Automatic captions from the recording, lightly corrected. Stretches with personal details, unsourced numbers, dated news or advice the class has since changed are left out. That's 17 stretches in all. If the recording and the steps above differ, follow the steps.
Open the transcript, 12 chapters
0:00 Why connect AI to your tools
All right. So, last week we had a good talk about what did we talk about? We talked about connectors a little bit and we talked about um just expanding on the context kit and building out kind of how to take that context kit and turn it into more assets and more assets. And the goal today is going to be to take the connectors, use those to establish distribution pipelines. Distribution in the historical sense has meant um external to customers. But we're at a place now where we can create so many things that the distribution pipeline could be internal to our own team. It could be um external to vendors. It could be internal to just you. Um, it could be just like a tool that you're creating, right? And so the connectors are the way to get the assets
that you build into the places that they need to go without you having to copy and paste and do all the things that you used to have to do with ChatGPT.
Do you have any questions that you really want to get answered today?
Um, no. I have used AI. We We use Claude a lot.
Oh, nice. Okay. Yeah. Yeah. Um, however, I find that sometimes, you know, when you're trying to say what what do you think this house is worth and here's all the information and, you know, pull comps, it can't like pull from MLS. So, you still have to do the work and you have to check it.
Yep.
Wrong.
100%. Yep. Yep. Perfect. So, th those are things that are definitely addressable for sure. Um, and there's
your own.
Yes. Yes, I've done that.
Well, and there's also so now there's um agent browsers. So you what you can do is actually use uh the app to actually go navigate to the website for you cuz
the way it kind of used to work is you
Do you find that ChatGPT is better at doing the market analysis rather than Claude? I feel like Claude is more co-workers.
Yeah. So, Claude has really carved out the business niche as far as like what they're targeting is like we want to help people do productive work. Um, so I'm definitely a big Claude person. Um, I the thing that I caution people on as far as any of the models go is getting stuck to one model. And we've kind of talked about this and this is kind of the context kit and stuff like that, right? So you want to have the information that's most important to you that you can take to chat to Claude to Grok to whoever has the best model at the time that you're working with. Um right now I way but the way all these models are going is you will be able to start eliminating software from your whatever st tech stack you use because the
foundation models are able to replicate what is being done. Like as an example, um most people I know do not use Word or Excel anymore because they can do it all within the app. Like if you need a spreadsheet, you just put it in here and tell ChatGPT or Claude, "Hey, edit this spreadsheet." It does it. So a lot of people are moving things that wouldn't historically be given to an AI to the AI because of the capabilities that they now have. like PowerPoint X. There's literally no reason to spend any time in PowerPoint at this point because you can tell Claude either go into PowerPoint and create this or create a custom slide deck. So like as an example, this is not PowerPoint or anything. I built this by from scratch. Like this is pure code.
Everything that you're seeing on the screen is pure code and I did not write a single line of it. I just told it what I wanted and then told it how to change
9:13 Iterate, then ask for the shortcut
what it made. And so that's the big one of the biggest skills that you can get is the ability to make adjustments to the outputs that you are receiving, right? Because a lot of people they they take their initial input, they put it in there, and it comes out and they're like, "Oh, I'm unhappy with this. AI sucks, right?" And so like that's not the most uncommon story I've heard. But what the best operators are doing is they're getting in there, they're iterating it until so they're saying, "Okay, I want this to be blue, not red.
I want this to be this wording, not that word." And they just iterate it until they get exactly what they want. And then they ask the AI, "Okay, this is where we started. I now have exactly what I want. How can I get exactly what I want in significantly less steps than it took me to get here?" Right? Because if it took you 10 questions to get there, it has all the information of what that journey looked like. And now you can just ask it to shortcut the journey so that you can get there in one question next time. Or better yet, you can have it present a markdown file and you can upload that to either an artifact or a project. And now you have all the context needed to go right there from prompt one. You don't have to do
kind of A to B. And that's where a lot of people are missing efficiencies with AI is they have a bunch of stuff that they treat the same exact way every time it comes into their inbox or to their wherever it comes into. And instead of building one solution that helps manage that task every time, what they do is they go and they treat it like it's an independent thing every time. So, as an example, if like one lender is going to be roughly the same as all the other lenders with how you speak to them and you know what's required as far as like what they need on their end and like what makes an agent useful to a lender and all that kind of stuff, right? It's going to be relatively the same. And so
what you do is you put all that relatively the same stuff into one place. So into a markdown file, into a doc file if you have to. But markdown is my favorite because there's significantly less uh what's called metadata. Metadata is like if you look at HTML, the text version of it, you'll see uh like parenthesis, head, parentheses, HTML, body, like all the different tags and stuff that come with it. Markdown is clean of that. And so with AI, something that you have to worry about is how much stuff am I giving it? Because if you give it too much stuff, the performance starts to degrade. And so if you can give a markdown file that's very very efficient and AI does really well with efficiency.
So if you can figure out okay this is all the stuff that's relatively the same from the lenders put it in a markdown file and then so as an example um let me actually pull up Claude. So,
I did upload my um buyer's consultation into Claude. It took us really a day of having to go back and say, "Okay, this is I didn't give it a whole bunch of instruction at first. I just said, "Hey, I want to convert this um and make all the font the same. Don't do any bolding, you know, things like that." Yeah.
And then so it went through that and then we went through and you know went into more detail with it and I kept telling it now go back and audit it.
Yep. Yeah.
Go back and audit it because I already told you to do this. Now you go back and fix your mistakes.
And
12:55 Validation criteria for every output
so
no and that's exactly right. And so the way we kind of want to transition our thinking is from creating individual assets to creating pipelines that create those assets. Right. And so like just like you're saying, hey, go back and audit this. go back and audit this. The thing that you'll want to add to that is go back and audit this and validate the output,
okay,
via this criteria and then you give it what the criteria is that's different from what it actually gave you.
Okay. So, we ask it to go back and audit and do what?
Validate. Yep.
Yeah. And so, what you want is something called validation criteria. And so what that is is it's basically your custom set of rules that decide whether the output is a pass or a fail. So like maybe you guys are transitioning from a black and white color scheme to a red, white and blue color scheme.
You would have the validation gate on that one be make sure that the brand is using this color, this color and this color and is not black and white dominant. And so what it would do is it'll produce the output. Like let's say you're doing a landing page or something. It'll produce the landing page and then it will actually call a browser to open that landing page. It'll take a screenshot, submit the screenshot to the AI and then it'll analyze it to say like does it meet X, Y, and Z.
Sometimes that's still like unless you get into where you're coding like Python scripts and deterministic stuff, you'll still run into every now and then where it'll be like, "Oh, I completely jumped right over that gate or whatever it is." But if you have them in place, it happens a lot less frequently where if you're like, "The last step of this process is audit X, Y, and Z, and if any of these things are prevalent, send it back in and fix it." and it will actually send it back into the loop and it'll go, "Oh, I see what she's saying."
output and then it runs that validation gate again. And so sometimes when you're first doing this, it'll create an asset, run the validation gate, send it back in, and it'll repeat that six, seven, eight times or more because it's like, oh, every time it's creating something, and then if you have it, okay, create that, learn from your mistakes, and create rules around those mistakes that you're making, it'll help with that, too. And so you just kind of get you want to think about one having insanely clear outcomes like what exactly is the outcome I want from this task.
And then two is un the the way to have insanely clear outcomes is to ensure that you know exactly what a good output is and what a good output isn't and that the AI does as well. Um because the more you can give it in clear terms, like if you just say, "Hey, I want a I want this to have a luxury feel to it."
AI tends to have the same definition of luxury across the board. So you're not getting an advantage by saying a luxury feel, right? Like
more detail.
Exactly. Cuz it doesn't know what luxury is. It just know what its interpretation of luxury just like we do, right? And so you have to define for it just like um I was telling someone the other day like just like when my daughter was very very young I was like hey can you go grab that ball she just kind of looked at me like what what are we what are we talking about here?
She didn't know what a ball was right and so like why would she know to do that and so AI is in the same boat with our context. It doesn't know what luxury is to you. And if you have a rapport with clients, if you have the ability to, you know, build those relationships, you need to be the one giving it the definition. You don't need to be the one receiving the definition. If that right,
Exactly. Yep.
Yeah. And that's a def that's like beautiful context is this is what I have. This is what I want. Do it right. And it's very very clear. There's no gray area because sometimes you do want gray area where you're like I'm not really sure what I want. Maybe the AI will give me a better solution. But then there are sometimes where you're like I know that for me luxury is elegant white space using it is you know sharp uh crisp lines it is uh shadows it is not being cluttered it's like giving it those things that yeah exactly it's whatever it is for you right and so what I've been kind of telling people especially for the context cards and so
17:42 Business, deal and client context cards
what a context card is is it is the things within your business that never change or are don't change that frequently. Right? So, if you're like, I serve, just as an example, I serve Hendersonville. I do luxury properties that are worth more than a million dollars and I have a strong uh proclivity for uh lakeside housing, whatever it is, whatever the things are that are very specific to you,
put those in one place in a markdown file, and that'll be your business's context card. And I have samples that I can give you all that kind of stuff as well. Um, but what that does is it allows you to move that piece of information without it impacting your business. So let's say that you're in Claude now. ChatGPT releases an amazing model that's better than what Claude is right now and you're like, "Oh, I want to try that." Having your context saved into a markdown file gives you the ability to move as quickly as you want.
Okay? And so you can jump from AI to AI. And it also allows you to if you do the same for your deals. So like if you have one that's uh for a home that is let's say it's a million-dollar house, they've got wood floors, they've got big windows, all that kind of stuff, you would put that in its own context card. You could even like add the the listing link, you could add pictures, you can add all that stuff, put it in the context card. And then let's say you have a client now. You have do the same thing with them. This client loves um vintage furniture. They love um you know lots of windows, lots of light, lots of whatever it is, right? So now you have something that's specific to you, specific to that property, and specific
to that client. So what you can do is you can use them in any combination to create whatever asset you want. So, as an example, let's say you have three clients that all might be interested in this home. What you can do is you can add your context card for your business. You can add the deal context card for that property. And then whichever client you're marketing to. So like if this one likes modern furniture, clean design, da da da, and this one likes gothic, like old, like bring me that thousand-year-old bookshelf type stuff, they're going to see that house in two very different ways as ideal for them, right? Just their natural context is going to be different. And so what you can do is you put those cards in and you say, "Hey,
send create a landing page that is for this home that is specifically for buyer Ted Smith." And it will create the landing page. You could even have it generate images of like their style of furniture in like some of the rooms. You can have it whatever it is that will maximize that context for that person. and the same and then all you would need to do if you wanted to send that same home to a different uh client. All you would do is switch out the context cards and so that different um it would create an entirely different page based on what that client has told you that they want.
So do you do that in Claude?
You can you can have it create the the stuff right there. So, you can say like, "Hey, um, I'm connecting I've connected you to my CRM. Go in, get all the details you can about this person. Get all the details you can about this home, and here's my context card. Use that to create whatever asset, the landing page, an email, uh, whatever it is that's specific to them." Because that's kind of where marketing is heading, right? It's very, very specific now. Because ads. And so when people see something that jumps out specifically to them, where the top of the landing page says, "Hey, Ted, I made this for you in exactly um what I think your style is.
let me know if there's things that you need modified or things that don't resonate with you. And it's one people the time to do this. They x, y, and z." But really, what they don't know is you have their deal card. You have their um their client card. And it's not that hard of a situation anymore to put together these really custom bespoke packages that say, "Hey, I really value your business." more so than just sending them to like a Zillow or an MLS or whatever it is. You're sending them something that's like, "Hey, this is what this house would look like if it was the dream house that you told me you would really like."
Okay.
22:38 Projects and context you can move
And so this is So this is Claude. Yeah. So if you go into projects and Claude
Yeah.
and you go new project. So you could call it context card or you could name it the deal or you could name it the client, whatever it is. And then what you would do is right here's a prompt that you would put in. This would be your system prompt is what they call it.
And so this would be where you would
Um, and then what you add is the context. So this is where you can add from Google Drive, from Notion, from GitHub. You can upload stuff from your device and it stays here in the project. So it it's never starting with a fresh um context window. It's always going to read these things and then create the output. And so that's how you get like highly customized stuff direct without having to go through that same process every time. This is probably the simplest way to do it.
Okay.
And so that's just in Claude. It's under projects and then just create a new project and you have access to all this and then just test it out and then have you can have it. So, this can't edit its own context, but what it can do is it can create the new context cards and then you can replace them out. And so, then you'll have the full asset there.
But that that's kind of how I use um we were talking about just kind of how which one is best. This is kind of how I use all of them. And so, I will try this out. And and the nice thing about it is because you have those contacts, you can try it out in Claude, you can try it out in ChatGPT, you can try it out in Grok and just see which one is giving you the best results. And then if something releases, you can move around. You can kind of do what is needed.
No, it's it's basically the the equivalent of building a new friendship
is like the work that's required to get it to where like, oh, it knows me again, right? And that's with the context cards. That's one of the things that it fights against because all these models know that. Everyone knows that the big like in the tech space like all of these models understand that the way to keep get and keep people is to trap their information in there as best they can and be as useful as possible with it. And so that's how they're operating. Whereas we need to be operating, okay, what is the best intelligence that is currently out there? I want to use that because if we operate like that, it forces them to make better intelligence. if we operate under their paradigm, it allows them to just kind of dictate the pace of
technology effectively. Um, and so by having the ability to move quickly, it it's a massive benefit because a lot of people, especially in the real estate space, tend to be slower adapters than some of the other spaces. And so there's alpha in being able to just move as quickly as you want to.
Okay. Now within Claude there is like a ChatGPT right?
Yeah. Well so it's its own little model. So well I say little but it's one of the major models but yeah so each one so like Grok has its own entity that is basically it's trained on its own data. It has its own compute power. It has its own brain basically. OpenAI has one. Anthropic has one and then Google is the fourth one. And so basically the intelligence is all raising to roughly the same level. Um the intelligence is that you can do is you can basically prompt it. You can ask ChatGPT or Claude to give you a data extraction prompt saying hey I want to get as much of my data out of your system as I possibly can. I need a prompt that will help me extract as much data as I can
reasonably get out of your system without me having to do any work. And it'll create a prompt. And then you go to a fresh window, a new chat window, and then put that in the new chat window. And it will give you I think when I did it, it gave me 10 pages of like who I am within ChatGPT. And so it was good because it allowed me to edit and like get the context really dialed in. And then I took that to Claude. And that is actually so when I did ChatGPT being locked in was actually what made me want the context cards to be able to move around as as much as I reasonably want to.
Okay.
But yeah, this is the projects is probably the easiest way to get what's called persistent context in one place and just use it regularly.
Right.
And if you can do that, it's just it makes your life so much easier as far as like generation of stuff. Yeah. Yeah.
Yep. Exactly.
Context.
Yeah. Exactly. That's exactly right. Yeah. And that's that's the type of stuff. And so you'll want that in like a markdown file or like something as long as you can move it to different models. Like that's the big thing that we want is just the ability to change if we want to. And there have been models that have come out and like people are just like I just like this. And it's like great. If that's what you like and it's effective and it's working for you, there's no reason to switch.
Yeah. Because everybody has a different utility for the AI, right? Like some people want to be computer scientists and do like physics research and you need all of the compute and the horsepower and stuff like that. And some people just want to build email campaigns and it's like you don't need that much as long as you know what you're doing,
right?
And so that's kind of where we're at, especially with the within the real real estate space is if you know what you're doing, you can get away with less intelligence because the value now isn't in what's public. It's in what's in your head that is not public. Because AI, like if it's a public data source, AI can go get it for anybody, right? And so for like I was uh for as an example like if you're doing working the 12 South neighborhood and you're like I know that everybody in this neighborhood wants at least something that's three bedrooms because the two bedrooms are just not set for whatever reason, right?
That's valuable context because that is not stuff that the market just can figure out.
Sure. because you know like you know it you've analyzed the data you have those things and you can then give your AI that context that gives your AI an advantage over the other ones. It's
32:02 Fixing your details across the web
on some of these um programs that you can use. I was just curious because
realtors in general I think we got information all over the place
for sure
is um that
a lot of the stuff that's out there may not be consistent
and so can you ask a particular program to go in find you on the web and make sure that all of the information that is on there is accurate and change it. Can it do that?
So, I don't know that we're at the change it phase yet, but I know it can go do the research and it can go find as much as it can
um and it can make suggestions like if you have context for it. So, if you say like so I would break that project down into like three small projects, right? So, step one is the research, go out and get the information, figure it out, kind of do that. Step two is the analysis of that research. So, okay, you've got the research, now analyze it and compare it with what you know is true and the information that I've provided to see if there are discrepancies. Once that's done, then you need to figure out what the remediation mechanism can be, right?
Is it I need to email this person? I need to go on Google My Business, I need to fix my YouTube channel, my in whatever it is. So, so you're kind of looking at it as like research, analyze, remediate to some degree.
All right. And so, and then once you know what it needs to go do, it can go do all the stuff for you. As long as it's uh what's called authenticated into that app. So, if you're logged into Google My Business, it can go do the Google My Business changes for you. Um, but if you are not logged in, you'll have to log in and then it can go do the changes for you. Um, or if someone in a a third party holds ownership of the site, you can draft the email to send to them to say, "Hey, I need you to change this on your website. It's inaccurate."
Or whatever whatever the remediation method is. It can execute it, but it might not be able to make the changes just depending on the authority required to do so. Okay. I always have a rule in these classes that if you have any questions because I talk fast due to being an introvert and nervous public speaker. So if you ever have any questions feel free to raise your hand and I will just stop talking because sometimes people need time to process. Um and then if I can do it literally anybody can do it. uh with as it relates to AI. Um and this is the part where I tell a story about me being an idiot at some point in my life that shows how anyone could do it. So I literally failed a military intelligence
test. So that's like literally failed that they were like I don't think we can use you. And the reason for that is that I ranked I'll never forget this. I ranked brand new car as like the they wanted you to rank desirable things and they just gave you this list of stuff and they were like rank it. So I ranked brand new car at seven. I was like, "This is not going to go well." And and I saw that as the questions were playing out. But what they didn't say is they didn't give you any additional contact. So when I saw a brand new car, I was like, "I don't need a payment. I don't need the higher insurance. I don't need all the stuff that comes with a brand new car." And so I marked it at
seven.
That was not the right answer as it was revealed to me later. You shouldn't think that deeply about stuff as it relates to military intelligence tests. But I literally failed it to the point where they were like, "I don't know. Your priorities are weird." So they like and that I was not trying to get into the military. Um but
Oh gosh.
Yep. So if I can do it, literally anyone can do it. Um
36:02 Source, destination, state and permission
so these are some of the things um when we're talking about connectors and kind of moving information around, this is some of the stuff that needs spelled out so that it runs really effectively. So source, what is basically what is the context? What are the approved assets that you're providing the AI for it to make whatever it needs to make? Destination, where exactly is this going? Do you want a draft email? Do you want something in your CRM? Do you want an HTML page uploaded to GitHub? Like where are we putting whatever is created? State. So is what is going to come out of this live? Is it a draft? Is it an experiment? Is it a scheduled item that we're going to run later? Like what is the state of it? Permission. What is
allowed now and what still needs a human decision? So what is the AI able to do versus what should a human be managing almost all the time? So the way I look at that is if something is irreversible, I manage it. So, if I'm sending an email, the AI can go as far as to put that email in my draft if I think it's an important email. There are certain emails that are like internal to my team or whatever where they'll email send something an email, my AI will catch it and be like, okay, we'll push this email back out to them. But if it's like a client email or something that I'm like, okay, once this gets sent, the there's no unend. It's it's in their inbox. They will have it forever. I review it. Um,
and then the next action. So, what happens after everything is ready to roll? Once you have this spelled out, you have a really, really clear idea of what the project is and kind of what the win state for the project is, like where are we starting, where are we going?
37:58 What a connector can do
And so, the connector is the bridge. So, the connector is basically it connects to Gmail, Google calendar, all that kind of stuff. And so it's the tool that allows the intelligence to get access to the tool. There's still a lot that goes on like do you want it to read? Do you want it to write? Do you want it to be able to delete from whatever the tool is that it's getting access to? And so if you have let's say a CRM that has a terrible backup system that like this is the only thing that holds your client data. you probably don't want to give it write or delete capabilities, right? You just want it to be able to go in and see the stuff and not edit things. Um, and so that's I always recommend having a backup as it
relates to basically any information that you have. And so the connector is kind of what we talked about. It's a capability ladder. So, it can read, which is pulling information, um, drafting, basically anything that's not permanently impacting whatever it is you're doing. So, that's read. It can create. So, then it can take things and actually create the stuff. So, if you're comfortable with it, it can go and create the entry in the CRM for you or it can create the data or it can create the email or the calendar appointment.
Um, the place. So this is something that gets underrated, but where are you saving stuff? Because one of the things that especially with the volume of data that gets created in today's society is I've created so many things that I don't know where some of them are. And so the tool provides me no value because I didn't have an organizational structure that allowed me to actually use the asset when it was ready. Um, and then confirm. So, you want to have the AI, just like we talked about validation gates, you want to have the AI confirm, hey, this is the real thing.
Everything's done the way it was supposed to be done.
40:01 The connector contract and its fallback
And so, um, if you kind of look at the AI um, integrations as a contract, uh, it goes back to roughly some of the same stuff. We've got source, we've got destination, we've got action, so what action is it taking? Permission. So that is kind of what we talked about just a second ago. Validation, that is what we were talking about as it relates to okay, everything's done. Now we send in another AI to actually validate that what is done meets what it was supposed to be done and then fall back. So if for some reason, let's say the connector disconnects and you it can't get into your Gmail, what's the fallback plan? Is it just going to create that email so that you can go copy and paste it? Are you is it going to give you instructions
on how to reconnect the connector? Like what happens if something goes wrong? Because something goes wrong more frequently than you know
more frequent than I would like it to at this point. If there's one thing you can count on, it's that not everything's going to go right. And so the other thing you can kind of think about is what is connected versus what is manual. So what are the things that I actually need to do with this project? And that's something that is changing on a daily basis. So connected basically means all of the connectors. The AI has every single tool it needs to do the entire job.
Manual means that there's going to be some parts of that job that requires human input, whether it's you or someone else, to get the job from the start state to, hey, we have a a finish. It's a full win. Um, and so that's just that terminology. Usually the finish line is going to be roughly the same. Um, and then the big thing that you'll get is fake completions. And that's why the validation is so important because if the AI is fully connected, it it sometimes will just blatantly lie. Um,
So this is a prompt. Um one of the things that I always tell people is uh content creators what they love to do is they love to put things on slides just like this so that it's you can't get it. Right.
Right. If you take a picture of it with your phone and then tell ChatGPT or Claude to transcribe that picture, it will give you this entire prompt. So, anytime you're watching like a YouTube video and you're like, "Oh, that'd be really cool to have." Take a picture of it, take it to ChatGPT. Upload that picture and say, "Hey, transcribe this." Now, you have it.
That's one of the like the little tips that people are usually um like, "Oh, I didn't know I could do that."
Yeah. Uh, it's it's very it's it's more useful than I thought it would be cuz all the time I'm taking screenshots and and sending it.
43:53 Handing work from AI to AI
So, this is um there's a couple communication styles that I think are going to really like dominate the world. I think communication is one of the meta skills that is required to really do well. Um, and I think that is AI to AI, human to AI, and then any permutation thereof. So what you're going to see is you're going to see humans start the communication chain and it's going to go human to AI to AI to AI to human to AI to AI to human. And so there's going to be all these different links of a and so understanding how the AIs communicate is a valuable tool because we kind of get how people communicate
and like if you think of like the best communicator you know they would probably do really well with AI because they probably have clear definitions of expectations. They probably have um an idea of what the constraints are. They probably have an ability to break the idea down so that it's able to be understood by everyone on the team. And that could mean communicating it in different ways to each person that needs communicated with, right? Because it's going to be different to a lender, different to a client, different to a broker, different like every person in that chain uses a different vocabulary.
And so AI-to-AI handoffs are one of the things that are going to be really really prevalent as AI becomes more powerful because we're going to be giving an AI a task and it's going to create an output and then it's going to give that output to another AI to create a different output. And so right now we're seeing this in the tech space where humans are starting the coding loop and then it's getting passed to AI to AI to for hours it's just going AI to AI to AI and it's like the old plane analogy where if you start a plane not directly at where it's supposed to be started at like a degree off or whatever and it's LA to New York you end up in like Kuwait or whatever it is right and so it's the
same thing if you're not managing the AI eye context and the handoffs well you end up in just a different place because the context is not where it was supposed to be and so like and this could be AI to human as well so you're identifying the payload so what exact files fields and instructions come out of the AI first so what is it actually trying to give you um the destination where is it sending it to is it sending it to another tool to another agent to you to review where's it one and then permission and validation exact same like does the AI have the permission to do the things it needs to do and then is it validating that it's actually doing what you asked it to do and obeying those rules
46:43 What goes wrong with connected AI
here are the big things that go wrong with the AI so invented facts um and you can definitely impact these with uh different prompting different context. So like as an example, if you ask for citations and links, it's highly less likely that you'll get invented facts. You still might, but it's less likely because you're making it give you a data source.
Wrong destination. So again, sometimes you'll tell it to put work in your CRM and it'll be like, oh, is she using uh Google Sheets as her CRM? Great. I'll put it there. and it's like, "No, I need to tell you that I am using Salesforce as my CRM or I'm using whatever as my CRM." Um, too much permission uh is something that can end up with your files being deleted, right? So, like if you have if you've given it too much permission to just go in and modify and like have a good time, it can be detrimental to like your real stuff. Um, silent sending. So, that is something that is basically it it sends it instead of creating the draft. Um, so it takes it a step too far.
So you just want to have your guardrails, this is where guardrails and validation fix a lot of this. Um, so like you can have a duplicate record. You can have a step in your workflow say check the CRM for all records containing this person's name, this person's da da, right? Like you can have it check for various things and then it'll say, oh,
Um, and then missing confirmation. So, sometimes it just doesn't send the thing that says, "Hey, I did what you asked me to do." Um, this is a prompt.
48:34 Recap: small tasks, clear rules
And then, yeah, just kind of going back over things one more time. It's literally what's the source info, where are we going, what state is it in, what permission is required, and then what's the next action after it's done. A lot of people break things down into too big of a job, right? So, what what I tell people is the thing that isn't getting replaced, it's not jobs that are getting replaced, it's tasks, individual tasks. So individual segments of people's job are what is getting replaced. If your job consists of a bundle of tasks that are low-level administrative work that requires no expertise, you will be replaced.
No one in real estate fits that bill because it all requires high-level relationship building expertise.
Right.
Right. And so what you're going to see is and My opinion is that AI tends to kick mediocrity in the teeth. So, if you're just average and you're just like, "Hey, I'm here to do adult babysitting and push a couple buttons on a computer," you're likely to be replaced uh quickly. That is not I don't think that that is going to be as widespread and prevalent as the fear-mongering craz are making it out to be. I think we will see restructuring of jobs, but I think as of right now, it's literally literally replacing task by task. And so if you can figure out like, okay, AI, if I use it to replace task by task, what else can I focus on,
that's where you need to be living is it's the same thing that the accountants went through when the calculator was invented, book publishers went through it, when the printing press was invented. Like if you look through history, while none of them have been this explosive, there have been technology revolutions where you look at and you're like, "Oh man, this is going to kill everybody's job." And it never has because people are inventive enough to create one new jobs but two new problems. So,
you know it's it's like all right like just it's it's a new economy new skills and like this is this is kind of where we're heading and It's going to be unfortunate for some people, but it's it's always unfortunate for some people and it's always great for other people. If you have a good mindset about it, it's going to be great. If you don't, it's going to kick you in the teeth.
And so, like, I'm I'm just of the valition that if we keep looking for what's the highest value problem we can solve, we solve the underling problems and we just keep focusing on that, you will always have some type of employment. uh it just may not be the same look exactly the same as it does right now. And so the idea that we're training our whole life for one job is like awesome for the people it happens for, but it should be the idea that we're basically training to be really good at change
and adapt and adapting to things, right?
