navarre.aiClarisTalk AI · the FileMaker AI podcast
Context Engineering
What's buzzing at the AI conferences Cris is attending? Context Engineering. Matt and Cris catch up after a while, and talk about recent FileMaker conferences (Matt) and AI conferences (Cris) Early on we had a podcast about Prompt Engineering – which is but a small part of Context Engineering. Follow Cris: https://www.youtube.com/@iSolutionsAI https://www.linkedin.com/in/crisippolite/ https://twit
Apple PodcastsSpotifyYouTubeAll episodes
Transcript
Transcribed by machine (whisper, large model), 13,181 words. Names and product terms are sometimes misheard; the audio is the record.
Claris Talk AI. Hey, I'm Chris Ippolite. I'm Matt Navar. We got some stuff on the plate today. So conferences, I've been to some FileMaker ones, and you've been to some AI ones, and you win this round. Well, shoot, let's catch up. I mean, we haven't done a show in at least, I mean, probably six conferences combined, I bet. Yeah, fortunately, there's been nothing happening in the world of AI. So it took a nice little pause, like everyone was hoping. So I might win in terms of the fact that my conferences were in Naples and Antwerp, two very cool places. Nice, yeah. Yours were probably in, what, Las Vegas and Chicago. Well, one was New York City. That was kind of cool.
Oh, that's cool. And then the rest are just, you know, everything's in the Bay Area, basically. So I went to the FileMaker Week conference in October in Naples. Hadn't been to Naples before. Oh, my God, lovely city. And I... Taught there, I did some work on kind of two areas. All the stuff I've done have gone really well. So I'll tell you two really big highlights from that one. First of all, I kind of introduced these topics of starting really small with AI and something I call buttonize it. And I've got a whole bunch of things that are really, really small, like something you can implement in an hour or two. And what I'm really hoping to do, we'll make plugs later, is help a lot of companies who are being told by their boss they need to put AI in.
And they have no idea. They don't know where to start. So basically... Which is a lot of companies, actually, right now. Call me. We'll spend two one-and-a-half-hour meetings. And by the end of it, you'll have something small integrated. Love it. Or a proof of concept, like loud lines. And, yeah, anyway, that's kind of my pitch. Basically, experiments. Like, we call them experiments, kind of a holdover from machine learning. You know, it's not... Maybe. It's a little bit proof of concept. But also, you're like, hey, is this thing I want to do going to work in this crazy black box of language models, you know? Some of it could be in production. Here's a good example.
Well, I mean, yeah. Look at us go off track. Precursor to production. No. No. Production. Yes. Production. Well, did you see? There was a paper that came out, actually, since we've done the pod. I don't know. Maybe it was. It was the State of AI report that comes out every year. And the headline grabber was, it was like, 63% of experiments fail. Or, no, 63% of projects fail or something like that, right? Pilots, they called it. Just what it was. Which was... And then, and I was like, oh, dear. Like, everything's over. Maybe AI is done now. And so, then I read. And, like, literally, the next paragraph and then the following five pages were about, like, well, they fail if you don't use these methodologies.
And when you do, like, 63% of them proceed into production or something. It's like, wow. Okay, that's an important piece of information. I'm away at the other end. Here's my little pitch, right? That this can be done in a couple of hours. Production and working. So, let's say you have a system, which is super common. Where, as part of the process, you have an email sent to you with a bunch of information in it. Case number, person responsible, city the event is happening in, dates, something like that. It's all just jumbled in the email and who knows where. You have a database that you have been copying and pasting that notes field in. But you also have case number, person responsible.
You have these six or seven fields. It's crazy easy, knowing what I know now, having touched on it. A bunch of times. Whatever. To have a script that basically takes a look at the text coming in, uses a quick little thing that says, where's the case number? Where's the person responsible? Where's this? And just populates the fields. Like, if it's got really high confidence on it and those fields are empty, it's saving you a bunch of click and drag. When you're doing JSON parsing, basically, you're converting the output. Or the output is JSON format and then you're using FileMaker's native parsing tools to put it into fields. No. Individual. No. Prompts per each field.
I guess I have to think about it. Yeah. I think I probably would do it as Jason. Because if you do JSON prompt and you get back, you know, give me the case number, give me the blah, blah, blah. Structure. Yeah. Why did I say no? It's the obvious way to do it. You love JSON. I was trying to tee you up. Yeah, I do. I do. Thank you. Of course. I whiffled that one. Sports. Yeah. And then the fields are populated, right? Simple model. No overthinking. Really simple. Well, I think what's really important about what you're pointing out there is that, you know, it doesn't have to be AI. This is an outcome that you're talking about, right? Like, we're, I think, maybe even Claris would admit this right now, that their focus has been really, like, hyper-focused on the AI stuff, like the terms.
Semantic search was just, like, pounded into the community for ages. But nobody ever talked about, like, what can I do with this? Or what business problems can I truly solve? And what you're talking about is, let me actually map to something that you can do, and we can concentrate on outcomes. That's fair. Because I think you can add. I think you can actually get through a conversation about AI without mentioning anything about AI. Not that you're trying to hide it, but you just don't need to have a conversation about it. Just, like, what do you want to accomplish? And you're doing the perfect thing. You're saying, because really, the best way to introduce AI into your organization is in those small, rote tasks that you're talking about.
Like, get rid of redundancy, reduce error. You know, I heard one of these conferences I was at, someone said this thing. They go, hey, AI isn't going to come from your job, only the part of your job that you hate. And I know that's a little convenient, but I think. It actually speaks to what you're saying. Like, hey, there's these things that just drive me nuts where I'm just copying and pasting all the time, and all I'm doing is introducing risk of error. Why don't I get rid of those and automate those? And it just happens to do it with AI. Yep. So, anyway, I'm thinking I've got about 20, 25 of these really little things, design patterns, whatever you want to call it, and implement.
So, the conference I did there, I showed 10 of them. Nice. And I taught a bunch more at another conference. So, the second thing that was a great takeaway for me at the conference in Naples was the Naples. Apple Developer Academy is there. Oh, yeah. I've heard a lot about that. That's really cool. I was in a room with over 10 FileMaker developers, all under the age of 30. What? Let that sink in for a minute. FileMaker Pro? FileMaker Pro. Current version. Were they looking for a different room or something? No. No? No. So, these are people who were. Oh, Wi-Fi. You had free Wi-Fi there. Oh, come on. No. These are just. This is real. This was exciting, right? So, there's a bunch of people who graduated from year two of the Apple Developer Academy where they get a lot of FileMaker.
Yeah, it's a really cool program. Several of them actually joined a company, right? So, they actually started together. Now, they're actually doing FileMaker for a living. Oh, I heard about those guys. A few of those people were there. Yeah. The Italian kids, if I'm not mistaken. Well, not all of these are because they come from different places. But to join the academy, you have to live in the city. You have to move there from wherever. And so, it was people from actually all over the world who were there. So, and a lot of other people in the room were graduates with serious FileMaker knowledge. These are people who are doing really, you know, real professional FileMaker work who didn't yet have jobs in FileMaker and kind of weren't sure what they were going to do.
Amazing. They sort of realized it's not going to be easy for them to make money as a Swift developer, which is what the main skill that Apple Developer Academy teaches. That was primarily what the Apple Academy is about. It was the native programming. Right, right, right. Oh, is that a depressed market space? To me, that was a huge takeaway. Interesting. Wow. Because Claris definitely is putting emphasis into the other locations of this and to really try to step up the training of these people. I'm really hoping to kind of have a little bit of a role training the trainers and being involved in helping it because it's just so exciting to see, you know, one of the plans that Claris has, the community has, of bringing new people in and who actually have enough training.
I'm really hoping that Claris will be able to do a lot of the work that they've been doing and experience and skill to be valuable as employees of partners and as internal developers. Well, as you know, you know, I was very near and dear to my heart. I did many years of the 42U school training that then eventually led into Quasar. And part of that, we actually did two pilots, one with actually two pilots with the Naples Apple Developer Academy. This is many years ago and they're still kind of testing it out. I met a lot of really cool kids out there. I did some mentoring stuff. It was a lot of fun for me. And that energy there is really amazing. And since then, the Detroit location has opened up here in the United States.
I've been sort of giddy to try to get out there and check it out myself. But yeah, really, like, really happy and proud of all the work that Claris has been doing in that area to really, I mean, I love seeing all the kids that I got to teach in the 42U school. I see them now working at partners or speaking at conferences. It's just awesome to see. How many people came through that? How many? God, I think at one point we counted it up. It was, you know, it was all cohorts of like 10 kids. I think we, I think I probably did like 200 over all those years. I mean, we did a good seven years or something where we were doing two or three of them a year for a while.
So it was a lot of fun. And, you know, the community, that's what we need. We need the good lifeblood in the community. So I'm really happy to hear that and glad that was such a focus of, I mean, it makes sense that Academy is right in Naples. Did Quasar just kind of roll up? What's, is that still a thing? Yeah. Uh, Quasar's not a thing. I mean, the school itself is not a thing. So the program as a result ended, but, um, but that was a really cool experience and I'm still in touch with, you know, a lot of the people that I met there and trained through there. So very proud of them. Cool. All right. And so what about the AI stuff at the Naples or the other conferences actually?
Well, so, I mean, like everything's AI, it's just kind of all across the map. Some great speakers talking about all kinds of things. And some of them were, were at both. I actually saw the. The thing on LinkedIn. Yeah. The EngageU one. I was just gonna, there you go. As a transition, I was looking on LinkedIn and it said, it was like, here's the, so it's amazing. Like what the first Austin Engage was, it was like two years ago, like, right. A handful of months and two years ago, there was only two, like two AI sessions. And now, and then I look at the EngageU and it literally was like 20 AI things. And then there's a little note. It goes like, and then there were also some things that weren't AI, like three things or whatever.
So it's really amazing. It's amazing to see how much AI has interjected itself. What were some of the highlights that you took away from, uh, EngageU? Um, let's see. Well, I taught a half day class before it started surprisingly on AI, uh, the same, the same real world implementation type stuff. I showed those perfect. I showed a lot of those features and I showed exactly how to implement them and then left people with a demo file with my API key in it for a number of days, trusting that they wouldn't spend all my money. Dude. Hey, by the way, we got to talk about, uh, true. And token, man, at some point today. Oh, yeah. Yeah. Uh, that's a whole, that's a whole episode now onto itself.
So gosh, uh, let's see. So what are the takeaways of AI stuff that there was so much, um, uh, always amazing. Oh yeah. It has amazing stuff. Cause he started on this so early they did. There was a lot of MCP. Um, yes. What was your, what were your thoughts on, uh, MCP? Well, I was last to the game of the beta program. So I've been looking at it and learning about it, but I didn't know there was actually an official thing. There's a way that you can enter the Claris MCP beta program because they have a product that's pretty cool. Like really cool. I'm a big fan actually of the concept in general and, you know, the direction that they're going with it. Well, that's actually, we'll have another podcast just about that.
We should absolutely talk MCP specifically, you know, who I'd love to get on the podcast to talk about that is, uh, Louie. Um, oh yeah, my, my buddy, Louie de la Para, he's the primary architect over at Claris on all the MCP stuff. Let's do it. Let's, uh, yeah, yeah, yeah. He was there talking to good stuff about it. So let's do a con, but for those that aren't familiar, um, let's define MCP. First of all, I think the most important way to define MCP is to define the P which is a protocol. So, um, it's a model context protocol, so it's a protocol, uh, that allows you to be able to pull. Okay. So it's a protocol context, which is both, uh, documents or, or information data, for example, let's say, or actions, which is a lot of people aren't paying a lot of attention to the action part it exposes, which is huge.
And then also potentially resources. I happen to think that leveraging, um, prompts like system prompts attached to your data sources and actions that describe them for LLMs makes MCP even more powerful because the, the, the C part is context, which is a little bit of a tease for what I'll be sharing about what I've seen in a bunch of conferences. But the context part is I want to go hit this data source. Let's say it's FileMaker. So we want to, let's say we're using Claude. That is a MCP client, right? And then what we need from the client is to be able to connect to an MCP server. So you can go spin up and configure your own MCP server. But what Claris did is they, uh, allowed, they have sort of a conduit for connect, you know, for creating your MCP servers.
And they also then did the engineering on the backend. Yeah. On the Claris side for actually having FileMaker server hosted applications, be able to be configured for MCP servers. So long story short is you can expose your FileMaker data and actually even scripts. That's the part that I think is so cool. Um, in, uh, in clients right now, I think probably the best one is, is Claude. Um, frankly, it's a little clunky, you know, as far as the Claude part of it, not the FileMaker MCP part of it, but this is exciting. This, this, this, uh, explains. This exposes safely and within the same bounds of all your, your permissions, uh, in user management, the same, uh, data that you actually have within your FileMaker systems and scripts and automation.
So tiny, tiny hello world summary, right? That's what you can do in like the first hour or whatever. So it's a website that you log onto. That's part of Claris. Yes. It's part of the platform. It can see and connect to FileMaker databases hosted on a FileMaker server. Kind of like Claris Connect can. Yes. You know, so you go in and actually you put in the credentials, you turn on. It's not too different by the way from Claris Connect for those that understand that it's kind of like the old paradigm for, well, the, the Connect paradigm was exposing actions and potentially data through connectors, but it was super proprietary to Claris. Yeah. So this is what, this is an industry wide thing.
And actually, hold on one sec. Actually, we missed an important part. Why is MCP such a big deal? Well, about a year ago, Anthropic dropped one of their blog papers. By the way, when Anthropic drops a blog paper. You can drop everything yourself and go read it back then. They dropped MCP. And essentially what they said was, look, we're seeing that people want to connect to data sources and, and expose tools. We're going to introduce our own protocol here at Anthropic and they offered up MCP protocol. Then everybody started offering the same. They're going to do their own protocols. Right. Uh, OpenAI was doing it. Google was working on one. What, what actually made MCP the, the sort of avalanche or the tsunami, so to speak happened was when we were all at Engage, right in the middle of that.
Uh, OpenAI dropped, they go, they said, Hey, we're not going to work on our own protocol anymore. We're just going to use the Anthropic one, MCP. Then all of a sudden, within a matter of like a week, everybody in the industry is like, we're all using MCP. So that's why it's such a big deal. It's not that it's perfect. It's still evolving, but it's the thing that everybody's doing. So that's why Claris jumped on that. All right. Before we get to the conferences that I want to talk about yours, I'll finish like the little quick tour of how, of, of how, how you can start with there. Right. So. In the website, you configure it, you list the specific FileMaker databases and you list the data tables and fields and scripts that you want to expose.
So you just list them. Cause you don't want to expose all of them. Yeah. And then you get a link, right? You definitely don't. Then you get a link and you can also save for each one, read, write, delete, et cetera. Yeah. Um, you go into Claude desktop and you can go into configure MCP. You copy and paste the URL in, and now in Claude on your desktop, you can say, uh, how many. Invoices that I have in July and it queries your FileMaker database and tells you the answer. Well, I'm potentially more stuff to do. Well, but, but at the end of the day, that's the outcome. And we just talked before about how we shouldn't focus so much on the tech, but what you can do with it.
And that's what you can do with it. And now in the absence of, I mean, to be totally honest, I think where everybody in the community is at, whether there's some that are a little bit more advanced than others, everybody kind of wants to chat against their data. Like some have gone and created their own tools. We pretty much. Build custom agents for all of our clients to be able to do that, but everybody understands that when they, like my dad, who's 80 understands what chatting over data is right. Like people get that concept, but we don't have that native function in FileMaker. Sure. We got script stabs, create all these things. It's fine. Now you actually can use the paradigms that are well, frankly, you know, chat GBT has this, uh, has MCP connectors.
Claude does as well, too. Those are the most used applications by far in the world. Even more than Instagram and all that silly stuff. Yeah. So now we can actually use those common paradigms to be able to communicate with our FileMaker data in the way that everybody else is doing. I'm going to push back a little bit about chat interface with your database, uh, way back in the day, we had a DOS prompt and you had a keyboard and you could start typing commands. Hey, all those kids that were in your twenties that were in Naples, tune out because we're going to lose them. Yeah. Yeah. I'm not going back to typewriters. Basically what I'm going back to is, you know, way back before graphical design, you had the ability to just only type commands and, and get data and write data to a system.
And so your only interface was a keyboard and it was nowhere near as smart as that we have now, but now I have a keyboard and I can type questions to my database and that's going to be really cool for about five minutes. And then, and then. I got work to do. I got to find a list of contacts. I got to make a record over here. I got to send an email to someone. I got to, you know, figure out some plans and get some data from here. I got to use my database in normal ways and with normal layouts and with normal fines and with normal things and normal activities that haven't changed because AI is here. And I'm not going to go automate those by typing into a prompt, no matter how smart that prompt is.
So wait, you're, are you lobbying? Against the idea of replacing your entire system with the chat, or you're just saying chat doesn't make sense as a utility. I'm trying to figure out what I'm saying. Like to be able to have a conversation with your database and ask questions over your data, or even give commands, like query, query it workflows. How often do you query my data, your file maker databases that you use to run your life in business? How often am I performing fines? That's essentially what you're saying. No. How often? How often are you, how often are you performing a fine to thus then work with the found set and do something useful? I think a lot, actually.
I think that's probably a pretty common paradigm inside of file maker usage. Yes. And how would you rather type a sentence or paragraph to do that or three or four words and have what you want? You know, I'm pushing back in a little bit because I just don't think position. I think when we get to a place where, um, where I can have. Way more power than just being able to text with it. So basically I can say, Hey, today I'm working on follow-ups for all of my new contacts who have come to my classes, create an interface and find all those people, uh, give me recommendations of the best way to communicate with these people with whatever, uh, and show me what you got.
And then I, and I see stuff and then I say, yeah, that's great use for a chat interface. Yeah, it does. Oh, that's what you're saying. You know how hard that would be to do right now? MCP or otherwise? Uh, not that hard. Is that a real, I don't know if you're actually asking me that question for real or not, or if that was hypothetical, because that's not hard. Not that hard. 20 hours, 40. Well, ironically, you can leverage AI to actually make that, to increase the, the, to get from idea to actual implementation there. So there's a lot of AI benefits in that. Yeah, I get it. But, but, but, but what I'm saying is like, the first thing I said was. Create an interface of that, and that doesn't exist right now with, well, you can't, oh, we're talking about file maker.
Okay. Yeah. I'm talking about just AI. And by the way, one of the emerging standards right now, probably the hottest topic right now is actually dynamic UI. Uh, ironically, actually one of them is a platform that actually emerged from through MCP for UI as well, too. So you're onto it. You're that itch that you're, that you want to scratch there is exactly where people like where the heartbeat is in a lot of this. Technology, but in file maker, no, dude, you can't do any of that stuff in file maker. Yeah. I a hundred percent agree with you. No, I can't, but I can run my business on it just perfectly well. And my clients can run their business on it just perfectly well.
And there's aspects of it that are hugely ripe for AI being added to it. And some of them are big and some of them are small. Yeah. I would go back to what I was saying earlier, where I said like the, the, the biggest impact you can have, especially when you're getting started with AI is in just little things. Like we don't even need to talk about AI, just like little tiny things you can do like this. Stuff that you're talking about in your classes, that's where you can make the biggest impact. And eventually the big stuff will follow. But 1.2 billion active monthly users worldwide are using chat to actually communicate with data. And I thought you were going to say Facebook now that's like 5 billion worldwide.
So it is a, it is a very it is a very profoundly here and now, and not going anywhere interface with data. So definitely for us to. Ignore it completely would seem very irresponsible to our clients that are relying on us to stay ahead of the times when it comes to, uh, new things in technology. So I think, you know, I, you and I both use AI all day, every day. Right. Yeah. Um, and I use it for everything from researching appliances. We're buying for the house we're building to learning Greek, uh, to God, a hundred other things, you know, medical stuff, helping with the dogs, barking their stupid head off right now. Like everything. None of which connects to my life as a filmmaker, developer, and helping me do the things I want to do, or the things that I'm helping their customers want to do.
It's all much more unstructured and much less connected with kind of wrote business where I have a whole bunch of records that I have to do something with and need to track long history and whatever. Yeah. Well, if the first thing is to support your point is crud operations are not common in. Chat interfaces, right? Chat is simply a query mechanism and there's a lot of querying happening. There's people at certain altitudes within certain organizations that all they do is ask questions of data. Like they don't, they're not fingers on the keyboard entering data. When we talk about data capture, that's just almost non-existent in these chat paradigms. Like you can tell it, go, you know, go create a record or whatever, but that's not as efficient as just simply doing it in a workflow.
All these things are true. So within FileMaker, yes, we're still, I mean. Workflows aren't going away. Like people still need to manage data, collect data. All that is all very true. There are organizations, there are roles within organizations. There are, you know, workflows where just trying to get information and insights from your data is critically important to the task at hand. So in those cases, I think people are very much leveraging and becoming aware of the power of integrating AI along with their data. As a matter of fact, I, I, to me, creating, managing your own truth. To me is the most important thing that an organization can do, even in the AI era.
And that involves creating it and capturing it and, you know, uh, keeping it, you know, uh, maintaining it, let's say, and that's really where an application platform like FileMaker comes into play. So, um, in that regard, I'm not trying to say that integrating AI and FileMaker is not useful. I think it's hugely useful. I'm just trying to say that I don't have any utility in having a chat conversation with any of my databases and I'm. I'm having a really hard time thinking about any of my customers, having a chat conversation with their databases to actually do their job. You're like, well, I mean, if you're a project manager and you did have a database with 400 and field different areas and all this other stuff and quotes with child records and this and project managers and POs and all that stuff, being able to say what's going on with this project.
It doesn't, that's great for five minutes, but it doesn't help me actually do my job. Oh, wow. Well, some people, that is their whole job actually. So, um, wow. Okay. Well, this is a really weird conversation because literally all of my clients have that in common that they want to interact with their actual proprietor data. Well, I mean, I'm talking about FileMaker people. I'm talking about, you know, the people that I've been working with for this whole life, I'm thinking about a whole bunch of really specific clients I've worked with in the last 10 years, all the way from the public health where there's, you know, some great utility of it for analysis.
And they're, they're college workers that have really important stuff to do, right? Like, like, I don't know, really good, a good example, an epidemiologist in the, in some program in public health has a bunch of information. You're trying to convince me that people don't need AI? No, I'm not, I'm not, I'm not, I'm not, I'm trying to convince you that there's aspects of AI that are massively useful to a lot of FileMaker people, but that they don't want the they don't want to prompt that they have to type into to do any of the stuff that they have to do that has utility for a little bit of use, but a chat interface with the FileMaker database right now with the camera just gets you about.
This is great. It's great. No, I've just put this on record, dude. This is awesome. I love that you're saying this and it's being recorded and we're going to share this. I know it's going to get where I want it to go. It's just not there yet. It's just not there. Okay. Yeah, no. FileMaker is definitely not there. It's not a FileMaker thing, it has nothing to do with FileMaker. All right, you can hold that position. I cannot agree with you, dude. Like, I am swimming in signals and I'm involved in a lot of active projects. But you're management, people, sure. If you're, if you're a... I don't know, Matt. This is a tough one. Why we look, I look forward to the comments of people talking about how AI is not useful.
No, it's not what I'm saying. If you're a project manager. You get a bunch of information in, you have hundreds of details to manage. You have to track all the people who are working on it. You have to, you have to send emails to the client. You have to manage everything. You have to make sure all the numbers are correct. You have to audit it all. You have to connect it to payroll. You have to get all those numbers right. You have to run certain reports to understand it. Yep. And all of that is communication with people and running around the building. I use AI for all those things as a business owner myself. Right. Everything you just said I do in my job is a guy who runs a couple of businesses and I use AI for every aspect.
That's right. But you don't use it for everything. I'm not chatting with it. You use the freaking chat interface with it. That's my point. In some cases I do. Yeah, for the reporting aspects you do. Matt, I don't know. I don't know. I mean, do you actually, do you have a chat interface that says contact all of the people out in the shop and find out if there's any problem with their hours? I, I already said I don't perform actions through chat. I don't perform crud through chat, but there's so many other things that I actually do. I don't know. I don't know what to tell you about. Well, you're at a different level, right? Maybe I'm just working with people. Uh, I'm just talking about as a business owner, I'm not even talking about as an AI guy.
I'm just talking about how I leverage the tools. Yeah. And I'm not talking about the business owner. I'm talking about the 400 people who work for him. I don't give a crap about the business owner. I care about, I don't know how to defend the, I don't know what you want me to do here, Matt. Like this is probably really interesting podcast, but you're, so this is the Claris AI podcast and you're saying AI is not useful. So I'm not even sure what you want me to do here. You're so not understanding me. I'm not. And I hope. The listeners, the three listeners are. I'm saying that the way that I use the chat GPT interface by typing and asking questions and getting answers and doing research and understanding stuff and having individual chats is not even related to the clients that I have who are working as a project manager who have to do all the things that they're talking about.
Yeah. I mean, I think there's some in the field who are saying, oh yeah, yeah. We're going to replace all of that with a chat interface. I don't agree with that type commands. Yeah. I don't agree with that at all. Okay. If people are saying that, I don't agree with those people. Well, there you go. Now we're finally on the same page. Yeah. I mean, I, I, I, you're right. Those people are wrong. And what I was trying to say is what, when we get to the point where I can say, as a project manager who has to do all that stuff, I can say, create an interface. These are the tasks I have to do today. And it goes and gets all your file maker data and it puts all that to get file maker data and puts it all together.
Connects it, has all the security layers, has all the people, you know, the guys. The guys in the shop, they just write down their stuff on I, on pieces of paper, right. I gotta, I gotta read it, audit it, get it in, put in all the exceptions. Um, yeah, there's that manual, there's people who have to type stuff in, right. Those people aren't using chat because they're working with, in this case, like making stainless steel things or whatever. Right. Anyway, um, all of this is a long aside because I really want to get to, let me, let me bring it back to MCP just to be clear. MCP, isn't about just chatting with your data. Right. One of the ways that you.
Can interact with MCP services through MCP clients right now, the most popular, because they literally have taken over the world from a total addressable market standpoint. There's nothing bigger right now than people who are chatting using these IDs. I'm sure they'll awaken one day to how useless and futile that those efforts are, but in the meantime, they are doing that. So a lot of people talk about MCP as just chatting over it. That's not it. Like if you really dig into MCP, it's like saying the only thing I can. Do. Do with API connection, you know, with API endpoints is like chat over them. No, you can do so much stuff. You can expose things. You can, you can have a entire, like, um, a cat platforms exposed to be, to integrations, just in the same way that you've integrated API.
So really, if, if you believe that you don't need chat, don't turn your back on MCP. That's that you can completely live. The C doesn't stand for chat and MCP. You can completely utilize MCP without ever doing chats. Just. So everybody knows that, that we're talking about, and we can dig more into that in a future, uh, episode. We don't want to just lose everybody on the concept because they don't want to chat with stuff. It really has very little to do with chatting. I got two things left on the agenda. Uh, you can take them, whatever order you want, uh, trillion tokens and getting listed on the open AI website. I got a, I got to know more. And then the conferences you've been going to in the AI world, which are completely different.
I see a little open AI dev 2025. Yeah. That was just one. Yeah. Um, with cute little ASCII, look at the little fun ASCII things they were doing. That was like the theme of their conference. Um, yeah, so, uh, well, one of the, yeah, one of the conferences was opening. I dev day. That was a, that was a real fun one to be at in person. I've watched those online, uh, the past couple of years and, uh, and got invited to come check it out. And, uh, it was awesome. What a cool energy. There's a lot of really, you know, it was really like a product release. Uh, conference, but I got to say like, just, you know, pound for pound. That was the coolest conference I've ever been to.
Just like every little, I mean, it was really a company that has way too much money to spend that put on a conference and, um, they did a great job with it. It was really cool. It was kind of a small group though. Um, like 1500 people, not because there wasn't big demand, but I think that's just what they wanted to do. And again, it's kind of a product drop, uh, thing. I learned a lot of really interesting things about their products that they're using. A lot of it was the stuff that really hit me hard was, um. They were releasing a lot of codex tooling, so sort of their answer to Claude code and, and there's some really cool stuff that they released into market over there.
So those that are actually doing coding work, it was really, um, really cool to see all the products out there. Then they did some, like some of the stuff that we're seeing in ChatGPT now, like, I don't know if you've done any of your holiday shopping and ChatGPT, but you'll notice like a little app pops up that helps you do shopping, which is really cool. Um, and then all these other connectors that are integrated, like if you say, tell me about. Homes in my area in Greece, like it'll pop up a Zillow or whatever they use in Greece, you know, or I don't, I don't know if it's available in Greece actually. Um, but, uh, but they're, they were, you know, profiling all that.
It was really cool. That was a very cool conference. So really there wasn't a lot of like general, you know, AI industry chatter at that one because it was all just about their products basically. Right. Um, but then I went to a couple others to the AI conference in San Francisco. It's kind of unsexy name. It's just called the AI conference. Yeah. And, uh, and that was definitely, you know, what's going on in the, you know, in the AI community. What is everybody talking about? It's really neat to see, you know, what some of the common ground is, what some of the topics are that you hear a lot about. Um, and then another one was, um, AI Dev Day. One of the conferences I really enjoyed earlier in the year was AI Dev Day.
That's put on by, uh, Deep Learning AI, Andrew Ng. Open AI Dev Day and AI Dev Day. Yeah. It's weird. Yeah. It, and, uh, Open AI's Dev Day was kind of like DevCon is for us. It was like, hey, developers, you know, it wasn't customers facing. It was like, hey, developers, come check it out. And this, the AI Dev Day is, uh, Andrew Ng, that's NG. He's like a luminary in the AI space. Like he's sort of the, you know, like he, he's at this conference every two feet, people are taking pictures and selfies with them and everything. He's a big deal. And he's one of those, when he talks, people listen. So he has his own Dev Day conference. He's focused on developers.
And he's a big deal for developers in general, because he's got this Deep Learning AI, which is sort of like the lynda.com of AI training. And by the way, a lot of really free, good free stuff, if people want to check it out. Um, and he put on a conference in San Francisco, which I attended, which was great. I thought it was a really good one. That's where I learned a lot about, um, some of the memory usage stuff that was really early before it had come out, which I thought was really cool. Uh, and then he did one in New York City, which was just like a couple of weeks ago. And. It was all new topics, huge attendance, a lot of real big industry players there, and some kind of commonalities between his, the Andrew Ng AI Dev Day and the AI conference.
There were, you can kind of see what, you know, some of the stuff is that a lot of people are talking about out there right now. So, um, that was a, that was very interesting. Um, and that, yeah, that's just a couple of weeks ago too. So I, I think if I was going to share with you what the most common topic, thousands of people, like how big are these? Yeah, like those were maybe, yeah, maybe, well, the AI Dev, the one, the, not the AI Dev Day, but the other one, the, the AI conference, just way too many people. It was just packed. It was like, like one of those peers in San Francisco and it was just elbows to elbows and like, there was a lot of people.
So, um, so yeah, no, really great energy, a lot of popular, very popular. I think the AI Dev Day is just kind of sold out after a while because of the venue restrictions. And so, um, yeah, it's hugely popular, just like, you know, tons of just really interesting energy there. Um, but the, the big takeaway from all the conferences and actually, which was sort of, um, sort of, sort of propped up by some papers that dropped recently as well too, was this idea of context engineering. Have you heard a little bit about context engineering? We've, we've talked about it a little bit and. Yeah, we've talked. I think there was another, like, uh, I think the common term earlier in the days was prompt engineering.
Yeah. Yeah. So we like, well, I think our first episode was on prompt engineering, right? Right. Cause that was like just in the AI culture and the discussion, that was what it was all about. And context engineering is, yeah, for those that are on, uh, uh, just listening to, for the one person that listens to us, you should go check out what the other two people that listen to us do online. You can see the simple graphic. A couple hundred people listen and a couple hundred more watch the YouTube. It's not, it's not like a, it's like maybe three to two ratio anyway. Awesome. Well, for those that are watching, this is like a, like sort of a, a bubble chart, kind of Venn diagram.
You did not just say the word bubble. Yeah, I don't, let's not get into the bubble conversation. Um, but, uh, but, um, but the idea of context, well, here's the, here's the good news context engineering came out. Um, recently, and by the way, this, this paper from Google called context, engineering sessions and memory. So for those of you that are listening, if you just Google the term context, engineering sessions and memory, Google will show you the Google paper. It's a huge, it's a 72 page paper that I love. Just grab it, throw it into a chat so you can have conversations with it. You don't have to read every word if you don't want to, or read every word.
This just came out in November. So this just came out last month. This was so important to them that they put out a 72 page, uh, document. There's been a lot of blog, uh, articles, uh, coming out. Um, a lot of people talking specifically about this and, um, just discussion everywhere. And you'll even notice in the prompt engineering guide, they talk about context engineering. So first of all, what is, what is a contact? And then actually, uh, I mentioned before how. Anthropic we'll put some of these links in the show notes. Um, every time anthropic drops something, you should run and go read it. So this really got everything started when they dropped their blog, effective context engineering for AI agents, which was in the end of September.
So really, this has been like the, the, the number one, uh, conversation. And really it's about, so here's the good news. It's not some whole new thing that you have to just be like, oh God, how do I keep up with all the new chatter and AI? This is really my reaction to this was, oh man, what is now what's contacted during? Oh boy. And then I look. And I go, oh, oh, oh yeah. This is all cool. This is all the way I'm doing it. Awesome. It was just like kind of good validation. And for those that are kind of on their learning journey, this can kind of help you like jump from just prompt engineering into context engineering. So first off, we're showing a graphic on screen about, and you teed it up.
We've talked about prompt engineering before. What prompt engineering really is, is the art of the question. Um, you're a hitchhiker's guide to the galaxy fan. There's a, a book and a movie of course. And there's the, we talked about 42, you don't even, okay. It's so bad. It's very bad. I I'll admit, but for those that watch movies, there's a chapter for those that read books. But the, of course the big one, we actually talked about 42, you before, and the way that 42 you school got their name is from the scene in that movie where it was, uh, what is it called? Deep think or big brain or something like that was the largest, fastest computer in the world.
Yeah. I mean, and then the smartest people in the world, the question, what is the meaning of life, the universe and everything. And then it answers with 42 and they're like, what, what are you talking about? 42. And then what big deep brain said was in response was, well, that's the answer, but you got to focus on the question, right? So that actually is what prompt engineering was all about. Like, how can I be really clever with my questions? And you know, what order do I use with all my questioning very much in art, kind of a science to it as well, too, but really what that is based on is like, we've talked before about single shot prompting, like, just like, right.
Like I I'm trying to wrestle too much into the format of my question to try to get what I want. It's really two things though. It's a system prompt, which is like, you are a something here's the industry you work in. These are good. These are examples of questions. These are example of answers. So that's cool. That's all part of system prompt. Yeah. Well, that's context. So, so that actually, that's a perfect way to talk about, so when you're just doing prompting where you're not doing a system prompt. Right. Um, that's like where you kind of start with prompt engineering. Then you kind of go from like just the prompt to what you just described before the prompt even happens.
I've got a system prompt that describes some context about this conversation that I'm having. That's really where people were going on their journey. What context engineering is, is systematizing, like very profoundly changing the way that you're providing context to the conversation and for those that are watching on YouTube, they can see that this involves everything from, well, actually, I'm going to go back to this little graph because, um, this is, yeah, so the bubble chart thing was really interesting because it was like a, it was a large circle that, uh, called constant engineering, which entirely contains prompt engineering, rag state history, and memory.
Yeah. And then structured outputs is half in half out of the, of the big bubble. Yeah. It's kind of interesting. Yeah. So really though, if you look at it, these are all things that we've all talked about before. These are actually episode concepts. So prompt engineering is one of them rag that's retrieval augmented generation. People I think are pretty familiar with rag and then, uh, state and history is where I'm, that's probably it's at state history and memory are kind of the same thing because you facilitate state through memory, but they're the management of them is a little bit different. But when we talk about memory, we're not talking about like just conversation memory, like where you can allude to something that happened in the conversation before, cause you're feeding the whole conversation.
You're feeding the whole conversation into the LLM. This is about like introducing state into AI. One of the things, the reason I'm so obsessed with stateful AI is I can't believe it doesn't exist. Like, I think this is because like, you know, guys like you and I, we come from database world. We're literally everything's memory, right? Like we're just, all of our data is in memory and this has been our whole lives. So to have this whole industry, like be born out of a single conversation. Memory. It's so bizarre to me. Like the, the equivalent of this would be, you have a family member or a neighbor or someone you see all the time. And every time you see them, you have to reintroduce yourself to them and explain everything about yourself all over again.
Like that would be incredibly, yeah, right. Incredibly frustrating. Yeah. It's like, yeah, it's totally like a 50 first dates thing. So in like so many plot lines now that you've mentioned it. So that'd be really frustrating. So I'm just fascinated, like why we're not all super frustrated with our chat. Experiences you know, introducing statefulness into that would allow us to persist things across different sessions. As a matter of fact, I think one of the most fascinating areas within AI right now is introducing complex memory systems. And I'm not just talking about databases that you're retrieving from that's more rag talking about like a memory system.
That's listening to your chat that you might say something like I don't know, you know, Hey, I'm connecting to JIRA and we use the. The, the field, the story points field in JIRA, we actually put hours in there, not story points. Well, having a memory system actually listening for those types of pieces of information and go, Oh, Matt just said something really interesting about how he uses JIRA. Let me save that over here in a very specific type of memory so that I can retrieve that in future conversations. And he doesn't have to keep telling us that. Right. And if you multiply that one, like little experience by just the hundreds of times, you share things that are meaningful.
To future conversations, you can really get a completely different type of experience. Yeah. So that's really. Way more human. Like in this one area. Yeah. Exactly. It's very much more human. Like as a matter of fact, the, if and we can do a podcast episode on this in the future, but the way that I would actually urge people to kind of look into some, how some of the AI memory systems work there's one from, there's one called laying mem. If you want to check it out, it's absolutely patterned exactly like the brain does it. There's like procedural memory stores, there's episodic memory stores, like there's different types of memories and our brains manage them differently.
I think the best analog to it is like Ram versus ROM, right? We get how those hard memories are different. Our brain does a whole nother sophisticated thing. And they're actually, they meaning the AI builders out there are, are mirroring these types of things in the products that they're building and building in a ton of us in there. So, so it's really about like preparing the context window. With tons of information, like application level information and tools and memory and sophistication to make it so that these conversations are not just like one offs and, and that's really prompt engineering is about one off conversations essentially. So that's, I don't know, maybe a, a long winded version of what context engineering is, but it's really what everybody's talking about right now.
Yeah. So I can see you've got documents, tools, memory, tools is a big one, dude. Yeah. Yeah. Yeah. Yeah. Yeah. I mean, like, like, like I have a client that's building a software product and they're kind of just, they don't really know anything about AI. So they're just like learning as they go. And they were trying to build it as just one single prompt that, but they have a multi-step like product process. Right. So you can't do all that in one prompt. At the very least you have to break it up into multiple different, you know, calls to the LLM. And then they were trying to do things like make it do math, like LLMs can't do math. And so. So you don't just stop and throw your hands up and go, oh, LLMs can't do math.
No, have, go have something that does math, do math. Like the most common is a Python calculator. You can do SQL queries, whatever. Those are examples of tools. So just the tons and tons of tools that you have at your disposal to, to take the load off of LLMs and to have actual purpose-built things do those is a very big deal. Okay. Documents, you know, you can have business context. You can have context about the work you're doing, the project you're working on, whatever. And those are documents that can be retrieved and pulled into context as well. Yep. And then all of these can actually be run in a single API call where it uses all of this information before it gives you the result.
Yeah. Yeah. It actually is. It, it, at the end of the day, you're still, what it's doing is it's packing in dynamically your API data, like payload. Right. And then these are getting sent over with the user's interactions so that the user doesn't even have to bother telling all these stories. It's the same thing as you run into your neighbor again. You don't even have to tell them who you are and your whole life history again. And I see like message history is in there. Right. So a lot of this stuff is more permanently stored, right. And stuff like that. But then you've also got, oh yeah, yeah. In this conversation, these are the API calls you've done recently.
Okay. This is cool. Cool. Cool. Cool. Yeah. And you see like domain knowledge and comprehensive. Those are like procedural memory stuff. Like I don't have to explain how I do my job or what my KPIs are or whatever. Like all that stuff is coming over contextually. There's a huge thing there that's called curation right in the middle. Yeah. So curate. Well, curation is actually kind of where humans come into play. Right. Like where consultants can come in and say, well, I can help you like make the data that would be useful to your AI interactions, make it available to your AI interactions. Right. So the curation, it's all the stuff on the left for those that are looking at the possible context to give just exists, but the context window is going to go retrieve that stuff or curate that in its own context, whatever's relevant to the conversation that that user's having.
Mm-hmm . That's kind of what that, that sort of means there. And I'm a many, many, right? They have those to be spun. Yeah. Oh, for sure. I mean, you know, this, this is context engineering is we're talking about it now because we have long context windows that we can actually work with, which means that we can stuff more things into it. But it also means we have to be more purposeful in our, our cramming stuff into context. It's signal noise stuff, right? Like we can't just go grab everything. We want to grab just the stuff we need. Mm-hmm . And in this case, like even tools to help us with that. Yeah. Yeah. Yeah. Yeah. Yeah. Yeah. Yeah. Yeah. Mm-hmm . We can't just go grab everything.
We want to grab just the stuff we need. Mm-hmm . And in this case, like even tools to help us with the stuff that we need. So it's, it seems really sophisticated, but really, if you just go on your own AI journey, you're eventually going to start doing all this stuff anyway. So I guess what I would say to new learners and those that are just kind of getting their feet wet or just have curiosities, skip over all that you got to go figure this out on your own and just dive right into context engineering and maybe learn backwards from there to where you're at. Because this is where you're going to end up anyways. Okay. And, and this was probably the single biggest message at two of those conferences that, that I attended recently.
So if that helps anybody. Yeah. Sort of defining what it is and what it's not and what areas to gain skill in, et cetera. Correct. Okay. And the good news is we, we've covered a lot of those things in previous episodes and we will look forward to covering a bunch of them in future episodes as well too. Okay. Okay. Okay. Okay. Okay. Okay. Okay. Okay. Okay. Okay. Okay. Okay. Okay. Okay. Okay. Okay. Okay. Okay. Okay. Okay. Okay. Okay. Okay, well, I'm here to tell you exactly how I use the Trillion tokens. So I actually have the breakdown right here, right out of my dashboard. So 200 tokens I used for meeting summaries, 350 tokens, I did some chunking, 790 tokens, image descriptions, you know, throw some images at it and have a look at it.
And then I have some of the other tokens that I used for the meeting summaries. And then I have some of the other tokens that I used for the meeting summaries. descriptions you know throw some images at it and have it describe it and then 999 billion 999 million 998 660 tokens were because i left an agent running without an exit condition they make a more efficient paperclip problem yeah dude it was it was a computer use thing for back in you know 2023 and i just didn't know it was running the whole time so it's a real nothing burger uh nothing to see nothing to see here folks come on that can't be it yeah that was it i mean so you know you really left something running yeah it would have cost you a lot of money yeah just yeah just it's like you know you leave the faucet running and then you go to your vacation home for a few months you know you're going to come back to a swimming pool folks i'm here to tell you you like make sure not to make the costly mistake of uh leaving a drip in your uh tokens uh that's going to allow something and just catastrophic end loop if token movement is going to happen and you're going to have to make a big mistake of uh leaving a drip in your window exceeds one trillion literally it was just that easy if i would have just just taken a moment i'm always so hasteful so um so yeah i'm just here to tell you guys like always add an exit condition when you're running agents very important well i mean i'm guessing you
probably had a legit billion or several billion uh and that this you know it probably wasn't quite 99.9 my wife misunderstood it to be like oh yeah i mean you use chat gbt more than anybody she's like yeah i believe it you totally did and i'm like well no that wasn't it she's like no that is it like that's all you're ever doing is using chat gbt so i'm like all right whatever what was the output of that crazy loop that it did it solve uh nothing it literally just uh you know what it was the answer was uh 42 we did it we brought it all the way back to uh nice 42 actually yeah and i got this uh sweatshirt too so you know so there's that uh yeah very good stuff now i don't have any more conferences before the end of the year do you no i don't have any till i don't know spring yeah same here i think nobody really does more with the holidays so we're gonna have a couple more episodes though uh before our end of year yeah we got to do our end of year where we're gonna like look back on the whole year and and maybe make some predictions for the coming year maybe we can even go back to that last years and like see yeah actually we should oh my god we should totally do that yes that's
what we'll that's what we'll do and then um if we can maybe we can cram in a little uh go get louis to talk about mcp and file maker for us um if that's of interest to anybody that would be amazing yeah um do you have any plugs uh i think i kind of hit mine i'm i'm trying to really change um uh my um i don't know business model maybe like training i really want to do a lot more of ai integration and ai teaching cool and try to move much more into that area than just regular being an immediate advanced file maker classes yeah i still definitely want to do those i love them um i don't have any on the schedule at the moment because you know holiday season and whatnot of course um but i'm i think i'm much more interested in like really helping companies who don't have any um i think i'm much more interested in like really helping companies who don't have any um i think i'm much more interested in like really helping companies who don't have any integrated and i i can't tell you how many really large claris partners who are presenting at conferences about ai and i ask them hey how many uh how many of your customers have any integrations yeah and how often i hear the answer is zero or one yeah no i that would be my guess yeah i have
eight in my little teeny tiny i think well and i think i think it speaks to your approach right so um i think it speaks to your approach right so um i think it speaks to your approach right so again i not just being critical of clairs but like pointing to that example that all of our list all five of our listeners are aware of clairs has been focusing on terms and concepts and acronyms and and it's just not landing and that's why you go to even the biggest partners and they're like yeah i'm not doing any of it because when i go to my customer and say you want some semantic search and my customer goes dude i don't know what you're talking about right and also no because it's not that good well yeah i mean but it is sure i have some tricks i don't know what you're that make it great hey that's why people should go to your class really good yeah but i think the difference is in my one of my predictions for the new year is that i think clairs will change the way that they're marketing and that what they'll say this is this is one of my predictions as a teaser to our end of year show i think clairs is going to start focusing on outcomes as they should um and what i mean by that is like um what you're basically what you're talking about like you like the exact same thing you're saying is like forget just the acronyms and the tech and the whatever just like what if i made you not have to copy and paste 300 times and it would just actually go right into the fields you're interested cool do you even care what tools
i'm using no great let me actually do that for you go plug this hole plug that hole solve this problem make this more efficient people will realize that their efficiencies have gone up their productivity will follow as a result and they've integrated ai it's not like you got to do this big pull the big lever and all of a sudden ai is doing everything it just doesn't make sense it's it's just another utility like like anything else and it's a very powerful one with a lot of things it can do but you don't just look at it piecemeal and do little tiny things i do believe it has to do with messaging though um i had i was i i'll give an anal i hope this analogy lands we'll see or anecdote anecdote lands i was in my kitchen eating dinner and uh i feed my dog table scraps forgive me um i won't feed your dog table scraps but i do that to my own you have to come visit me first chris of course that may happen yeah very we'll talk about that offline but um so i put the bowl down on the ground my dog comes over and i go oh wait there's some crazy stuff in this recipe they were like curants and and like um some other wacky stuff so i go let me ask chachi bt if those are toxic for dogs onions too or garlic yeah very bad for them and by the way curants were toxic as well too so i'm glad i caught that
moment and then i i said to my wife i go you know i got that damn home pot in here it's just worthless it's listening to us all the time but what a real good ai utility would it should have been hey you knew i was cooking or let's say i asked ai for help with my ingredients or did the recipe you know that i've got stuff toxic there you hear me actually call my dog over you can contextually understand i'm about to feed to him i shouldn't have to go ask the ai implicitly hey can i feed this to my dog it should be proactive right i thought well that would be a cool product your wife would right or yes your kids or other people in the house yeah thinking okay dog okay what do we have to wait wait we had you know yeah piecing all the context together and making sense of it now i said to her well that would be a cool product but then i realized this and this is where i'm going with the way we market ai there's two ways you can market the product i just described you could say hey you guys want to put ai llms language models in your you know always listening to you to guide your home constantly listening to you ai llm gpt or do you just want to have something that prevents you from killing your dog by accident like which one of those two things do you think people will gravitate towards go to the outcome one instead of the thing with all the acronyms and all the tech in it what's going to happen chris is the real world situation is going
to be you're having a discussion with your wife and yeah it's going to go actually chris two days ago what she said was quote and then your wife's going to laugh because she was right and you were wrong or whatever damn you stateful and see that's the same thing it's not you know it has to be able it is it's that stateful ai dude all that memory integration so i guess like the point is what you're doing with your customers i think is perfect it's just what can we fix uh you know the the whole ai is not going to replace you just the stuff you hate to do that's kind of a silly way to say look there's a lot of rote tasks that don't require your knowledge as a human just get rid of those and make those easier so you can go do human things more often and better right yes i mean but i think if i could summarize my whole approach of what i want to try to do i like you said as i'm just going to replace the things you hate to do it's basically you know have these have this assistant do stuff so that you can do the core of your job the things that matter most the things that you love most about your job that drive you forward so you don't have to do this busy work of copying pasting and you know i translating or summarizing or analyzing or it's exactly exactly it matt like i couldn't agree with you more i on friday i'm actually speaking at a conference i was brought in to do a couple ai workshops for educators and my message is literally
like hey ai can't replace that human element that empathetic listening element that an educator you know brings that you know that teachers do in the classroom there's so much human touch stuff that's incredibly important but you doing lesson planning and grading and feedback that's just killing like a bunch of people so i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really
important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really important to do that and i think it's really