you might kind of realize that there are opportunities to further improve the quality of what we're talking about Uh by the way this video is almost 15 minutes so I'm going to post this to LinkedIn If you want the rest uh request my YouTube just drop Don't tell me your favorite pizza topping because you did that last time and it was Someone said mayo Don't do that So why don't you comment your favorite flavor of ice cream that's like more tame Yeah Yeah I mean the the company's context is really important and you really need to get this out of the leader and also you need to offer things that you can do So you have to understanding of your own capabilities like you can search the web you also can do deep research you can write code you can design things So you need to have that kind of baked in sometimes the AI models don't really know what they're capable of Um um I don't I think that the pattern is really just the model asking questions from both angles sort of top down and bottoms up sort of the inverse framework to really understand what is happening and so it should be sort of a longg guided conversation before the AI takes a crack at creating an artifact that is sort of the AI blueprint for the organization My company's called blueprint and you also should brand the out of this when people do it say like invented by Jordan Crawford Put my name everywhere I want to make sure that people know that uh that I'm the guy Uh I'm just being I'm being a little physicious Yeah I mean like I definitely think that data is sort of a moat here I'm going to put a data prompt that I've used in the past uh here so you sort of can understand what that looks like Uh yeah I mean I I I think that it really should it should focus on create like I I you should go and look at and figure out how many tokens can Claude Opus 4 actually output and you should do as much as possible within that framework and then you should basically tease additional value It's like hey do you want me to do this thing next like so this sort of mega prompt should really know everything that is going to be important here Oh wow As AI capabilities evolve maybe the first thing the prompt should do is get an understanding of what AI is capable of that day That's a great point Um it's a great point It should be designed to evolve Okay Ask me more questions unless you feel you have a good understanding Is it weird to watch videos of me talking AI while you're here as a person i think it could be weird Waiting to super whisper a transfer thing is a great time to meditate to find your center to figure out what makes you you I like how it moves to nine 9% You ready eight more I like how it's like you know he's going the distance He's going for speed all alone All alone in a time of need uh execute the uh yeah I have to say that because I pasted two things and if it's pasted here I don't know it's like great I will maybe I think I should just never edit my videos unless I can do it Yeah AI system design prompt blueprint framework A specialized system designed to help organization fundamentally remember context constraint creativity Yes pattern gray [Music] Yes Yes Bam Inversion The five W's Yes This is great Wow All right 20 minutes Uh if you would like access to this comment your favorite ice cream below Thank you so much I'm so grateful I really appreciate you coming and watching my stuff Yeah I hope that you found this talk Very useful And have a beautiful Oh I like I forgot that I put all that LinkedIn stuff in the mail It just ignored it Amazing All right Later folks