Read the transcript, 12 chapters
1:01 What this class covers
If there's anything specific that you guys want to learn, just let me know. Um, we are going to go over prompting. We're going to go over context engineering. Uh, if we have time, I'm going to try to get into tools, but I don't know that we'll have time just because it's only an hour. possible.
Yeah. And it's this class will be um I'm trying to keep it kind of universal. So if you use ChatGPT Gemini like whatever AI you use this class will be applicable for it. And if there's anything specific that you want to learn, let me know. Um because I'm Yeah, I've got I've got planned about an hour or so, give or take for questions and if I can get to tool use and stuff like that. So,
extra papers.
Thank you. Nothing fancy, but enough to take notes on.
You're so welcome here in case anyone else needs it. Good. How are you?
8:50 Early questions from the room
I mean, I'm excited to learn whatever we're going to learn today. I read the summary, but I was like, who knows? But I'm really intrigued by the idea of like creating a AI assistant.
Okay. Yeah.
Real estate. Yes. You know what I mean? And
I've got a buddy that's explored that not in real estate, but like in his own,
you know, environment, if you will.
And so I'm actually not as techsavvy as I should be in a lot of ways. I would love to learn how to do that. And then also, am I allowed to? Because like legally or whatever, am I going to be having AI do tasks that have to be completed by a realtor or you know what I mean?
Element.
No, 100%.
I'm very interested in all that.
Yeah.
Yeah. And that's um it's that's actually I'm glad you said that because one of the things that we're going to talk about today is kind of engineering your context. And so how do we take uh it's kind of the next level of prompting, right? Prompting is what you say to the AI right then and there. Whereas context is going to be here are kind of the persistent details in my life. And then also you can you can throw in stuff that's dynamic and changes. Um like right now I have on my phone um I just set it up so that I can talk to Claude via Telegram. And so I have all my context context in there and it I'll say like these are the of tasks that I got done today. Here are new projects that I
added like update my um personal operating system and tell me what I need to do, right? And so because one of the things that happens especially in today's society is you get so much digital clutter like digital data whether it's good bad otherwise um that what they call entropy management is like really really um really powerful concept because it allows you to stay organized and this is stuff that AI does really really well is uh helps you stay organized. So you can put like I put all of my stuff that I get for the day into a folder and then I run a command and I have an AI that goes it opens each file.
It looks at it, tells me what business it's associated with, renames it, rehomes it, and then logs it. And so now time doing,
right? And I'm not saying that I'm above like labeling files and stuff, but I'm saying if I don't have to do it, I'm not going to do it. Right? And so like that's the kind of stuff that AI is
But yeah, and like and that's that's one of the biggest things is giving it giving your AI the ability to kind of pull um contacts and be like the because everyone in here has different resources like there's a billion ways to sell a house, right? Like you can do cold email, you can go to networking events, you can like and that's just to get leads, right? And there's different steps that you can take throughout the whole thing. And the thing I love about AI is it can impact every part of that.
The thing I hate about AI is it can impact every part of that. Meaning that a lot of people will start to build half bridges and then they're like, "AI sucks. I tried it in every part of my business and it's just not getting me anywhere." And it's like, yeah, it does suck if you get halfway with it, right? And so, like, I would much rather you guys focus on saying like, "This is the one thing that's most important to me right now. Let's do that." and then um kind of expand the capabilities because a lot of people like I said play with it they get halfway to the destination they run into some type of challenge or it gets hard or it gets unconventional and it's just like okay well I guess this is
as far as this can go right instead of using AI like a co- intelligence and being like hey this is the problem I've run into here are my constraints here are the resources that I have to solve it help me
right
right so out of curiosity the stuff that you're doing in your own personal, you know, AI agent stuff. Uh, is that do you have
like bridging code to do? Are you taking it that far or you're literally just able to use the AI on its own to go from point A all the way to the complete task.
So, it depends a lot on the task and how much of the inputs and outputs are what we call fuzzy, meaning that they're not a definitive thing. So with your business processes, um the things that don't break are going to be what they call deterministic. Meaning if one thing happens, this next thing happens, right? Whether that next thing is nothing or whether that next thing is making a decision, whatever it is, it goes from boom to boom to boom, right? And that's effective in the context of a business, right? And so what you want to do is you want to have as much deterministic as you can and then everything else should be intelligence based. And so you do have like Python scripts and so it'll say like go from here look at these variables and see what the situation is.
um if the situation is this, run this deterministic code. If the situation is this, you have to make another decision. And so it's always kind of building that decision tree of like, and this is one of the reasons that I would recommend that everybody starts if you're going to implement AI, start with something you do really, really well and you know really, really well. Because if you start with what you know and you kind of compound your strength, you don't have to learn a system plus how to use AI at the same time, right? You can just figure out what you do well and then add AI into it.
18:30 Searcher vs. delegator mindset
Woohoo. All right, so I have a critical fear of public speaking, so I'm just going to be standing here silently and pressing a button on my computer. I'm just kidding. Um, all right. So, these I I uh very much love questions. So, this is an hour class, so I'm going to try to kind of minimize questions, but I also want to make sure you guys have a way to get questions answered. So, if you have a question that's preventing you from kind of advancing with what we're studying, whether it's prompting, context engineering, raise your hand and ask it. I will answer to the best of my ability. If you have a question that is maybe like, what does GPT stand for? I don't know, but I do want to know. scan this and ask it to this and I will
answer it uh in an email that is going to be uh forthcoming after the class. So that's what these are. These are for questions that may not be uh relevant to kind of advancing the subject considering we're here for an hour. All right, that being said, let's get right to business. So um most people with AI, the way they start using AI is the same way we use Google, right? Like most people when they use AI they go they ask it a question they get an answer and then they do kind of what it is with that answer. We are moving into a place where AI has access to tools. It has access to uh what we're going to talk about like additional context and a lot of um a lot of utility in moving towards a mindset
of delegating to the AI. So, we're moving from a searcher mindset to a delegator mindset where it's actually able to pull some level of task off of your plate, right? Like, we don't want just the AI to live in a chatbot. If the AI stays in the chatbot, there's a lot of manual work for you to do, copy, paste, going and finding, doing all that kind of stuff. So, we want to get to a place where we are um going from Google to basically using it like an operating system. I was kind of uh talking a little bit earlier about how I have kind of all the context that I need in my phone and I can send a telegram message.
AI will route through it and tell me like, hey, these are the tasks that are most important for me to do today. That's kind of the the difference in an operating system, right? Like an operating system is something that is dynamic and provides you value on an ongoing basis. Whereas if you're kind of the searcher mindset, you you just go and you get value one time when you execute a search. Um so my biggest filter is if it doesn't save time or make money, I don't care about it. Um that being said, I don't care about it as it relates to kind of teaching in the class. I think AI is super fun. I love like designing logos and being creative and stuff that isn't making anyone money, right? like you you've seen some terrible logos out
there of companies that are doing pretty well. Um, but that's kind of one of the things I said in the beginning of this class, AI is great because it can do everything. AI is terrible because it can do everything. This is one of the side trails that you have to kind of look out for. You have to whatever your personal filter is. You have to apply it or you'll because it can do anything. You'll just go off on side trails and you'll be like, "How am I looking at logo color shades and it's been three hours sitting at catching a tea?" Right?
because it's just very easy to do. Um, so what we're going to learn today is we're going to learn mental models of working with AI. We're going to learn prompting, context engineering, and how to verify results. So, uh, context over cleverness is basically the idea that prompts can get you so far. So you can have a prompt and it can get you really really good results, but if you want to be efficient, if you want speed, you're going to want to develop a kind of a p what's called a persistent context. Meaning if your name is Larry, your name is likely always to be Larry. You can put that in your context, right? And so now the AI when it says um write a listing description for X, Y, and Z, it'll pull like let's say you hate the word dream
home, like let's say that's one of your realtor pet peeves is you absolutely dislike the word dream home. You can throw that into the do not say list for AI and it will not say it, right? Because one of the things, especially with realtors, right, is a lot of people with AI use it and what they don't realize they're doing is they're actually kind of chopping off human connection. What I mean by that is I I personally am like I'm not a realtor, but if I were to write a listing description, I could go to ChatGPT and I could have a reasonable listing description written for me, right, without very much experience or anything. However, everyone in this room has expertise has um this is very very important. It's something called non uh public proprietary data, right? So chat
GPT is trained on the internet. So if it's public data, it's likely been in some AI somewhere. Your advantage is the things that are not publicly accessible, that you're giving it the right information.
24:02 Mental models for AI
So mental models for AI. Uh the reason I teach this and I personally am not someone who enjoys theoretical teaching. Uh and so this kind of uh crushes my soul a little bit to teach this but it is super duper important because if you don't have the fundamentals for thinking about how to use AI, you get locked into the model that you're using. What I mean by that is let's say you know ChatGPT, you know it really well. Let's say that Claude comes out with an AI that is leaps and bounds better than ChatGPT. If you don't have thinking tools to be able to switch providers, you're going to be stuck to ChatGPT and you will always be at a disadvantage to people who are able to move quickly model to model because it changes so
fast, right? It's kind of like um it's just it's similar to like cell phone providers, internet providers, they all have things that they do better, worse, all that kind of stuff. So, mental models for AI. Um, and it's all about kind of the thinking tools because you can ask AI, get a generic output or you can the the text is definitely not big enough for you guys to read. Um, but it it helps move that generic output to what success actually looks like for you, right?
Because there are what what success looks like for me is going to be different from what it looks like for you. Let's say you guys are really good at sending cold emails. I'm not really good at sending cold emails. So, if you guys are really good at it and you guys have good context and AI, you guys are going to be really, really good at it. It jumps your strengths significantly farther ahead. The things you want to look for when identifying is a task useful for AI is one, how well do you know the workflow for that task? So, if you know from A to Z and you have that process down really well, that's a really good place for you to start with AI because you don't have to learn AI and a system, you can take the system
you have and put AI into it. Um, and that kind of teaches you, okay, this is how I can implement AI and then you can start thinking about using it with your weaknesses as well because you have to develop kind of not linearly but it helps if you have kind of a process in place to work with AI and kind of understand how it can best uh best impact your business.
26:30 Start with the win
So first principles thinking define the outcome not the process. So we live in kind of a world where the processes are changing really really rapidly and I think we're going to see that more and more as AI continues to become prevalent. Um and so it's having people completely reimagine things uh such like even the internet right now is a there's a lot going on on the internet which is completely invisible to the human eye but is very visible to agents things like uh generative search. So how do you get recommended in ChatGPT? How do you get um how does an AI um kind of find your business? All that kind of stuff.
So the the fundamentals of discoverability are changing, right? And so you want to look at what outcome you want. So the process may have been, you know, okay, I have to go and um collect emails off a website and then I have to load them into a spreadsheet and then I have to, you know, x y and z. But if you look at kind of the AI and you say, "Hey, these are the resources I have." And it again, context, context, context. These are the resources I have. This is the problem I have. These are the things that are kind of guardrails or restrictions that you can't do. Help me solve it, right? And so start from the beginning and kind of engineer the solution. Um, so as an example, help me with social media turns into I need
three posts to generate at least one inbound DM this week. here's what worked last month, right? And so you're moving from a very generic prompt that you would just use one time to something that eventually you'll be able to just copy paste, copy paste. Um, and so the questions uh that I always ask before I work with AI is what's the win state? So if my dream outcome, if I were to get everything I wanted out of this interaction, what does that look like? How will I know it worked? What does good look like for this task? So, what would be some things that would make this task fail? Like if I saw this and I was like, "Nope." Um, do I have an example to show AI? So, if you've done something, you've done it well. Examples
almost always work better than description. So, if you have something that you can put directly and you're like, "This email, I've sent this to clients. It always makes them super happy. Like, they're like they love it when I follow up this way, blah blah blah." Put that in as context. like make sure that you're giving the AI as much information to win as possible. Next up is the udaloop. So may or may not be familiar with the udaloop, but it is observe, orient, decide, act. It was basically what it is is this is kind of the loop that you want to run when you're prompting in your own head.
So observe what is the reality of the situation right what is um what is true right now right and then you want to orient so of what's real where am I going which direction do I want to go then you need to decide on the plan to get there and then you act so for example if you say to AI write a uh listing description for this house you know whatever so you observe the output you say okay this is really good, but it doesn't have the bedrooms. It's lacking this water feature. It's lacking, you know, this this thing that gives the house extra charm. And so, we're going to add that to it. Um, we're going to orient. So, we're going to say that, you know, buyers in this market are really looking for the following
things. So, we want to make sure that's highlighted. Then, we're going to decide, okay, do I want to go in this direction or that direction? Let's focus on this one decision. And then you ask. So, then you send the prompt. It comes back and you set you just start it again, right? you observe okay what what is in here that I don't like that I need to change that you know whatever that is right and so the goal is to if you can get good at this you can get good at AI because it's the less dynamic decisions that you have to make and the more you can put everything into a framework where you don't have to make a bunch of decisions in a row you can just kind of be like okay it's entering observe okay what's
true okay or the more you can put this kind of running in in the background of your mental framework, the easier it is to be like, okay, this is what I need to do next. And so the speed with which you can get through observe orientide act speeds up dramatically as you do it more and more. Um, and just so you guys know, I am I have like a client portal or a student portal that I'm working on that is free that you guys will be able to log into.
You'll get the slides. I have a workbook that goes with the class and there's a bunch of other stuff in there too. I don't remember exactly what it was.
Yeah, I'm on I think I'm on that right.
Okay. Yeah. Yeah. I spent all week
going through the slides like automatically actually.
Yeah. Yeah. I mean, there's a bug in that. Yeah.
Sure. I'm right.
Well, it's funny because I actually fixed the bug for my presentation like probably 10 minutes before class because I was like, "Oh yeah, it's auto advancing."
Just making sure it wasn't just me.
Yep.
Um and so, but yeah, you'll have all of all of the details. So, one of the things that's really useful with AI, um, there's, so I think it was starting from nothing, right? And so, one of the great things about AI is you can go and you can just start. Um, because speed, if you go in there and you start to build momentum, it will get you where you want to go. But if you're just staring at the blank screen trying to figure out, okay, how do I make this perfect? You I mean, it's not it's it you're going to spend a lot of time and very little action. And the cool thing with AI is you can go in there and just start action and end up at the same place much faster than if you tried to
be a perfectionist. That was me. I was the perfectionist. So this was something that was really really useful for me is getting in there and just being like, you know what, I have an iter iterative mindset now. So this first output, it's not the final output. I'm going to be able to do this as many times as I need to get what I want. And then once I get what I want, I'm going to ask the AI, okay, from where we started to now, how did you end up going from there to there? And now I'm going to have a prompt that shortcuts all of the 10 messages that I said, right? So instead of going back and doing that every time, I now have the AI telling me in its own words, this is how we arrived at this
solution, the solution that you want, right? And so you want to make sure that you're kind of just making mental notes and logging that kind of stuff. Um,
33:14 Where AI needs your judgment
skill gap is something that is really, really important. Um, because what it does, AI will make whatever you do go faster. It'll make it bigger. it'll make it whatever. If you have a process that sucks, it will suck faster. It will suck bigger. It will be terrible, right? So, it just makes bad worse if that's what the goal is. Uh if the goal is to just go as fast as you can, right? So, you want to have an understanding of like really what you're after and have it so that the system is dialed in so that you can go faster. you can what they call scale at an appropriate rate and you're not scaling something that's going to break in a month when it gets volume that is um you know commensurate for it to to handle right so
but yeah so this is just you know if you're a novice uh or if you're photographer and you use AI editing you're going to have polished garbage right so that's that's where you're at um but if you have expertise and you have AI to supplement that and you can call the AI out on its bad decisions. It's like I don't think that fact is real like all that kind of stuff. Then you have a much better and everybody has some type of unique like everyone in this room has a unique expertise that they can apply to AI and get a disproportional advantage over everyone else in the room, right? Like every single person has something that they can use AI to gain a big advantage with.
Um, and so AI can help us draft and research and do kind of the uh the grunt work to making a really good output, but you have to be the one that is verifying those outputs. You have to be the one that's, you know, getting a feel on your clients in person. You have to be the one that's knowing the market. Um AI can speed up everything you do again but again you have to be that human in the loop that is judging the quality of the output that you are getting. It is always and I I say it this way um AI does not have a any type of uh legal representation. So if in the absolute worst case scenario if you go crazy and your AI outputs some hot garbage and you get sued over it, they're not suing your
AI agent, right? they're suing you. So, make sure that like you're assessing this through does this meet fair housing. Even though you've said make sure this is compliant with fair housing, you you still want to read it and be like, is this compliant with fair housing? Right? Because interpretations, all that kind of stuff, you are the final decision maker for the AI. Um, circle of competence. This is similar to skill gap, but it is kind of the opposite where that is way too small. uh where you kind of know your strengths and you use the AI to make it bigger.
And what you're doing is you're also paying attention to everyone else's circle of competence, right? Because you don't want a painter who's like, "Oh, yeah, I can build a deck." And then, you know, I've never built a deck before. Probably don't want to do that, right? So, you want to make sure that you're staying within your circle of competence with the AI and you're expanding gradually, not just um dumping a bunch of stuff into the AI and then just saying like, "Oh, here I made a bunch of outputs.
Great." Right? That that's not going to help. You need to be uh it's ideal to be living where you are the most proficient as it relates to the AI because you can uh kind of guard rail it back to the truth once it gets off base. um if you're if you don't have any subject matter domain subject matter expertise in that domain you'll need to do a little bit of research on your own and I also something that's like really really useful is asking the AI to provide links and citations to the research that it's done right because then you can just go and click the links and see exactly what it looked at um if so most people don't realize that they are the constraint as it relates to AI and that's just because the AI can go
faster. It can do more. There's just what tools do you have access to to be able to make these things automatic to have the context like context can be as simple as putting it into a Google doc that you copy and paste that Google doc into your AI every time, right? It's very very simple um stuff. And then 8020 rule. This is probably my favorite one because it means the AI does 80% of the work. um what you want to do is kind of view yourself as being the first 10% of the project and the last 10% of the project right so when you get started with a project you have to give it clarity and direction the more clear you are with what you want the better the output AI is going to give you um and
the more so basically it looks like 10% is create um give clarity provide context to all that kind of front end work. Then you have 80% of the AI doing research, building like doing all that kind of stuff. And then you have the 10% of polish. Um, and that's where we we talked a little bit earlier about um how a lot of people use AI and it kind of blunts human connection. That polish is what puts it back into a human connectable way. Right? So, if if someone knows that you're always so friendly and polite and you have big energy and blah blah blah blah blah blah blah blah blah blah blah blah blah blah and um you email some them and you're like to whom it may concern. They're going to be like, "Ah, I don't know
about this. This doesn't sound like Jerry or whoever it was that sent the email, right? Like you want it to be as indistinguishable from your voice as possible." Um, and like I said, you can get quality outputs. like I can go get a great listing description, but if I'm like funny and engaging and all these things and I can add that to it plus local knowledge, I now have a listing description that is very different than the other stuff out there. And I think just the idea of doing things in a more novel way is going to be beneficial reasonable quality. But right now, we're trying to get to the ceiling, right?
Like the floor is very easy with AI. The ceiling is much harder. And that's what we're trying to get. I I don't want you guys to just be living at the very base of AI. I want you guys to feel comfortable being like, "This is like really moving my business forward." Um because a lot of people um a lot of people use AI and they're like, "I don't know how this is actually helping me, right?" And so it's because yeah, you want systems in place that allow the AI to do as much for you as it possibly can.
So 20% that you don't want AI to be doing is if the sellers are going through a divorce, right? You know that the AI doesn't. So you want to treat that with care. Um the buyer is nervous, not cheap, right? Like so there there's things that the AI will infer that are not true or that is just missing, right? And so that's when I I keep going back to it, but that proprietary non-public data that you have in your brain is the the advantage uh that you can give to your AI. So if you can use AI well and you have that, you have an advantage. Um so context over cleverness. So this is again um just kind of talking about the fact that prompts isn't all you need, but it is a great start. Um and so it
goes from, you know, help me with this listing to imagine you're paying a $500 an hour consultant. They've never met you, never seen your client, never heard. Would you uh say help me with this or would you brief? So like you're getting a much different um different output by ensuring that again you're just context context context and being specific. So I said earlier how examples feel like you've got three emails that are written I I'll give you an example because this is probably more pertinent. So, I'm uh very very uh classical introvert. Like, lock me in a closet with a computer and let me go to work all day. I'm super happy. Um not great at customer service as it turns out, right? Engineer brains don't work that way. So, I for me personally, I was
like, who is the person uh in my life that is the best with customer service that makes customers feel like, oh man, like every time they walk away, they are happier with the company for having interacted with her, right? And so went to her Now, I have a similar customer service skill set for her as it relates to warmth and friendliness, right? That no one is going to I have never had someone uh respond to one of my emails with, "Oh, that was so nice of you to say."
Because that's not that's not what I do. I'm very direct, very like what it is. But now people are like, "Oh my goodness, you're so nice." Not really. Uh the AI has you. Um, so but yeah, you just have to change kind of the way you're thinking about the problems, the way you're thinking about using the AI, and it'll change the outcomes that you're getting with it.
43:39 Older prompting method, now retired
if you can.
Are these chats saved? Like you could go back and say, "I want to use the same with a new thing." Yep. Yeah. So, you can do that, but it's it also kind of lends to that old phrase, uh, no man crosses the same river twice because it's never the same man, never the same river type thing, because every time you add context to it, it changes just a little bit. And so, what what I would recommend doing in that situation is having the prompt somewhere. So whether it's a Google doc or something like that and then doing fresh context window every time meaning go to a new chat have what you wantformationally that you know works put that in there and then start it with a new chat. Um because yeah and at a certain point
there's something called context rot or context bloat and that basically means that um you'll notice that the like let's say you said don't ever use the word dream home and you notice that it's starting to use the word dream home. What's likely happened is your guardrails are gone and you have too much in the context window that it started forgetting about things that you told it to do. Right? It's a very common thing where you have like this massive chat and then you're like, "All right, I think I'm ready." And you ask it for something and it forgot everything that you said within the last 20 messages, right? And now it's just and you're like, "What happened here?" And so that's that's why I would recommend trying to keep the context fresh situationally if you can.
1:00:42 Context engineering and Context Cards
Context engineering. So this is kind of what I've been talking about for majority of the class. Um, so prompting is that one time asking a question. Context engineering is how you get an assistant. That is how you create something that it knows your projects, it knows your tasks, it knows your personality. So it knows even a little bit the way you're going to complete the task, right? Like if you're someone who hires out instead of developing the skill as an example, the AI will eventually learn, hey, this person hires out doing things that they are not proficient with, instead of learning how to do it. It'll start making recommendations for vendors instead of making recommendations for courses to learn how to do the skill, right? And so that context just continues to build and grow and it kind of it learns and it
knows about you. And this will be one of those things that um people are going to build their independent systems for it. And I've seen this happen a number of times. So people are going to build independent systems. I build an independent contact system for me. What is going to happen is eventually the frontier labs are going to crush that eventually. So like Claude, OpenAI, whatever are eventually going to solve the problem of personal assistant and everything I've talked about here today will go away. So just be aware that you're learning a depreciating skill right now, but only because the skill will be um basically crushed by the AI companies themselves. And so at that point, everyone will have access to like a full context system on themselves. You guys knowing how to manage the context
and use it better is going to be an advantage for you in the long run. But in the short term, um, the AI companies do not have this. So understanding it now is going to be beneficial. And then when the AI companies do get it, you'll have a deeper understanding and be able to be more proficient with it when it does come. So um, so this is very similar to front door key versus master key. Um, context engineering basically gives you the master key. Not only does it give you the master key, and it can also give you a master key to a nicer house, right? So, because of all the time and effort and things that it can save you by building on the front end, and it don't get me wrong, I'm not going
to sugarcoat this. It is a pain in the butt to build on the front end, but it is very worth it. Um, and it's becoming less and less of an obstacle to actually build. So, what you want to do is, yep, context engineering is the delicate art and science of filling the context window with just the right information for the next step. So it's not filling it with everything, but it's filling it with the ability for the AI to make better decisions. So again, you this is all included. So this is a context card. So this is kind of basically the identity, the voice, what knowledge you have, and what guardrails are you interacting with your AI app. Um, so again, another one of these, it's just kind of if you load the context card, you get better outputs.
Um so the way this most people um do context engineering is they set up the first three layers and then that's your reusable um your reusable window. So the system is what's your persona, your voice, your constraints, your expertise, kind of everything we talked about in prompting, right? Like everything that we've talked about in prompting what goes in system context. So, one of the things that you'll want to think about is your personal what they call system prompt. Okay? A lot of people don't realize, but when you open that uh ChatGPT window or anthropic or whoever provider you use, there is a prompt in there already that you never see and it's the what they call the system prompt. So, it is the prompt that all users funnel ChatGPT or Anthropic through and it's kind of the start. Um,
and so the better you can get at understanding what your system prompt looks like. So what are the things you do well, the things you do poorly, the things that um the like the problems you solve really well, the resources you have available, all that kind of stuff should go in there. So memory is a big big problem with AI. It's something that they're working on. Uh the smartest some of the smartest people in the world are working on it right now. I do expect them to solve it. um to where it's good enough to the point where chat GBT will be able to go back to like your initial conversations with it and pull context from that. It's not anywhere close to that yet. Um, but what that means for right now is that let's say you have a
client that loves whites space and they want luxury everything and da da da and then you have another client that is value conscious and wants to be as affordable as possible and find as much value as possible, right? Like the same communication is not going to hit the same way for each of them, right? Like it's going to be very different. And so just like you can build a context card for yourself, you can build a context card for your clients and you can build a context card for your deals and your properties, right? And so now you have context for three different entities. You, your clients, and your deals.
1:06:09 Room questions on Context Cards
I'm a little confused on context cards. Are those supposed to be just in a prompt that we save and then start?
Yeah, exactly. So what's really the context card what you want to do to build it is kind of the way I did it is I uh went to I think it was quad I transcribed my me talking because it's a lot and I just basically had a brain dump of everything that I thought was useful for it and I said this is my personal context help me sort out what's meaningful and what's not meaningful and then it sorted it out and then I went back through and read it and I was like no I think these things are more meaningful than you're going to be prepared for blah blah blah blah blah Y and so it built it out. And then what you do is you get that and then you take it to like a Google doc and you save it
in there or save it even in like a cell on a sheet and then you can grab it and paste it in there. Or if you're using like a desktop version of the app, you can actually have the app go in and grab it out of the doc and pull it right in for you. And so, but yeah, so you're going to be pasting that into basically every interaction that you're using for kind of a productive work output conversation. So, the conversation um like we talked about is what is discussed this session. Um, and you'll want new conversations, new context windows for different um tasks that you're starting because it it will impact uh the next thing in the conversation.
So, if you have like a 20-minute conversation with with it and then you go to a different topic in that same conversation, it's being shaded by what you previously talked about. So, if you don't want that, you need to just bounce to a new conversation. Um, and this is context rot and context bloat is what uh causes your output quality to start degrading as you have longer conversations task. So this is just exactly what you're doing right now. So like why are you here to use AI? Um, and so if you can make layers one through three consistent and as easy as possible, the prompting layer for number four becomes literally as easy as saying, "Hey AI, write a prompt to get me this outcome."
Because if you're not using AI to write your prompts, you 100% should be. Most people like, like I said, go type and then send it to get the response. One of the things that I love to do is say, "Turn this into a system quality prompt that gives me elite level outputs and then tell it exactly what I'm saying." And so it's not going to give me the output. It's going to give me the system prompt to give to the AI to get the output, if that makes sense. It's very, it's a little convoluted, but the AI is smarter than me, so I want to use it in as many places in my workflow as possible.
Do you also save those prompts as well? Like you can go back and pull
It depends. It depends on how much utility I've gotten out of the prompt. Um, but yeah, so like a majority of the time I will if it's something that I think has any potential to be used again and I really like the output.
Um, and that's something else that is a lot of people don't think about. If you save your prompt, it may not matter the next time you go to the AI model. And I say that because the AI's advance so quickly like the model comes out and so you may put that prompt in and you may get a different result be like what the it just happened here right and so and it may be better it may be worse like you might put it in you'd be like holy cow I didn't expect this thing to make a bar chart you know and it's like oh wow I made a whole presentation right so it you just have to be aware that the models are changing too and that's something that a lot of people don't account for when they're
building this stuff Uh,
1:10:00 When a chat starts drifting
so here's some signs of context rot. If it start AI gives generic responses, if it forgets the things that you put as constraints, if it the outputs start to drift, like if you're like, I don't recognize the kind of language it's using, or if you find yourself having to reexplain things over and over again, you're likely at a place where you should ask the AI to give you a summary of the entire chat and then move it to a fresh context window. Uh, So identity. So voice. So this is just kind of a recap. So I'm just going to kind of bl through it. Identity. Who are you? Voice, how should it sound?
Knowledge, what information does it have? Guardrails, where needing to stay on the road. Um, dream. So the do not say list. This is again one of the guardrail things. Like I highly recommend if you have pet peeves that or if you have things that you would never say, stuff like that, add it to your do not say list, add it to that. Um, and then again, going back to examples being better than descriptions, if you can have it match your style and tone for the things that you like, have it do that 100%. Give it really good examples. This is actually uh something that should kind of be running all the time in the back of your brain. Like you send an email and you're like, "I nailed that email. That email's great." Save it. Save it and keep it as an example
for your writing style, right?
Do the guardrails travel to each chat or do you have to say again, "Hey, these are the guardrails."
So, if you put it in your context card, it'll travel with the context card. Um, and depending on how you have your AI desktop set up, it can. Um, but a majority of time it'll be you have to either have it in a context card or put it out there. um each time. But and this is a great way to do it. Super simple prompt. Um again, you guys will have access to this. Um but you can have it analyze the writing samples and then you paste your examples and it'll give you kind of your writing style.
Uh so write in a warm professional tone, write in my analyze style, short punchy sentences. And so it gives you your writing analysis and then it with your examples it can really really do a good job kind of matching um kind of predicting what you would say in that situation. And so once you build out the
1:12:25 Context for clients and deals
templates, you just continue to use them and then you end up, you know, with listing description contacts and you can build different context for different things, right? So client communication, neighborhood expert, market update context, lead qualification, and then again, not going to go super in-depth with this, but this is how you kind of stack that into a workflow. So you can have the transaction deal sheet, the client persona, objection handler, and then you just kind of all parts of the deal you can filter through the AI and it's like, oh, we're having a pricing objection. What are some things I can do to overcome that? And it will have the context on that client. So it knows like, hey, it's really important that they're in the school district. Use that in the negoti or like whatever it is, right? So, like
having that additional context gives you the ability to really um really push forward. Uh 20 message rule. I don't personally have this, but I know people that use this to kind of avoid getting into that context rot like bad AI situation. Um and so you just new window. Um and then the these are the biggest mistakes with context engineering is pasting everything just in case. Um, same context for everything. Never updating. Like a lot of people will build it and be like, "Yeah, we're good." And then six months later they're like, "This sucks." Yeah, it's never been updated once. So, make sure and you can update. You can have the AI actually do the update. Like you can copy paste and say, "Hey, update this for me."
Um, and then each platform, uh, you can put it into the baseline of the platform. So like chat GBT for example has like personas and you can paste right into that and so if you always use chat GBT for your real estate stuff paste that in there and it will always take that mindset to kind of work with you. So custom instructions in chat GBT also projects projects and claude and gems in Gemini. All right, action plan. And um so the goal isn't to prompt a chatbot. It is to build a digital assistant that is actually helping to move your business forward. And um we Yeah.
Yeah. Yeah. No, it is.
Yeah.
Oh, 100. It's it's um
it's completely like another
it's I I often imagine that this is what it was like when the printing press was developed. You know what I mean? Like it's just a complete technological shift of like the entire culture
and the slides and stuff get
Yeah. So if you have the um if you create an account to the student portal so it's and I I'll send the link to everybody just create a link to an account and everything will be in there. You copy and paste through the whole thing. So
thank you.
Thank you. Absolutely. Yes. And if you guys have any questions, um,
1:18:29 After class: more questions
Hi. I think you kind of got around my question that I submitted, but um I'm not using actual context cards. I'm on chat. I'm at the top level. that's fun. where I've been using it for a while and it's my contextes are getting mushy where the it but I have not been like thinking anything new I'm trying to figure out how to tangibly
I'm like I have buckets it doesn't have buckets
I was the same way yeah and so yeah so it's funny so like I have like I have my client based business that I for it's very very different and
like all of it and so yeah absolutely and so um the way there's two really the simplest way to do it is just to be like ChatGPT is my real estate stuff and qu is my something is the simplest way to do it
you can do um I think it's a project in chatbt where you can kind of have that be sitting there separately and you could have have it that way
um
I have a question for speed just go start a new chat and then later on like I should drag this to the project.
Well, and you can always Yeah. And you can always ask it like um
summarize this in a way that I can put it into a new chat with fresh context to get this output or whatever it is. So, you can always do kind of little tricks like that where you have it basically
do that the grunt work of like building that out for you and then just taking and copying it.
Yeah. Um
that's I'm like, man, I wish I had three versions. Right.
Well, I mean, I kind
Thanks for coming.
Yeah, you as well. Take care.
