Latest / CE Pro Podcast / CE Pro Podcast #183: How Integrators Are Using AI, from the CEDIA Expo 2026 Smart Stage

Transcript

CE Pro Podcast #183: How Integrators Are Using AI, from the CEDIA Expo 2026 Smart Stage · CE Pro Podcast

• 0:02
  SPEAKER_01: Good morning, everybody, and welcome to AI for Integrators, the tools that save hours every week.
• 0:08
  SPEAKER_01: I personally feel a little intimidated by this discussion.
• 0:12
  SPEAKER_01: Not because I feel ill-equipped to talk about it.
• 0:15
  SPEAKER_01: We have a tremendous panel gathered here today, minus Jared Ross, who unfortunately couldn't make it due to a last-minute emergency.
• 0:21
  SPEAKER_01: He is being replaced, though, by Chris Worthington, who is just as capable of talking about this subject.
• 0:28
  SPEAKER_01: But I'm intimidated because how do you even introduce AI at this point?
• 0:33
  SPEAKER_01: I know it, you know it, the American people know it.
• 0:35
  SPEAKER_01: As business owners who stand here today, because you know how important it is, I mean it's in the title, the tools that save hours every day.
• 0:44
  SPEAKER_01: Honestly, I feel like I don't even need to be here.
• 0:46
  SPEAKER_01: I can probably just swap my mic off to these guys and just sit in the audience with you and listen.
• 0:51
  SPEAKER_01: But I guess the what and the why are kind of self-explanatory.
• 0:56
  SPEAKER_01: What I find most exciting though, and what we're gonna try and talk a little bit about today, is the how, how AI helps your company, how AI can affect this part of the business versus this part of the business.
• 1:09
  SPEAKER_01: And it's the how that's always changing as we more learn more about what AI can do.
• 1:16
  SPEAKER_01: So today, we're going to explore some of these use cases, and ideally, we'll be able to piece together how the conversation about AI has changed from when it first started, how some of our panelists are using AI to boost their businesses today, how the different tools available can be used specifically for building out your custom integration business, and how you can get started with those tools today.
• 1:42
  SPEAKER_01: So, I'm gonna start off with a general question for everybody, uh, and I'll let you guys fight over who gets to talk first.
• 1:49
  SPEAKER_01: How do you use AI in your business today?
• 1:57
  SPEAKER_05: Hey everybody, I guess I'm starting first.
• 2:00
  SPEAKER_05: Um we use AI to bridge gap between software packages that we use in order to eliminate uh manual uh uh data transfer between several software packages uh in our company, and then additionally on top of that, we also with some functionality gaps that we also bridged by using AI, writing our own software layer on top of all the SaaS software stack that uh our company is using, and that helped us tremendously to reduce errors, increase speed of uh communication between uh different departments to pick up uh uh steps in the workflow of the project going through.
• 2:51
  SPEAKER_05: So that's our primary uh primary way we save a lot of time, and then obviously everyone is using it to write better emails and analyze long uh support uh threads and things like that where you kind of start losing the ins and outs out of uh subject.
• 3:10
  SPEAKER_03: Like like Serge, I started using AI to build tools to bridge gaps in the software.
• 3:16
  SPEAKER_03: We all use software and it's not perfect, it it is lacking in certain areas.
• 3:20
  SPEAKER_03: We use a bunch of different ones, and so I built tools to start to fill those gaps and move data around.
• 3:27
  SPEAKER_03: I migrated from there to just building the software itself because I realized that I was using tools to fill holes because the software itself didn't do what I wanted, and I came to the conclusion that if I just built the software myself, then I wouldn't have to build 20 different tools to move data around.
• 3:45
  SPEAKER_03: So today we use uh uh a software that we wrote with AI, and we use AI actually almost instead of the software to build beautiful, accurate proposals quickly, um, send those out to customers, generate sales, and then take that same data set, that same source data, and then have that propagate through the life cycle of a project from purchasing, accounting, inventory tracking, delivering, revenue recognition, et cetera, um, so that it's as the process is as automatic as possible.
• 4:16
  SPEAKER_03: Um, and everyone on the team, whether you're an accountant, um, office manager, project manager, et cetera, has access to all the information that they need at any given time.
• 4:26
  SPEAKER_03: We found that one of the biggest problems in my business over time is just people not having the information that they needed and waiting, delaying, asking questions, et cetera.
• 4:34
  SPEAKER_03: So our goal is to make sure that every team member, whatever their role, is as well prepared and well-informed to execute as they could possibly be with a singular point of data entry.
• 4:47
  SPEAKER_04: Um I'm Ari with Sonance.
• 4:49
  SPEAKER_04: I'm the only non-integrator on the panel, so we're a manufacturer, and we're using AI throughout our entire business.
• 4:54
  SPEAKER_04: Um, I would say that all types of knowledge work are out of our 360 employees, we have about 125 actively using cloud co-work to get knowledge work done.
• 5:06
  SPEAKER_04: Spreadsheets, slide decks, um, and everything in between.
• 5:10
  SPEAKER_04: And then uh much like the rest of the panel, um, we are also building our own software.
• 5:15
  SPEAKER_04: Um, not necessarily to replace the SaaS tools that we have, but to build an abstraction layer above them.
• 5:20
  SPEAKER_04: Um started off to solve problems and uh in different parts of the business.
• 5:25
  SPEAKER_04: And as it grew, now we're bringing it all together to build one entire operating system to run the business on.
• 5:32
  SPEAKER_00: And now Mark DePetro with Sound Vision and Sound Decisions in out of North Carolina.
• 5:37
  SPEAKER_00: I don't want to mimic, waste time and mimic what what these other gentlemen have said.
• 5:41
  SPEAKER_00: Very similar.
• 5:42
  SPEAKER_00: I would say two things that we do.
• 5:44
  SPEAKER_00: We are using it in a lot of different areas of our business.
• 5:47
  SPEAKER_00: One is to make people the best version of themselves.
• 5:52
  SPEAKER_00: Uh, if you want to think about having like an agent as a as an assistant, uh something that can take the routine, simple tasks that they do every single day and remove that from them so that they have more time, so they can be more productive, or so they can concentrate on work that really requires a human.
• 6:12
  SPEAKER_00: So that's one thing.
• 6:14
  SPEAKER_00: And then the other thing is like these gentlemen have said, you know, being able to take knowledge that you may have somewhere and actually use it.
• 6:25
  SPEAKER_00: We're very process heavy, very process-oriented, and we've we've spent hundreds and hundreds but thousands of hours of creating processes, but they're just in OneDrive.
• 6:34
  SPEAKER_00: Or they're like, you got the document, but how do you actually use that document?
• 6:39
  SPEAKER_00: Do people know where it is?
• 6:40
  SPEAKER_00: Do they consult it?
• 6:42
  SPEAKER_00: And do they actually like follow it?
• 6:44
  SPEAKER_00: So being able to implement that into whatever your systems are has been really uh really helpful for us.
• 6:51
  SPEAKER_01: Now, a big theme I am seeing here across all of what you guys just said is a lot of this work that you're doing with AI seems to be in the software development aspect, whether it's developing like full-blown like programs or just kind of like finding augmentations to existing programs that you're using throughout your business.
• 7:11
  SPEAKER_01: Would you guys say that's kind of where like the cutting edge of uh AI use in the business is today for you?
• 7:19
  SPEAKER_05: Uh are we going in the same order?
• 7:21
  SPEAKER_05: Okay.
• 7:23
  SPEAKER_05: Sure.
• 7:24
  SPEAKER_05: The end of last year, I would say November-ish, December, basically breached the paradigm of programming.
• 7:34
  SPEAKER_05: It made it accessible to anyone.
• 7:36
  SPEAKER_05: As long as you have an idea and they can and you can verbalize it, which I'm terrible at, but it still understands me.
• 7:43
  SPEAKER_05: Um you can write a software uh for any purpose that you possibly can without actually understanding any ins and outs of uh um back end of the software stack, how the database is created, how to host it.
• 7:58
  SPEAKER_05: There's a wonderful thing you can do with AI.
• 8:01
  SPEAKER_05: You can tell it, do it for me, and then literally solve all the problems that used to take an engineer to solve.
• 8:07
  SPEAKER_05: So the moment that happened, and that was the end of last year, we basically started to look at AI only as a tool that uh helps us build tools that we need to bridge the gaps in the operations.
• 8:23
  SPEAKER_03: You can verbalize it in Russian too, as long as you can verbalize it in basically any language it that works.
• 8:29
  SPEAKER_03: Um, one way to think about the the way that I started thinking about um what the these software programs, Claude and Chat and all these others, they're you know, they're large language models, right?
• 8:40
  SPEAKER_03: And what does that mean?
• 8:42
  SPEAKER_03: They're they're translators.
• 8:43
  SPEAKER_03: They you can speak English and it can spit out Russian or anything else like that.
• 8:49
  SPEAKER_03: If you think of the software that you use as just the expression of a language, it kind of clicks to me, it clicked to me a while ago that you're really just translating into a language that is then producing a web page or a software application, et cetera.
• 9:05
  SPEAKER_03: So that's the sort of fundamental way I think about what is the actual tool in the first place.
• 9:10
  SPEAKER_03: Um yeah, software development's a lot of what we do.
• 9:12
  SPEAKER_03: I think that we're all kind of those types of people.
• 9:14
  SPEAKER_03: We all love software that makes our life easier, so we do spend a lot of time on that.
• 9:18
  SPEAKER_03: But I encourage my um everyone on the staff to, and I myself, to use it for anytime you have a question or just to help you think through something.
• 9:28
  SPEAKER_03: Um so just problem solving broadly, but I think, yeah, 75% or more is probably developing software.
• 9:35
  SPEAKER_04: Sure.
• 9:36
  SPEAKER_04: Um so what I think of as the um cutting edge of this is yes, writing software, but there's some fundamental things to do prior to that to be ready to write the right software, which is to get data in the right place and clean and accessible and get information out of people's heads into a place.
• 9:54
  SPEAKER_04: And so we're building a knowledge graph of our institutional knowledge, we're building a data lake house for our data to all build software on top of that is uh tailor-made for our business.
• 10:05
  SPEAKER_04: And I think that that's definitely where I think things are going.
• 10:09
  SPEAKER_04: And then we haven't really spoken too much about agents, but uh they're here, and uh they are our coworkers, and that's the I think really what the cutting edge is to build your own software, build agents that know how to use your own software, and have those be coworkers to your people that can collaborate and get things done more efficiently and with fewer mistakes.
• 10:31
  SPEAKER_00: So I'm gonna take it a little different angle here.
• 10:34
  SPEAKER_00: What I see a lot, and this is not cutting edge at all, but what I see a lot is people are scared.
• 10:41
  SPEAKER_00: People are scared to like try, like maybe, maybe you'll dip your toe in and rewrite an email or or read a document or or something.
• 10:50
  SPEAKER_00: But like like Serge said, you know, if you just all you have to do is try, like just talk to it and ask it.
• 10:58
  SPEAKER_00: I ask it the dumbest things in the world.
• 11:00
  SPEAKER_00: Sometimes I ask it two and three times.
• 11:01
  SPEAKER_00: I ask it, where did I talk to you about that before?
• 11:04
  SPEAKER_00: Because I lost where that was.
• 11:06
  SPEAKER_00: There's I cook with it all the time.
• 11:09
  SPEAKER_00: So, you know, it yes, all this is true, but if you're in that camp that you're like, wow, like programming, okay, I hear you, but you guys have done this.
• 11:19
  SPEAKER_00: It it is truly that easy.
• 11:22
  SPEAKER_00: And if you just dip your toe and try, you'll be amazed what you can do and produce.
• 11:28
  SPEAKER_01: So I think that's a very good point to jump into one of the other questions that I had, which was um, what tools do you recommend that you know, people that are hearing us talk about this, um, like what what what are the good tools for them to start using to kind of like explore and get into this stuff?
• 11:48
  SPEAKER_05: No one explains this better than Ari.
• 11:52
  SPEAKER_04: All right, I'll take this one.
• 11:54
  SPEAKER_04: Um so I I think the thing that I would suggest to people is that they recognize that we're beyond chat.
• 12:00
  SPEAKER_04: Okay, chat is ask a question, get an answer.
• 12:03
  SPEAKER_04: We're in the get stuff done phase.
• 12:05
  SPEAKER_04: And uh to experience that, and I believe that um in order for a company to adopt AI correctly, a leader doesn't have to be the leader, but a leader has to have their own wow moment with it.
• 12:18
  SPEAKER_04: And I think the tool that I have seen produce the most wow moments for people is Claude Cowork.
• 12:24
  SPEAKER_04: And I would have to say that ChatGPT's product, ChatGPT work, has really caught up to it.
• 12:30
  SPEAKER_04: And um, I would say one of those two products.
• 12:33
  SPEAKER_04: Um, and they are more of an agentic tool that instead of asking it a question, you use um delegation terms like build me this, design me this.
• 12:43
  SPEAKER_04: And I think that is the gateway to really understanding how powerful and capable AI is today and how to harness it.
• 12:52
  SPEAKER_04: And then once that ball starts to roll, you start to build dashboards that lead into software, and that's where I would start is either cloud cowork or chat GPT work.
• 13:03
  SPEAKER_00: Can I add one thing to that?
• 13:04
  SPEAKER_00: And and that is one of the issues with AI now is it's it's moving so fast, like every day if you're on X, there's like the next thing and the next thing and the next update, and you feel almost like you're behind all the time.
• 13:15
  SPEAKER_00: I should say I do.
• 13:17
  SPEAKER_00: So I 100% agree with cowork or chat GPT work.
• 13:21
  SPEAKER_00: It doesn't really matter which one you pick, pick one and get comfortable with it, and don't necessarily feel like you've got to get the newest thing, latest thing.
• 13:29
  SPEAKER_00: Just get one that you're comfortable with and then move as you get more comfortable.
• 13:35
  SPEAKER_04: I want to add one thing.
• 13:36
  SPEAKER_04: Also, learn how to do what's called reverse prompting.
• 13:39
  SPEAKER_04: Everybody thinks that they uh they remember this idea of prompt engineering.
• 13:44
  SPEAKER_04: To me, it's dead.
• 13:46
  SPEAKER_04: That was a concept from when the models were not as smart as the people.
• 13:51
  SPEAKER_04: And you had to be very specific about what you wanted the model to do.
• 13:54
  SPEAKER_04: We're past that.
• 13:55
  SPEAKER_04: So, reverse prompting is where you say, given what you know about this project or about me or my business, what do you think we should do next?
• 14:02
  SPEAKER_04: And I think that will unlock a whole lot.
• 14:05
  SPEAKER_04: And then lastly, I would say is um sometimes AI will say, This is how you do something, and it'll give you these instructions and learn to say, can you do it for me?
• 14:14
  SPEAKER_04: And you'll be shocked at how often it can.
• 14:18
  SPEAKER_03: Yeah, I think I agree with all of that.
• 14:20
  SPEAKER_03: And to put a little bit of a finer point on it, just take one task that you're spending time on in your day, especially if that task involves a spreadsheet of any kind in any software, and say, you know, if you're spending an hour a day on something 30 minutes a day, um dump that spreadsheet into cloud, chat, whatever, and just say, can you help me do this or or figure out a better way?
• 14:41
  SPEAKER_03: Reverse prompting is a great is a great place for that.
• 14:43
  SPEAKER_03: Um so start with something small.
• 14:45
  SPEAKER_03: That's how I started.
• 14:46
  SPEAKER_03: I just started by looking at deficiencies in my process, and just one little victory, understand the tools, your confidence will grow from there.
• 14:54
  SPEAKER_04: A lot of people don't know that you can open up cloud or chat GPT inside of Excel and then watch your spreadsheet transform in front of your eyes, and I think that's a great gateway into this as well.
• 15:06
  SPEAKER_00: That works in all the Office uh applications too.
• 15:10
  SPEAKER_00: Word, PowerPoint, Outlook, all of them.
• 15:14
  SPEAKER_01: So uh Ari, I was talking to you a little bit earlier about this, just kind of like uh, and we we sort of brought it up just now.
• 15:22
  SPEAKER_01: Um, the notion that AI is always changing.
• 15:25
  SPEAKER_01: Like next, like literally the next day, like what we're talking about now could be somewhat invalidated essentially.
• 15:32
  SPEAKER_01: So I am curious to hear from both you and the rest of the panelists how how do you keep up to date on this stuff?
• 15:40
  SPEAKER_01: Not just in like what the new releases are, but like how people are using those releases.
• 15:47
  SPEAKER_04: YouTube.
• 15:50
  SPEAKER_04: I'd say uh Twitter, I still call it that, uh, and YouTube.
• 15:53
  SPEAKER_04: I would say those two things.
• 15:55
  SPEAKER_04: Um and my what I try to do is I watch somebody do something.
• 16:00
  SPEAKER_04: I pause, I try it my well, I make it my own, I make it personal to my life or my business, and then I try it myself, and that's how I've learned everything.
• 16:10
  SPEAKER_03: I just wait for Ari to tell me what to do.
• 16:14
  SPEAKER_03: Talk to your friends, talk to other people that are doing it, seriously.
• 16:17
  SPEAKER_03: Um and and you know, like Nick said, you you don't feel like because something came out ten seconds ago, you need to switch everything you're doing.
• 16:23
  SPEAKER_03: You you know, if you find something that's useful and helping you, it's fine to be a week or two or you know, a little bit behind whatever the newest cutting edge thing is, because like in another week it'll be something different entirely.
• 16:34
  SPEAKER_03: So start doing something useful, understand the tools, and then talk to the other people that are using it.
• 16:41
  SPEAKER_05: I just want to add take breaks, have some sleep.
• 16:46
  SPEAKER_05: Oh yeah, you know, um AI is not everything, it will be, but not right now.
• 16:52
  SPEAKER_05: We're still time to sleep, and then um uh in all seriousness, the uh tools um they get pushed forward and then sometimes they retract backwards in capabilities, but essentially, anytime you touch anything, this is the worst it's ever gonna be.
• 17:10
  SPEAKER_05: Right?
• 17:10
  SPEAKER_05: So you always wanna if you stumble across something, you you can always wait a week and then go back to it.
• 17:18
  SPEAKER_05: You'll have that idea how to uh push it forward, and then the tools by that time might get even better.
• 17:24
  SPEAKER_05: So just uh taking breaks and uh your um use of AI is also very helpful.
• 17:32
  SPEAKER_00: Uh here's a practical tip.
• 17:34
  SPEAKER_00: I use uh Twitter and I just create like a little group uh on whatever you whatever you're using.
• 17:40
  SPEAKER_00: Let's just say it's Claude and you have a Claude group, and then in the morning or whenever you see it, I'll take that and I'll actually share it to my LLM and just say, hey, can we use this?
• 17:50
  SPEAKER_00: I don't even read it or watch it or whatever.
• 17:52
  SPEAKER_00: I just have it do it for me, and it knows what I'm working on and knows how I work and that sort of thing, and so it knows how to apply that.
• 18:00
  SPEAKER_00: And I tell it, like, don't be syncophantic, like I tell me if it's actually useful or not.
• 18:07
  SPEAKER_00: So that's that's something to do.
• 18:09
  SPEAKER_04: Yeah, one thing a lot of people don't know that you can do is take these uh posts from Twitter or a YouTube video and send that to Claude or ChatGPT and say, hey, given what we're working on, do you think this thing that they just talked about would be helpful?
• 18:23
  SPEAKER_04: And often it'll be like, well, I would skip that first part, but that second part we can modify and use in this way, and that's how I I've done a lot of advancement.
• 18:34
  SPEAKER_01: So I guess that's leads us into our next question, which is we've already talked about like some of the tools that people can get started with, and we've kind of touched upon different use cases that we can use those tools tools for.
• 18:46
  SPEAKER_01: But I guess like in it's specifically in their organizations, like what are some use cases that you know if they're not already using AI, they can kind of like start there to like ply with.
• 19:00
  SPEAKER_00: Well, I'll give you one we used, I stumbled on about a year ago.
• 19:03
  SPEAKER_00: Uh, as an integrator working with builders, we get uh subcontractor agreements.
• 19:09
  SPEAKER_00: And I used to read them.
• 19:10
  SPEAKER_00: I mean it's a you know, it's a legal document.
• 19:12
  SPEAKER_00: And I used to read them, and the ones I get are about 15 pages on average.
• 19:16
  SPEAKER_00: And I just one day said, I'm just gonna make a project with this, and I want you to read this, and here's what I'm looking for, and then tell me, you know, the insurance and blah blah blah.
• 19:25
  SPEAKER_00: So I did that.
• 19:26
  SPEAKER_00: Well, a week later or so I got another one from another builder we work with.
• 19:30
  SPEAKER_00: And I said, Well, okay, read this one.
• 19:32
  SPEAKER_00: Oh, by the way, compare it with the first one.
• 19:35
  SPEAKER_00: Tell me, like, you know, what are different, what are better, worse, whatever.
• 19:39
  SPEAKER_00: And so now I've got like 14 in there, and it it has it it built itself a custom grid based on what I thought was important and what it found to be important.
• 19:50
  SPEAKER_00: And literally one builder had insurance that was five years old.
• 19:56
  SPEAKER_00: And I sent it back to him.
• 19:57
  SPEAKER_00: By the way, I'd never read on the contract, I think I was too scared until this.
• 20:01
  SPEAKER_00: I read on every single one of them now.
• 20:04
  SPEAKER_00: And I told the builder about their insurance was too old, and they were like, wow, thank you so much.
• 20:09
  SPEAKER_00: So that's that's a use case.
• 20:12
  SPEAKER_04: Uh I would say um any kind of dashboard.
• 20:15
  SPEAKER_04: So just bringing information from different software packages that you have, maybe it's even just an Excel spreadsheet, a CSV that you exported, and stick it in a folder, open up Cloud Cowork or ChatGPT work, and say, given the information in this folder that you can see, there could be seven spreadsheets in there and two PDFs.
• 20:34
  SPEAKER_04: What kind of dashboard do you think we should build?
• 20:37
  SPEAKER_04: Now build it.
• 20:38
  SPEAKER_04: And I think that's probably the easiest way to see value.
• 20:44
  SPEAKER_03: Uh similar along the similar lines, one of the things I did that took zero engineering of any kind was just take our company personnel documents, health plan benefits, et cetera, put all that stuff on a website that's you have to log into, and just make it accessible to our employees, or code of conduct, et cetera.
• 21:01
  SPEAKER_03: Um, all that kind of HR stuff.
• 21:03
  SPEAKER_03: It's already documents, we already have these, but they're strewn about.
• 21:06
  SPEAKER_03: No one actually knows how to find them or read them.
• 21:08
  SPEAKER_03: Um and that took nothing.
• 21:09
  SPEAKER_03: I just dumped a couple of PDFs in, made a little dashboard, and now all of our employees see our values, our code of conduct, if they need to look into their health plan, et cetera.
• 21:18
  SPEAKER_03: Very simple.
• 21:20
  SPEAKER_05: I hate uh doing takeovers and looking at people other people doing takeovers because I feel like it's a crazy waste of time.
• 21:27
  SPEAKER_05: So the the greatest time saving that doesn't require anything, any kind of building, you just uh linear lighting is uh the most terrible thing to calculate on a plan.
• 21:40
  SPEAKER_05: So I don't never do this again.
• 21:42
  SPEAKER_05: Nobody does it this uh in our office anymore.
• 21:46
  SPEAKER_05: We just load the plan and we just ask how many linear footage do we need?
• 21:51
  SPEAKER_05: Uh and it just does it for you.
• 21:53
  SPEAKER_05: And you can use that the PDF reading of the plan is so now.
• 22:00
  SPEAKER_05: Native to all of the large models that you can ask any questions, how many light switches in there, how many zones.
• 22:07
  SPEAKER_05: If the zoning is not on the plan, you can ask to give it like a best approximation of zoning for the lighting, and it will just do it.
• 22:15
  SPEAKER_05: Like literally doesn't all you have to say how many switches on this plan, and it will give you the answer in like five minutes, and it will save you hours and hours of work.
• 22:29
  SPEAKER_01: So we have about 10 minutes left, a little under 10 minutes left.
• 22:33
  SPEAKER_01: Uh I can keep asking these guys plenty of questions, but I feel like you all probably have some questions that you might want to ask them.
• 22:40
  SPEAKER_01: So I think now would be a good time to open the floor up to uh questions from the audience.
• 22:46
  SPEAKER_01: So um I don't know how the logistics of this are gonna work.
• 22:50
  SPEAKER_01: I would just say raise your hand, and whoever I see first, I'll bring the mic over to them.
• 22:59
  SPEAKER_06: Thank you.
• 23:00
  SPEAKER_06: So you're using Chat GPT, Claude, or whatever.
• 23:03
  SPEAKER_06: Have you ever heard pushback from your customers or clients about their data getting put into those models?
• 23:10
  SPEAKER_06: And if so, have you ever thought about pushing more things to local, more secure uh models?
• 23:17
  SPEAKER_06: Things like that.
• 23:20
  SPEAKER_03: Yeah, I've heard a few people mention something similar to this.
• 23:23
  SPEAKER_03: We work on and projects that are NDA, et cetera, et cetera.
• 23:26
  SPEAKER_03: Um, and I don't want to dump that data in there.
• 23:28
  SPEAKER_03: And I think it's a it's a very valid concern.
• 23:31
  SPEAKER_03: I think it depends on your business, right?
• 23:33
  SPEAKER_03: Some of us do more of that than others.
• 23:34
  SPEAKER_03: Um people might be more sensitive to risk there.
• 23:37
  SPEAKER_03: Um you already put their data into QuickBooks and you know, maybe the other software providers, DTools and portals.
• 23:44
  SPEAKER_03: So like that data is already being exported in ways that I'm sure guaranteed, you know, violate that NDA, um, even though that's obviously not intentional.
• 23:51
  SPEAKER_03: But um so I personally don't worry about it a huge amount the way that we use it, but I do understand that there are um, you know, your business may have more of those types of projects and they may be maybe more sensitive.
• 24:05
  SPEAKER_03: Usually those are, you know, that that actual person's information is sort of obfuscated in some way, um, you know, through an LLC or something like that.
• 24:12
  SPEAKER_03: Um so the name of that LLC is public, right?
• 24:16
  SPEAKER_03: But um yeah, I think it's valid, and I think there are certain documents that I you wouldn't um want to dump in blindly.
• 24:22
  SPEAKER_03: Um any type of plan set, et cetera, is already anonymized to the extent that that makes that customer happy with it.
• 24:29
  SPEAKER_03: And there's you know reasonable use um you know uh um uh of that document, you know, you're able to analyze it and share it with people, et cetera, to try to get the job done.
• 24:38
  SPEAKER_03: But I don't personally spend a lot of time thinking about it, but it's a valid concern.
• 24:42
  SPEAKER_03: I totally understand.
• 24:44
  SPEAKER_03: And I think that yes, if your business does a lot of those types of NDA projects, you may want to par back or crop out that information out of that PDF before you dump it in.
• 24:54
  SPEAKER_05: Just uh just two sentence on that.
• 24:57
  SPEAKER_05: Um ultimate goal of like experimenting with AI is to get a software written.
• 25:03
  SPEAKER_05: And once you write the software that does it without use of tokens, you now localized all the processing to basically uh uh a Python script or whatever that will actually do the things that you need to do.
• 25:17
  SPEAKER_05: So experiment with any anything that uh uh uh that doesn't have client info, but uh long-term output from that is actually a piece of software that doesn't uh pull LLM into the process at all.
• 25:34
  SPEAKER_04: I would I would agree with Serge on that.
• 25:35
  SPEAKER_04: I I always tell our employees I I'm comfortable with us uh using Anthropic to build tools that our business runs on, but not running our business on the Anthropic.
• 25:45
  SPEAKER_04: Um so spend the tokens to build the software that doesn't need to do API calls, but also your point about local models is very valid as well.
• 25:54
  SPEAKER_04: Um just two days ago, Perplexity announced a really interesting hybrid solution where it detects sensitive data and routes it to a local model running on a Mac and then sends the other information out to a frontier model in the cloud.
• 26:06
  SPEAKER_04: So I think that could be a great solution as well.
• 26:09
  SPEAKER_05: And that's why we're like talking to Ari because we just learned something new.
• 26:21
  SPEAKER_02: So I'd like to know a little bit more if you're building an app or you're building a website or something like that.
• 26:27
  SPEAKER_02: I've I've gone through this a little bit, but the question is how do you protect that?
• 26:32
  SPEAKER_02: I mean, they they're gonna host this app, and I've I've built you know temporary apps that last a day or two, that's all I needed them for, that's fine.
• 26:39
  SPEAKER_02: But if I build a complete flow through in my business that's not based really on a spreadsheet, let's say, how do you protect that?
• 26:49
  SPEAKER_04: Protect it uh from someone copying it?
• 26:52
  SPEAKER_02: Or uh I mean, how is it hosted?
• 26:58
  SPEAKER_02: How are you how are you making this your own software and kind of having ownership over it?
• 27:03
  SPEAKER_04: So there's a lot of different solutions to deploy apps.
• 27:05
  SPEAKER_04: The one that I think is most popular is Vercel, V-E-R-C-E-L.
• 27:10
  SPEAKER_04: Um you don't really need to know how to use it, um, but you can deploy it just AI knows.
• 27:15
  SPEAKER_04: It's a connector for it.
• 27:17
  SPEAKER_04: And um it'll deploy your app and I think it's ten dollars a month uh for it to continue to host, and then you can create your own URL for it and you own it.
• 27:26
  SPEAKER_04: So sorry, specifically I would say if you're kind of graduating from a basic dashboard into a full app, then you need sort of a back end.
• 27:37
  SPEAKER_04: And GitHub to store the code, some type of database to store data that gets entered into the app, we use Supa base as that, and then a way to deploy it, and we use Vercel for that.
• 27:50
  SPEAKER_00: And with GitHub, just make sure it's a private repository.
• 27:55
  SPEAKER_03: Yeah, and the I think the biggest thing to this is you know if you're running your whole business on a platform, which we are, um somebody could log in and you know, obviously do damage there.
• 28:05
  SPEAKER_03: So not only is your authentication very important, right?
• 28:08
  SPEAKER_03: Your user needs to sign in and you may have them uh do two-factor authentication, et cetera.
• 28:12
  SPEAKER_03: And Rasell and get and um Superbase have methods for this.
• 28:16
  SPEAKER_03: You can use Google as an OAuth uh as well.
• 28:18
  SPEAKER_03: Um and then also include a terms of use with it too.
• 28:22
  SPEAKER_03: So protect yourself not just physically from someone accessing it, but also when you give this app to your employees, they are clicking and accepting a terms of use saying, I'm gonna use this for authorized company purposes and I'm not gonna go delete all your projects, et cetera.
• 28:37
  SPEAKER_03: Um you can also certainly have your platform duplicate and back up that data.
• 28:42
  SPEAKER_03: So if something were to happen, um, you know, you would have a way to restore it.
• 28:46
  SPEAKER_04: Yeah, if if people aren't aware, there's there's a concept called MCP, model context protocol.
• 28:52
  SPEAKER_04: These are the connectors between a large language model and software tools, and that means that you can create, you you just connect GitHub, Superbase, and Vercel to Cloud or ChatGPT, and you now have a full stack application software machine.
• 29:08
  SPEAKER_00: You do not need to know how to do that.
• 29:12
  SPEAKER_04: You just ask your LLM to do it.
• 29:16
  SPEAKER_01: All right.
• 29:17
  SPEAKER_01: Well, we are unfortunately at time, and Mark unfortunately needs to start running.
• 29:22
  SPEAKER_01: However, um I think some of our panelists are gonna stick around.
• 29:26
  SPEAKER_01: Um, maybe not so much uh talk on the mics.
• 29:29
  SPEAKER_01: You can come up to them and ask.
• 29:30
  SPEAKER_01: We do have to make way for the uh the next smart stage session that is going to come up.
• 29:34
  SPEAKER_01: But uh, you know, thank you everybody for coming.