research-lab-canada-az1-ai-entrepreneurs-mike-smith-hal-casteel-2026-01-15-18-58-est-notes-by-gemini-1

Jan 15, 2026

Canada AZ1 - AI Entrepreneurs Mike Smith/Hal Casteel - Transcript

00:00:00

 
Hal Casteel: Hey there.
Mike Smith: Hey,
Hal Casteel: Is it That's not for real. The snow, is it? Is it snowing up there?
Mike Smith: it's we we got we got we got a hammer today.
Hal Casteel: Oh, did you really?
Mike Smith: Yes. So,
Hal Casteel: So, you're feeling snowy, huh?
Mike Smith: so I I think I I've already I've already went outside and shoveled the driveway I think uh like three times.
Hal Casteel: three times in one
Mike Smith: Yeah.
Hal Casteel: day.
Mike Smith: Like I think I think literally literally we've got probably 16 inches of snow just
Hal Casteel: Wow.
Mike Smith: overnight.
Hal Casteel: Is And is there more sitting on the ground already?
Mike Smith: Oh yeah. Yeah, there's more on the ground.
Hal Casteel: Yeah. Well, well, I won't try to make you too jealous.
Mike Smith: So,
Hal Casteel: It's what I think right now it's 25° centigrade and I just turned on the air conditioner cuz I
Mike Smith: well, you're Yeah, this is this is this is your summer down there.
Hal Casteel: Yeah, exactly.
 
 

00:03:04

 
Mike Smith: So,
Hal Casteel: Hey Jared, is that your number one?
Mike Smith: Yep. That's uh that's my number one son, Jared.
Hal Casteel: Great. Good to meet you. I've known your dad for gosh it's I don't know how long it's been seven eight years I think isn't it going on probably yeah and uh where where are you are you in the same neck of the woods as your
Jhared Smith: Uh,
Hal Casteel: dad
Jhared Smith: no. I'm more southwestern Ontario and Woodstock.
Hal Casteel: okay yeah are
Mike Smith: So he's about an hour away. An hour and a bit away.
Hal Casteel: is it you is that close to Lake Superior then or Whereabouts?
Jhared Smith: Uh yeah,
Mike Smith: closer Lake Hiron,
Hal Casteel: Lake you're on.
Mike Smith: isn't it?
Hal Casteel: Okay.
Jhared Smith: I was going to say like here on Yeah.
Hal Casteel: You get Do you get the lake effect type snow then?
Mike Smith: Yeah.
Jhared Smith: Yeah.
Hal Casteel: Yeah.
Mike Smith: Yeah. He's he's he's in the snow.
Hal Casteel: Yeah. Yeah.
 
 

00:03:58

 
Hal Casteel: So, you get even more snow than Yeah.
Mike Smith: So
Hal Casteel: I I lived in the Midwest when I was a kid. I mean,
Jhared Smith: Yeah.
Hal Casteel: my first entrepreneurial experience was buying a snow shovel and getting my neighbors to pay me 10 bucks to shovel their snow for the rest of the winter. And uh I got 30 people to give me 10 bucks. I was six years old and I wanted a pair of metal skis, which back in those days were 20 bucks. I mean, now it'd be hundreds, right? But I bought snow shovels for every kid in the neighborhood. And I've been I've never shoveled any snow. I mean, I'm not But I've been waking up early and watching the weather ever since. For sure.
Jhared Smith: Yeah. Yeah. We got a third 35 cm last night.
Hal Casteel: Wow. Yeah. Yeah.
Mike Smith: That's why I put this background on because it's like it's
Jhared Smith: So,
Hal Casteel: Yeah.
Mike Smith: a Yeah, it's snowing.
 
 

00:04:51

 
Mike Smith: So,
Hal Casteel: Well,
Mike Smith: but I think it was stopped.
Hal Casteel: tell me tell me a little bit about yourself, Jared, if you don't mind. What's your you are you
Mike Smith: So,
Jhared Smith: sure.
Hal Casteel: it type type of guy or what are you doing for a living?
Jhared Smith: Uh well, I'm an IT guy through osmosis of my dad.
Hal Casteel: Yeah,
Jhared Smith: Uh just or orbiting him for for years.
Hal Casteel: for sure. Yeah.
Jhared Smith: Uh I'm I'm an insurance broker.
Hal Casteel: Oh, okay.
Jhared Smith: Um so I'm in the insurance industry,
Hal Casteel: Yeah.
Jhared Smith: been in the industry about six years. Before that I was an accountant. Um so I was an accountant for five years and then switched over to insurance.
Hal Casteel: Uh-huh.
Jhared Smith: Um and been doing that. Um kind of interested to see how I mean there's a lot of regulation when it comes to AI and
Mike Smith: Oh,
Hal Casteel: Looks like Yeah. Look like he froze a little.
Mike Smith: he froze.
 
 

00:05:41

 
Hal Casteel: Did he freeze on your end, Mike, or Yeah.
Mike Smith: Yeah, it froze. Yeah.
Jhared Smith: oh yeah I think it's the the the
Mike Smith: Yeah, you froze a little bit there,
Hal Casteel: Uh yeah.
Jhared Smith: wind here is really bad
Mike Smith: Jerry.
Jhared Smith: so
Hal Casteel: Um so regulation in AI or reg I mean there's obviously a lot of regulation in the insurance business as well, right?
Jhared Smith: yeah there we're actually supposed to be a part of a big campfire chat with our regulation body in February in regards to how we can and can't use
Hal Casteel: Oh,
Jhared Smith: it. We've kind of been instructed at the moment we can't and we were supposed to Yeah,
Hal Casteel: you cannot. Yeah,
Jhared Smith: we're supposed to avoid for now,
Hal Casteel: actually I I just spent the day uh while I was at
Jhared Smith: but
Hal Casteel: I I was at a company called Grail. left Oracle uh four years ago to come to it was a health I've got four decades in healthc care looks like David Chen uh joining
 
 

00:06:38

 
Mike Smith: Y. Yeah, he's the Yeah,
Hal Casteel: too
Mike Smith: he's um here dude.
Hal Casteel: David how you doing Okay.
Mike Smith: I don't know if you can hear
Hal Casteel: Yeah, it looks like maybe any Can you hear us,
Mike Smith: us.
Hal Casteel: Dan? David,
David Q Chen: Sorry, I can't hear anything right now.
Hal Casteel: uh, we can hear you.
David Q Chen: Figure out why it is.
Hal Casteel: Uh, um, check. Are you on a Mac or what? What? Turn up your volume. Sort it out.
Mike Smith: I'm not sure if he's muted.
Hal Casteel: He Yeah,
Mike Smith: I'm not sure.
Hal Casteel: you He's on He is on mute right now.
Mike Smith: He is unmuting. I can't unmute him from here. So,
Hal Casteel: Actually,
Mike Smith: you you have to do it now.
Hal Casteel: I can't. It looks like he's got it mute muted.
David Q Chen: Okay, I figured it out. Hi. Hi,
Hal Casteel: Okay.
David Q Chen: everyone.
Hal Casteel: Yeah,
 
 

00:07:52

 
Mike Smith: Okay.
Hal Casteel: we can hear you. You can hear us.
David Q Chen: Yes. Yes.
Hal Casteel: Okay. Uh I think we'll go ahead and get started.
David Q Chen: Finally.
Hal Casteel: Do you guys mind if I record this session? All right.
Mike Smith: And we might have some other people come in
Hal Casteel: Yeah, that's fine. Yeah, we can just do a quick uh Yeah.
Mike Smith: probably.
Hal Casteel: So, and if you guys are interested in the I I'll do a transcript, too. And if uh you guys are interested, there may be some interesting stuff come up. Um I thought what we'd do is, you know, start with a real quick round robin. David, did we meet before? I think we did, didn't we?
Mike Smith: Yeah. Yeah,
Hal Casteel: Yeah. Yeah.
Mike Smith: we did.
Hal Casteel: You were the CTO working
David Q Chen: Yeah,
Hal Casteel: with
David Q Chen: I was a CEO of Benchai of the startup that I co-founded with
Mike Smith: Yeah.
 
 

00:08:53

 
David Q Chen: Abby.
Hal Casteel: Yeah, exactly. Yeah.
David Q Chen: I think I now I left a while ago.
Hal Casteel: Are you still there now?
David Q Chen: I'm just sort of a semi-retired at this
Hal Casteel: Okay,
David Q Chen: point.
Hal Casteel: one more joining us here right now. So, Mike, welcome aboard. Can you hear us?
Mike Wicks: Okay. Okay.
Hal Casteel: Okay. Uh, I am recording if that's okay with you, just to let you know. Um, so let's go ahead and get started. Um, let's do a real quick round robin intro. I mean, I'm obviously some of you know each other maybe more than I do know you. Uh, and David, even though we've met again, if you can kind of do an update on what your current uh status is and what you're up to and what you're interested in. Um, Mike, everybody knows you, right? So, I don't know if we need to introduce
David Q Chen: Yes.
Hal Casteel: you.
Mike Smith: No. Uh, actually not everybody knows me.
 
 

00:10:00

 
Hal Casteel: Okay.
Mike Smith: So, uh,
Hal Casteel: So why don't you start why don't you start the ball?
Mike Smith: I think I think Mike's coming in.
Hal Casteel: Yeah.
Mike Smith: I think Ross I think Ross.
Hal Casteel: Once you start the ball rolling
Mike Smith: So, yeah. Sure. I'll I'll introduce myself.
Hal Casteel: then
Mike Smith: So, uh, yeah. So, I Mike Smith. So, uh, I've been working with kind of Hal, you know, just kind of kind of back and forth. Um, you know, Hal and I actually worked together at uh Oracle Netswuite. Uh, Hal left about four years ago. Uh, so we were on the same team. Um, I'm now a solution consultant um at at Netswuite. So I basically run all the demos. Um again because of the massive you know uh investment of of Oracle uh AI is just taking off in the in the ERP space. So uh so um I'm looking to learn uh and see more um really just you know from from Hal's perspectives.
 
 

00:10:54

 
Mike Smith: Yeah, I think what Hal's kind of put together is kind of really really cool and uh and just yeah, just to see what kind of ideas are out there to uh hey, you know, maybe we can kind of take some ideas and actually turn them into something.
Hal Casteel: uh I'll go ahead uh real quick. So I when I left uh uh Oracle I was working with Mike I was a what's called a solution principle basically blueprinting businesses and software health tech fintech uh you know kind of in between direct sales and uh delivery uh prior to that you know almost 40 years in health care from uh medical laboratory you know robotic labs healthcare delivery and then cancer research at scripts pharmaceutical startup and then I you know after leaving uh that startup which got acquired by Sophia I started about 20 startups during the dotcom years and then seven years on the beach in Brazil I'm actually in Brazil right now uh came back into the states in 2010 and then started working with Oracle Netswuite prior to meeting Mike But you know more on the on the customer side originally implementing CTO level type of uh handling business is getting onto the
 
 

00:12:20

 
Mike Wicks: That way.
Hal Casteel: platform. uh Capgeemini a few years a couple billion dollar plus revenue uh Oracle customers where I worked you know on the customer side but directly with Oracle and then left for Grail which is a multi-cancer early
Mike Smith: Attention
Hal Casteel: detection test that uh launched a product built $2
Mike Smith: test.
Hal Casteel: billion startup acquired for seven billion dollars and then we started rolling AI out three and a half years ago and I was part of this seminal team that helped assist in that. So um first building a risk management framework working with you know upper management to make them aware of what we and the troops were doing what you know we're everybody was going wild on AI and IP was going out the window and then left uh it's been over about a year and five months uh in September with some of the ideas that I formed while I was at Grail I founded a company that awan.ai AI and we're we're launching our pilot right now. We have a tool called CODATECT ch which I'll demo later towards the uh as we do around
 
 

00:13:38

 
Mike Smith: Okay,
Hal Casteel: Roman.
Mike Smith: Jared, you want to go next? Because just so everybody knows, Jar Jar's my number one son. So
Jhared Smith: Yeah. Uh, yes. My name's Jared SM. I'm gonna apologize now if my network gets a little wonky and we have above ground lines here and the wind here is crazy. So, when that happens, our internet goes a little weird. Um, but yeah, so I'm Jared Smith. I'm Mike Smith oldest son. Uh, I come from a background of finance and insurance. I was an accountant for years before actually I started seeing AI shrinking that field. Um, which then led me to move over to insurance. I've been in sales and servicing insurance, uh, specializing in sales over the last six years.
Mike Smith: Thank
Jhared Smith: Um, and the more I see AI,
Mike Smith: you.
Jhared Smith: the more I become interested in wanting to see, especially how it can affect the financial insurance sector. Um, but just in general, just curious to see um, and kind of get ahead ahead of the curve and see what it's capable of.
 
 

00:14:30

 
Mike Wicks: Turn it off.
Jhared Smith: So,
Mike Smith: All right,
Hal Casteel: Abby, thank Abby.
Mike Smith: David.
Hal Casteel: Thanks for joining.
Mike Smith: Yeah, I think I think Abby's he said he's he's I think he's got a he's got a little one,
Abhi Pushparaj: Yeah,
Mike Smith: so he might for might be
Abhi Pushparaj: apologies folks. Uh yeah,
Mike Smith: on
Abhi Pushparaj: I uh just got a baby over here uh dealing with. But uh uh glad to be able to join and uh hear the conversation.
Hal Casteel: Yeah, thanks for joining. We're just doing a round robin of introduction, so your timing's pretty good. So, uh, David, do you want to go?
David Q Chen: Yeah. So I'm David. Uh previously I founded a AI search engine um company on antibodies with Abby. Um that I stayed with that company for about five six years. Uh and then you know during 2020 2021 with COVID that's when I thought maybe it was a good time to take a break. Um so yeah I'm still here taking my own break.
 
 

00:15:34

 
David Q Chen: I'm sort of looking for the next um I guess it's a period of you always take a rest and then look before you're looking for the next thing. So I'm sort of around looking for the next thing I'm looking uh I like to explore. Also recently I've been working with Abby um on looking at I have a background in neuroscience. So looking at uh this particular compact called ebook game and doing some neuroscience uh re scientific research and publications and from that angle.
Hal Casteel: interesting. Yeah. Uh,
Mike Wicks: Yeah, I I'll jump in. It's It's um Yeah,
Hal Casteel: Mike.
Mike Wicks: it's uh it's actually uh Ross. I'm using a a company um company email, so I don't know why it keeps coming up as Mike Wick, but um yeah,
Hal Casteel: Oh, I'm
Mike Wicks: I I was on that first call,
Mike Smith: I mean,
Hal Casteel: sorry.
Mike Smith: you look you look familiar. So, I didn't I didn't I didn't know I came out of mic.
Mike Wicks: but
 
 

00:16:28

 
Mike Smith: I'm like, "Okay, well that's okay." I think because I think you were you wanted to you invite somebody yourself. So,
Mike Wicks: yeah. Yeah, I um I met Mike last summer and I joined the initial call with with
Mike Smith: okay.
Mike Wicks: uh with Hal and Mike. I think David was on there too. Uh my background it's it mainly on the sales side of uh ERP solutions.
Hal Casteel: Yeah.
Mike Smith: Yeah.
Mike Wicks: So I've worked with uh bigger companies like Alysia on the Microsoft side and uh typically it was discrete manufacturing but I uh I always ended up
Hal Casteel: Yeah.
Mike Wicks: selling uh food processing for whatever reason. Uh but that's a bit of my background. Uh right now I I am working for a smaller uh partner uh strictly just data and uh data and uh business intelligence tools. So a lot of it's kind of advanced analytics building uh building out reporting for companies. So little different from the the ERP world but uh that's that's what I'm doing now.
 
 

00:17:33

 
Hal Casteel: And I mean it looks like Jared I think the only per are you using AI in any way personally these days? I'm sure Dave Yeah.
Jhared Smith: person. Yeah. Personally, yes.
Hal Casteel: Yeah.
Jhared Smith: Y
Hal Casteel: Absolutely. Which which tools are you tend to use the most or you use some or all or what
Jhared Smith: yeah.
Hal Casteel: what's your experience?
Jhared Smith: But um it's like I mean maybe it's like chat GPT is probably the one I use most predominantly
Mike Wicks: I don't want it.
Jhared Smith: just for researching stuff quickly. Um you know that's the one I probably use the most, but I've kind of kind of been dabbling in some of the few. Dad's been sending me some ones to to play around in. So, um, things have been a little bit busy lately, so I'm hoping, uh, in the new year things open up and I get a little more time to play around in some of these suites and figure out some stuff.
Hal Casteel: Yeah.
Jhared Smith: So,
Hal Casteel: And David and Obby, I'm sure you you guys are deep in whatever you're doing.
 
 

00:18:29

 
Hal Casteel: What kind of tools are you guys using the most or
Mike Smith: Watch
Hal Casteel: typically
David Q Chen: I mostly use well personally I if I'm coding stuff or side projects and do
Mike Smith: two.
David Q Chen: that I use uh all the cloud code with cursor and then
Hal Casteel: are using cursor with
David Q Chen: also I do both I I usually use
Hal Casteel: uh
David Q Chen: cursor and claw but then if I have something there's also the cloud code itself uh I use as a site tool as well by itself um I also use grock mostly if I need to do research
Hal Casteel: Uh-huh. Yeah. Interesting. How about you, Obby?
Abhi Pushparaj: Yeah, I would say predominantly uh Grock and Claude.
Hal Casteel: Yeah. Okay. Um, for me the I mean I've How How many people are are using Perplexity for example for research? Anybody?
Mike Smith: I use
Mike Wicks: uh yeah, I'll use it just for um like uh tweaking tweaking messages.
Mike Smith: it.
Mike Wicks: I I find it's a little bit better than than chat GBC, but
 
 

00:19:36

 
Hal Casteel: Uh-huh. How about you, David? With research or anything, do you ever start on perplexity or anything of that nature?
David Q Chen: Not really. Uh mostly because yeah I subscribe to Brock so I I rely mostly on uh on that.
Hal Casteel: Yeah. And Grock is really good at research.
David Q Chen: Yeah.
Hal Casteel: I find the same. Yeah. Um my predominant tool is claude code you know running inside of uh VS code with you know with basically code attack which is my agentic platform uh which kind of you know it hijacks cloud code and you know basically using the framework of anthropic skills uh their model basically Basically, David, you're probably pretty familiar with how they've they built out their platform. Have you built your own agents, skills, commands, or any of that kind of level out onto the platform at
David Q Chen: I I think I experimented with a little bit.
Hal Casteel: all?
David Q Chen: Um, but I never really got deep into agentic workflow
Hal Casteel: Okay.
David Q Chen: yet.
 
 

00:20:47

 
Hal Casteel: All right.
Mike Wicks: All
Hal Casteel: Uh, how about you, Abby? Are you using Agentic Ace workflows at all yet
Mike Wicks: right.
Abhi Pushparaj: No. Uh,
Hal Casteel: or?
Abhi Pushparaj: haven't got into that myself yet. Primarily working with Grock for research. Um, but curious,
Hal Casteel: Okay.
Abhi Pushparaj: very curious to see what you've got going.
Hal Casteel: Okay.
Abhi Pushparaj: Uh, no.
Hal Casteel: Um and then so from a kind of a general interest point of view, I mean my everything I'm going to share is I mean it's not public, it's private, but you know my goal with, you know, with Mike all along has been to, you know, help accelerate people's learning as well as help anybody that's got either entrepreneurial or personal projects or, you know, or wants to experiment. I think it's all, you know, all of the above. So everything I'm going to share, I'm willing to, you know, give people access to what the platform I've gotten. It's I'm I'm rolling out a pilot right now with a number of selected uh people.
 
 

00:21:46

 
Hal Casteel: So I'm looking for people who do who do use the platform uh to give me feedback on how to improve it and uh and what their experience is because you know whether it adds I mean it's an amazing time of change, right?
Mike Wicks: Stop.
Hal Casteel: Every day is something new and it's coming fast and furious. uh co-work has anybody uh played with co-work yet? Are you familiar with Anthropic's latest uh what they've done is okay co-work is anthropic uh basically running a version of
Mike Smith: No.
Hal Casteel: uh claude code in a in a kind of a I would say a paired down Linux sandbox that gives you some security probably not a lot. Uh and then they recommend you create a folder and then put you know if you wanted to take a bunch
Mike Wicks: Sure.
Hal Casteel: of screenshots of uh expense reports for example it'll build a you know a uh a spreadsheet for example it'll do a it can do a whole lot of things. It's it's a really interesting kind of evolution of taking cloud code more to to the uh public level where maybe people aren't so sophisticated but can can basically automate.
 
 

00:23:02

 
Hal Casteel: It's probably the next generation of what where anthropic is going which will probably move beyond just enterprise but uh they're also beginning to do some stuff that looks like vendor lockin uh hasn't affected me yet but I can imagine you know everybody's trying to build a moat around what they're doing. I'm kind of anti-mote, meaning that I want to work with all the LLM providers, you know, as agnostically as I'm able, but I've really built my platform on top of cloud code at this point just because they had the better coding engine and the LLM seem better, easier to program. So um let me share real quick and I'll kind of take you into what I am doing it. Uh those of you who are not familiar with in integrated uh development environments, David and Obby certainly are. Mike, I don't know if you've used any your yourself, but this kind of give you an idea of um and you guys can see my screen, right?
Mike Smith: Yep.
Hal Casteel: Okay. Um so what you are looking at for those of you that aren't familiar with you know what this you know allows you to basically open an
 
 

00:24:24

 
Mike Wicks: What?
Hal Casteel: integrated development environment in any folder. So this is inside uh CODATECT ch itself which is made up of right now it's a it's I think it's 84 GitHub repos uh and the
Mike Smith: What
Hal Casteel: organized here what I call
Mike Smith: happened?
Hal Casteel: core is kind of the brain of the operation so it's what what it does is it has lots and lots of agents. These are specialized agents that do a wide variety, you know, for example, research, uh, you know, cloud architects, you name it. There's a as the system has identified gaps, it is autonomous and it'll build out additional combination of what are called skills which are skills are similar to what uh most people think of in terms of prompts. skills uh do very specific things that are geared around one's an orchestrator for example which you know can control many agents uh you know you may want to have you know loop analysis which means you can control some of this is very technical uh but they look they're they're easy to read so if you're looking at you know this is a this is an agent for example.
 
 

00:26:04

 
Hal Casteel: So when you define an agent, you define what its primary function is, what its purpose is, you know, kind of what it how it will do what it does, not at the skill level, but then it'll, you know, can reach out and use a variety of different uh tools. Some of these are combination of uh uh anthropics tools as well as some that I've built. There are many there's over 2,000 components in CODATECT ch and you know including scripts. The the idea is not to just use an LLM to hit a hammer you know to hammer everything but to use programs in a controlled way to really direct the LLM to do exactly what it is. So, you know, I I I don't have problems with hallucination generally. Uh it'll break things down into workflows. So, when you you know, here's here's a skill for example. Uh skill uh you know, define when you're going to use it,
Mike Wicks: Go.
Hal Casteel: what it may what its particular focus is in that skill set. And they can be versioned I you know so they evolve as the system runs they they can be self-improving you know what you expect what you don't expect you know when you shouldn't use it and what patterns to avoid so uh there's a couple hundred skills in CODATECT so the combination of the agent the skills and then a combination of scripts and scripts are kind of the you know
 
 

00:28:00

 
Hal Casteel: the in a programmatic fashion. It's you know neuros symbolic programming and this and the scripts are once again very specific around particular types of uh they may be they can be JSON they might be Python they might be shell scripts they may be a wide variety of different things depending on the activity of what it is that we're we're trying to control. So you know here So, is am I getting too technical? I mean, I know I have a kind of a a variety of Are there any questions so
David Q Chen: How how did you boost traffic?
Hal Casteel: far?
Mike Wicks: Oh, right there.
David Q Chen: Like did you did you did you write it or did you ask the clock hoster agent to write all these?
Hal Casteel: Uh, I seated it. I mean, I started development back in March of last year.
Mike Wicks: All
Hal Casteel: uh it's in you know basically the seventh version and I mean basically code of tech has built itself at you know since August 27th a lot of the code was uh you know built on claude code early on and then you know as kind of code techch evolved you know it's all
 
 

00:29:19

 
Mike Wicks: right. And then
Hal Casteel: run through uh code tech on top of cloud code right now so if you think of it as a kind of a symbiotic uh relationship. You know, there's now the one thing that I have done that's unique to uh CODATECT in particular, let me show you.
Mike Wicks: Excellent.
Hal Casteel: Uh I have the ability to export and manage memory. Um, so one of the challenges in uh let me just get out of this here real quick. So you can run multiple uh sessions in parallel. down here you can see you know where I am, what GitHub repo I'm on, how much of the context is used at this particular point of view uh and and a few other pieces of information. Now in in the anthropic world you have this ability to do what what are called slash commands. Uh are you familiar with slash commands David?
David Q Chen: I think
Hal Casteel: Yeah.
David Q Chen: I
Hal Casteel: So you know you type ahead and let's say I I want to export the context for this particular
 
 

00:30:42

 
Mike Wicks: Yeah.
Mike Smith: Hey.
Hal Casteel: session.
Mike Smith: Uh how sorry to interrupt. I think uh Attilla is looking to get
Hal Casteel: Oh okay. Let me get get back to a window where I can see
Mike Smith: in.
Mike Wicks: f***.
Hal Casteel: here. See here. This is it. All right.
Mike Smith: all in good. Hey there, Aillaa.
Hal Casteel: Is this is this is this your brother
Mike Smith: Sorry. No,
Hal Casteel: too?
Mike Smith: this is a riding partner in the summertime.
Hal Casteel: Oh,
Mike Smith: Although he he rides all winter,
Hal Casteel: okay.
Mike Smith: so Oh,
Hal Casteel: Why don't you ask him to introduce himself real quick?
Mike Smith: yeah. Hey, Tilla. Do you want to introduce yourself just really quickly?
Dunnville Grand Tour: Absolutely. Um so my name is Otila and uh last time I visit joined you guys as well and um so basically I'm an aid cyclist running uh cycling tour business and I am um deeply involved with ERP implementations for uh logistics and warehouse operations.
 
 

00:32:17

 
Dunnville Grand Tour: uh and u AI uh really strict my interest uh the last few months and uh since our last meeting actually I started digging into deeper and deeper and now I'm using for from marketing to my I'm I totally switched over how I designed jerseys and everything and my clothing lines it's all AI now my marketing is all AI so it just opened a brand new world for me so I'm really interested
Hal Casteel: Wow.
Dunnville Grand Tour: to taking A deeper
Hal Casteel: Yeah. So, I was walking through just real briefly uh Koditech,
Dunnville Grand Tour: end.
Hal Casteel: which is kind of the agentic uh AI platform that I built. It my original thing. It was originally because coming out of a reg regulated industries like food distribution development you know certainly uh medical devices or you know medical applications uh insurance he anything that's heavily regulated uh
Mike Smith: Stop.
Hal Casteel: you know so I I I wanted to develop a platform that kind of took regulation and
Mike Smith: s***.
Hal Casteel: compliance at the root uh so the idea that you know building up from
 
 

00:33:27

 
Mike Wicks: The idea.
Hal Casteel: the idea that you know AI is a tool and like any tool uh you know while it may have utility it's also fraught with risk and you know may may need uh more than guardrails and really understanding the either the legal or legal requirements which vary by country. Uh sometimes I mean they're talking about regulating at the state level in the United States which we're trying to block at the federal level. uh the you know Europe has their own set of rules. uh so I built things like riskmanagement frameworks into the into the platform but I was showing where the tools and the how but at the core you know can be any set of you know either regul regulation or it could be policies uh that companies use internally could be at the departmental level. uh it's designed as a multi-user multi-tenant meaning multiple businesses with many users and multiple teams and you know multiple projects running in parallel on the platform as well meaning because when you work in an agentic system while I may be working individually on a project uh I may have co-workers or consultants or auditors also
 
 

00:34:50

 
Mike Wicks: All
Hal Casteel: uh going through uh either using the platform itself for as a
Mike Wicks: right.
Hal Casteel: tool or they may be you know if we're developing software say for a medical device company and any change anything that goes into uh a medical device company including the software has a procurement life cycle and in the United States the FDA regulates that and it's you know we may need to track you know using digital signatures every aspect of everything that's procured and then anything that changes in the life cycle
Mike Wicks: Sure.
Hal Casteel: of and David I'm kind of expecting you guys have had some of those experiences is that a good guess or you know you know you've developed things that need to be monitored and are regulated is that true or am I just reading into but I think you
Mike Wicks: All
Abhi Pushparaj: Yeah,
Mike Wicks: right.
Hal Casteel: guys
Abhi Pushparaj: I I work with several clients uh trying to uh incorporate machine learning or AI aspects into their their devices. Um and yeah, it is I mean even FDA's regulation is still evolving.
 
 

00:36:06

 
Hal Casteel: It is.
Abhi Pushparaj: We haven't gotten to anything that's uh we're we're still in the state where FDA is
Hal Casteel: Yeah.
Abhi Pushparaj: requiring things to be you can develop them with these types of tools, but then they require things to be frozen before they're they're put out commercially.
Hal Casteel: Exactly. Yeah. So I mean I'll just give you an example of you know this is a a riskmanagement framework that's built on top of the NIST as well as the GDPR uh kind of requirements uh that are you know kind of geared around you know these kinds of things. So, you know, what what regulatory framework are you dealing with? And this is a fast I mean, I wouldn't say it's evolving rapidly. There haven't been a lot of changes right now, but I think I'm kind of expecting and you're probably expecting this year or certainly by next year that some of these things are going to be more nailed down,
Mike Wicks: Okay.
Hal Casteel: right? Isn't that kind of what we're expecting?
Abhi Pushparaj: Yeah,
 
 

00:37:07

 
Hal Casteel: Is it the
Abhi Pushparaj: that's the that's the mandate um coming down that FDA is
Hal Casteel: Yeah.
Abhi Pushparaj: supposed to be putting out a lot more firm guidance in this year around this Yes.
Hal Casteel: Right. So, so what I do as the mandates, you know, as the frameworks evolve, you know, then I update and incorporate the actual changes that are, you know, I mean, this is more geared towards the EU, but there's, you know, I mean, every every uh country I mean, I think it'll largely be split between North America and the EU initially.
Mike Wicks: Where's the mission?
Hal Casteel: I'm I'm not sure where Canada is in the evolution right now. Do you guys know that where they're at in their frameworks? Are they going to adopt what we adopt in the US or do we know?
Mike Wicks: Not.
Mike Smith: No idea.
Mike Wicks: Yeah, I assume they're they're similar in
Hal Casteel: Yeah. So, so my thinking was is that you know because dealing in an agentic network where you know
Mike Wicks: nature.
Hal Casteel: you're you know you're I mean essentially you're managing agents the way you would enjoy you know manage employees.
 
 

00:38:19

 
Hal Casteel: So I have a variet sharing here real quick again. In fact I'm not sharing am I?
Mike Wicks: All right,
Hal Casteel: I thought I was sharing.
Mike Smith: No, you're not sure.
Mike Wicks: perfect.
Hal Casteel: Am I sharing?
Mike Smith: I was going to ask you.
Hal Casteel: Okay. Yeah, I think I was under the impression I was sharing. I wanted to share what I was showing.
Mike Smith: Yeah.
Hal Casteel: Um, so yeah. Okay, let me get back to the So this is a a riskmanagement framework built you know basically on
Mike Smith: Sorry.
Hal Casteel: the foundation of the NIST which is the National Institute of Standards and Technology in the United States which you know is part of the Department of Commerce in the United States. that kind of writes the book on the you know technology uh guard rails around things like that and the EU has their own compliance that is quite different really than than what is happening in the United States. So I I write a you know I say a core uh framework that then will evolve as the as the regulations evolve.
 
 

00:39:38

 
Hal Casteel: I mean you know they need to stay in sync with the time but this will provide you know say let's say you had an either a new business or an existing business that needed a riskmanagement framework. You know what does that mean? Well, you you start with things like governance,
Mike Wicks: Yeah.
Hal Casteel: right? You know, and what is governance? It's it's you know, basically, you know,
Mike Wicks: All right.
Hal Casteel: how the business is going to manage how AI is going to be managed across the organization. So, no business should use AI without having a you know, basically a riskmanagement framework in place. I mean, and I would argue even as a startup, you should put one in right at the very very beginning. And you know, then that goes into what typically is a charter. A charter is what guides the safe and responsible adoption or not of the AI across the org,
Mike Wicks: Huh? Heat.
Hal Casteel: whether they're used by vendors or whether they're developed internally. And you know the the main goal there is once again compliance and that both within the organization along with any overarching rules or policies that may need to be enforced.
 
 

00:41:00

 
Hal Casteel: And you know there's a process that goes into assessing risk you know and how to assess risk you know and that there's a process in terms of you know how you document that and manage those types of things. So you know while this is just one aspect of CODATECT ch you know if you ever had a
Mike Wicks: What?
Hal Casteel: business that needs a framework this is a good place to start you know you can d you know this is part of the platform and then you can you know kind of customize it to fit fit the business by working with your teams internally and you document everything as you go and you build it out as a you know as a as a set of documentation internally as well as policies and processes. that will you know govern how the business will operate internally. Does that make sense everybody? This is really essentual I think and you know and then there's things like you know
Mike Wicks: Yeah.
Dunnville Grand Tour: Yeah, this is very good.
Hal Casteel: how do you explain it to management right you know because you you know your executive
 
 

00:42:05

 
Mike Wicks: Too bad.
Hal Casteel: team this is the framework of documents themselves and the executive summary you know geared towards you know leadership understanding. So part of you know the job of anybody working w with AI is to inform the people you're working with both as either internal team members
Mike Smith: Yes. That's right.
Hal Casteel: or uh you know and it you know like anything uh business-wise you know you really you begin with policies but then there's training as well right you know and how do you train people to get to be aligned and then you may
Mike Wicks: Bye.
Hal Casteel: even need to to you know board presentations to explain at a higher level because they'll be asking you if you haven't prepared them. a a a smart board is going to be wanting to know what their risk and what you know what the what's happening inside you know so organizing you know committees and boards and understanding that so you can communicate and automate safely within guard rails and I think you know you were talking about the insurance industry I'd be really interesting from what you hear uh in that presentation that you're going to be attending.
 
 

00:43:28

 
Hal Casteel: You said in February, right?
Jhared Smith: Yeah, they I think it's scheduled for
Hal Casteel: Yeah.
Jhared Smith: midFebruary.
Hal Casteel: So, you know, industry is beginning to wake up. Uh so, it's going to be coming at us from all different types of uh uh uh directions. So, it's better to be proactive than reactive. you know, if you come if you come in informed, you know, ahead of the curve, you know, they're going to look to you for, you know, guidance and leadership. And that puts you into also a position to help the organization be safely adopting AI rather than, you know, issues arising and then leadership saying hell no. And then your competitors who may be adopting AI safely actually then move ahead rapidly and and could actually do serious damage to your competitive advantage because AI may or may not add value and how you want to add value to the organization. So these are things right here that I highly you should start from the you know never use personal accounts don't use brow AI browser extensions that are you know not approved it should start with a process where the organization uh you know informs itself about its risk profile and then smartly moves into addressing that
 
 

00:44:50

 
Mike Smith: Smart.
Hal Casteel: and managing that and then avoiding these types of things where individuals may be operating uh within the organization and then go through a process where you actually request large organizations like Oracle you have to go through very rigorous vetting process for you know what may be new adoption of any new uh tools right it's not something we can just use whatever we want.
Mike Wicks: process.
Hal Casteel: And then you may need to, you know, begin to process and develop your own internal audit process, which means that you document all of this stuff from day one,
Mike Wicks: Which one should
Hal Casteel: which gives you that framework. So later on, you know, if you do get audited, you're prepared. And how to be prepared. Make sense? So that I I you know and I'm willing to share this with any any of you uh freely.
Mike Wicks: Yeah.
Hal Casteel: So you know if if you want to just be more informed if you you know and it's a good place to start you know you could do your own you know googling or AI interactions with the documents and you know you'll you'll find that you know this does kind of meet the standard of required at this point in time.
 
 

00:46:28

 
Mike Wicks: with with my current company. Um, we we have tried to target law firms and uh quite often we'll do like a a proof of value. Uh sometimes it's just like basic co-pilot agents and things like that. So the directors are always on board but as soon as you know we we try to move forward uh FITO comes in and just halts the the project and uh it's it's typically lack of information and like like you went through uh just now like that's a lot of companies lack that ability to show um the the guard rails and everything else. So
Hal Casteel: Yeah. Well, that that was really where this uh Looks like somebody else maybe wants to join here. Uh I think it's another one of your sons there, Mike. Jacob.
Mike Smith: Yep, that's Jacob.
Hal Casteel: Okay. Yeah.
Mike Smith: That's number
Hal Casteel: So this is this is actually what started uh CODATECT ch or you know my founding
Mike Smith: two.
Hal Casteel: of this company that I'm involved in now because we didn't have that at Grail and you know we were you know we were I had two ISO certifications for an ERP and a procurement system and we were going through FDA approval for you know a a medical test which is basically falls under the medical device regulations from the CFR point of view in the United States.
 
 

00:47:54

 
Mike Wicks: Sorry.
Hal Casteel: So that really what we found was senior leadership was not very aware executive leadership was totally in the dark basically. uh now I would think this was a three years ago that we started and I don't know industry by industry it's going to wildly vary and I think business by business you know depending on maturity you know how you know how the you know does the organization have you know what your chief information security officer if they have those u structures already into their executive leadership they're probably already addressing this But but Mike to help overcome some or not I'm sorry calling you Mike uh the to overcome those objections you know if you share a framework that they can
Mike Smith: Yeah.
Hal Casteel: use uh to help that process get going right because people say no out of fear which is smart right that's a good thing uh honestly but it can also you know really stall out initiatives or or you know good ideas may get shot down uh that are bubbling up through the organization you know rather than you know really taking you know a strategic point of view of the opportunity that AI is presenting organizationally and
 
 

00:49:22

 
Mike Wicks: You don't request
Hal Casteel: then managing it smartly. I mean that's where so you know we I started my you know generative AI career at Grail by telling the lawyers what was going on and immediately they were like stop the bus which was partly I guess what I wanted them to do because I didn't want us to get into trouble with the FDA now. So when once we stopped the bus then you know then we had to you know build out that framework that I showed you uh smartly and get executive leadership on board form the committees then go through department by department to kind of do an internal survey to see what people were actually doing. Now your employment agreement may or may not have any clauses in there yet. Now employment agreements are going to be changing in the near future rapidly to kind of put these things in clear clarity for people who are on boarding. But I think any you know person that wants to be in business should really address this with safety first.
 
 

00:50:39

 
Hal Casteel: Uh these are very powerful tools. They're moving and evolving very very quickly. people like David and Obby that come out of machine learning backgrounds. You know, machine learning is a different animal than Gen AI alto together,
Mike Wicks: Good friend.
Hal Casteel: right? You know, the the risk factor for those tools is is very different. So people coming out of those backgrounds, they bring a lot of wisdom and experience because they've been using these tools oftentimes for decades. uh and then those like people like David who are earlier adopters of the genai are probably thinking about you know the risk and how to manage it. Is that true or do you think about risk David when you're working
David Q Chen: Well,
Hal Casteel: or
David Q Chen: I mean just for my own personal projects, but I mean at this point I I have a large I still have some doubts about well I guess it's a long-term memory problem context.
Mike Wicks: I Yes.
David Q Chen: It often loses context throughout as a project gets bigger and bigger and so at some point it it does you start to hallucinate in a ways that like you don't you don't actually know things or not.
 
 

00:51:51

 
David Q Chen: I think there's a lot of um yeah I think there's a lot of concern around that and also there's concern like you said around like people's misunderstanding of what AI is capable of and a misunderstanding of how it actually generates output reliability of that I think it's like uh it's very hard to communicate I think
Hal Casteel: Yeah.
Mike Smith: Actually,
David Q Chen: it's I don't know how to communicate that effectively at
Hal Casteel: Yeah. Let me in fact that's a really good segue into was something that I wanted
David Q Chen: all
Mike Smith: I think Attilla I think he raised his hand,
Hal Casteel: to
Mike Smith: so I think he's got a question.
Dunnville Grand Tour: Yes. How actually um I'm just looking I was uh very intrigued by your um framework what you draw because my um uh company where I'm working and we implementing a new system is actually um I'm facing with this very issue where people using AI within our company
Mike Wicks: Thank
Dunnville Grand Tour: and they don't know how to use it and a lot of time I'm running into reg uh
 
 

00:52:44

 
Hal Casteel: right?
Dunnville Grand Tour: regulatory issues with uh employees who's running AI and getting uncorrect information and uh here in Canada, we have the CFIA, which is the Canada Food Inspection Agency. And uh I'm getting, you know, false information to them and or companies getting targeted by the CFIA because some of the employees are using AI tools, not certified or not knowingly what they do with it. So this is when I when you walk through your uh presentation that framework will be extremely interested for me because I would love to implement something and I brought this to the attention to the executive leadership about this. I said we have to uh put some safeguards or put some stops to how we use AI at workplace. And then when David said you know the safe way and you know he triggered all the same same words but coming through
Hal Casteel: Exactly.
Dunnville Grand Tour: my my mind when I see employees just using AI and spitting out information that
Hal Casteel: Yeah.
Dunnville Grand Tour: is not factually checked or they don't prompting for correct correct information.
 
 

00:53:49

 
Hal Casteel: Correct. Yeah. So that so that's what I I mean I personally believe you know in any any size organization from startup to big uh starting with that as a you know fundamental uh and people who actually you know so if you bring that to your organization you're going to actually get more uh people are going to really appreciate you for bringing something to
Dunnville Grand Tour: Yeah.
Hal Casteel: the table right yeah what because you can jump you can jumpst start a program that's
Dunnville Grand Tour: And but you know what I can see already the issues what can raise.
Hal Casteel: what essentially what I did and and when that happens the you know
Dunnville Grand Tour: Yeah.
Hal Casteel: what the fear begins to be managed right managing fear and risk is how you know that's what management does right as we right and
Dunnville Grand Tour: Exactly.
Hal Casteel: so you know and I want to be able to share what I've created with you guys because you know I really wanted
Dunnville Grand Tour: Yeah. Great. Greatly appreciate it.
 
 

00:54:43

 
Hal Casteel: yeah we are facing right now uh the beginning of a an what I call an AI back. There's a couple things happening. I don't know if anybody there's what we call the the hype cycle, right? The hype cycle is we have a new technology and we all get very excited. Interest goes up, investment goes up and then we reach a peak of disillusionment. That's when you know in a way it started in July of this year. There was an article published uh by MIT, not a great paper from my point of view, but it it was based on real research over about a
Mike Smith: Oops.
Hal Casteel: year that found that 95% of AI projects that were internally developed or even with third parties were failing. Now, but there's two sides to 95%. One is that means 5% of uh projects were successful. Now, so what happens with the hype cycle is it goes up and you go over this peak. Then you go down into what's called the trough of disappointment.
 
 

00:55:52

 
Hal Casteel: This is where you know the the on the back side of that uh hype cycle. The f this is where we've actually learned something smart. We've learn 5% of the projects were successful. This is where typically where smart leadership and smart business starts to actually invest. invest not wildly like up into the, you know, into the into the hype, but take the lessons that have been learned from those front runners that took the arrows and then, you know, take those lessons, build your frameworks, understand what was successful, look at how what would be successful in your organization, break down what might be the hype into practical pilot projects that are managed internally.
Mike Smith: So one
Hal Casteel: Now what another thing that people found is that uh projects that were uh being managed by third
Mike Smith: second.
Hal Casteel: parties were actually uh yielding higher fruit better returns than ones that were uh being developed internally. That makes sense because you you know you've you've had you know the beginning of concentrated intelligence in the companies that are solving these problems learning from all their engagements with multiple companies beginning to you know do what they do better and recommending better uh focal points.
 
 

00:57:16

 
Hal Casteel: So there's a a business uh called work helix for example work helix u they're economists actually originally the co-founders one from the digital uh transformation lab at Stanford and then the uh MIT they teamed up and they were focusing on what is work how is work done meaning job description workflows roles permissions and actually activities and breaking that down. So if you think about it, when you look at what each of us is doing when we're working, you know, some activities we can do fully automated. I mean, that's what ERPs have. I mean, we've been in the ERP space in particular, we're not afraid of automation. In fact, automation adds a lot of value and actually save and and minimizes risk if it's implemented smartly, right? So you know we look for those opportunities where we can automate something uh to reduce
Mike Wicks: Stop
Hal Casteel: the number of uh manual work interventions and the potential risks that humans bring to the
Mike Wicks: there.
Hal Casteel: workflow you know or or managing the data like you know David and Abby do with machine learning where they can handle massive amounts of data in an automated fashion and bring a lot more value to an organization.
 
 

00:58:37

 
Hal Casteel: So it's not the automation that's the danger. It's how we're you know how we're analyzing that work and the workflows and where AI should be appropriately applied or not. So one of the my core thinking beyond the framework was then let's look at work how is work managed um work in any large uh enterprise or even I've recommended who who here has a PMP certification Mike you probably do right you
Mike Smith: No, I've looked at it a few times, but I I don't
Hal Casteel: yeah never did okay yeah but basically project management is a very
Mike Smith: have
Hal Casteel: valuable skill Right? You know, we create a project plan. You know, we figure we break it down. We, you know, we figure out in a software project, for example, we may break it into epics and features and functions and we break it down into smaller and smaller components until finally we can approach it say from a coding point of view. Now, in a business, we may also approach you know say a a change in the organization and we should also write a project plan for that.
 
 

00:59:51

 
Hal Casteel: So code of tech is kind of it's code and business process driven. It's not you know if you look at code code is actually a language. It's a domainspecific language geared towards you know particular tool sets that developers may bring to the table. But then the the language of development is also you know a software de design document a a technical design document an architectural design record and a whole lot of other components that we then end up defining as what it is that we intend to build. Now intention turns out to be one of the most important things when you're dealing with an AI system. Does everybody know what intention means?
Mike Wicks: It's not
Hal Casteel: Intent intention is not only when I define what I
Mike Wicks: true.
Hal Casteel: want, but I really have to give it that feeling of what it is I want. I have to be detailed. I have to really make sure that I, you know, we, you know, if I, if I do it with any ambiguity, ambiguity is like poison to a a generative uh large language model because in the absence of disambiguity, which means the absence being very clear about what it is that you intend, the when you have that to the specification level Before you start any kind of development, business or uh coding development,
 
 

01:01:29

 
Hal Casteel: it turns out that you should spend about 50% of your time in the specification to really make sure that your intention is thoroughly documented. When you do that, you're going to get much higher quality outcomes. David, I'm sure you're an expert at that at this point, are you? Do do you know what I'm talking about? Specification
David Q Chen: Yeah, I find it Yeah, I think I think working with uh like with with the AI agents is basically most of the work is in definition of the task itself and actually it's the meta work of organizing
Hal Casteel: Exactly.
Mike Smith: Okay.
David Q Chen: the definition of it and then it's work of organizing the evolving project and your like
Hal Casteel: Yeah.
David Q Chen: history of the context that you're going through right I find that it's uh so eventually you get you
Mike Smith: Come on.
David Q Chen: get you essentially become programming the meta structure of the of the task which is turns out well programming languages are literally that right it's a blueprint down to the very like granular definitions of what you want to do yeah it's uh it's kind of funny it's kind of funny that it turns out that everything
 
 

01:02:34

 
Hal Casteel: Yep.
David Q Chen: is approaching programming languages for general tasks
Hal Casteel: Yeah. And it's it's so we're in a really interesting time because in a sense we're we're now in the I feel like in the stage of uh you know where we're teaching the general public
Mike Wicks: General.
Hal Casteel: humans why do we have ambiguity in our language. It it turns out that ambiguity reduces conflict. Yeah. Humans are are when we disagree, we end up in a in a conflict state. So our lang our natural languages that we use to communicate are full of ambiguity.
Mike Smith: Oh,
Hal Casteel: We're not really and and the way we think turns out to be kind of geared that way. So when we start you know interacting with a system like a large language model where ambiguity is actually dangerous hallucination is a direct consequence of ambiguity as well as other things which I'll talk about in a minute but you know if you if you if you're not clear an a large language model is basically doing I mean here's a little bit of technology ex there is a basic basically a translation of our natural language into what is called a token.
 
 

01:04:00

 
Hal Casteel: That token is a mathematical representation of our language. That's what gets passed to the large language model. So when the the tokenization process happens, which is really kind of upstream before even it gets sent to the LLM, it's translating our intention or our our first prompt
Mike Smith: Thank
Hal Casteel: into what maps to a a series of tokens that
Mike Smith: you.
Hal Casteel: then get passed. If those tokens are not explicit to map our intention clearly, then what happens is we're going to get back what the LLM's statistical evaluation of what makes the next logical followup tokenization and then we get off track. Now the David was talking about context. Think of context like memory. Okay. uh you know we all go to a party and we you know we we're all going to see that party from our own unique points of view right we're in different parts of the party talking to
Mike Wicks: All right.
Hal Casteel: different people you know we all know it's you know how's birthday party that's the one thing and he's 73 years old but our collective experience and our memory of that event is going to be very very different based on our individual point of view and our experiences at that party.
 
 

01:05:27

 
Hal Casteel: Now, it turns out that when you're interacting with, we could take the very same idea for the group of us, there's what, eight of us here, and we, you know, if we were to take, you know, do a a thought experiment where we said, you know, uh, let's create a a a bicycle tour of Italy. Okay, very high level, right? Well, that's not Mike. you know Italy, you know your plan is going to be very detailed because you know where you would want to go and ours might be quite generic, right? So our output from that particular prompt would be wildly
Mike Smith: It's correct.
Hal Casteel: different.
Mike Smith: But but don't tempt me and don't tempt me to go and do that. So
Hal Casteel: No,
Dunnville Grand Tour: I think that's a good idea,
Hal Casteel: but it would be but No,
Dunnville Grand Tour: Mike. Let's team up.
Hal Casteel: but it would be fun because you know because now we're in the age where building an application is really easy. Okay, this is where you know kind of what's happened with the latest iterations of the engines and particularly
 
 

01:06:23

 
Mike Smith: Yes.
Hal Casteel: with things like kodch and others that are evolving in a parallel space like what I'm doing you know if
Mike Wicks: I'll be
Hal Casteel: you can be very specific and specify what you want we can actually build stuff autonomously without any I I have not used a software developer or directly modified any code since August 27th myself. David, you're still modifying code, right?
David Q Chen: No, not
Hal Casteel: Not really. Okay. Yeah. So, no. So,
David Q Chen: really.
Hal Casteel: now now I think what we're doing is maybe we're measuring the gap between what we our intention was and what we're getting. Right? And then we're we're iterating now to get to where we want. Right. There's that iterative process where we're we're refining.
Mike Wicks: Yes.
David Q Chen: I think so.
Mike Wicks: Thank you.
Hal Casteel: Got
David Q Chen: It's uh but there's there's also the gap of do you trust the LM to
Hal Casteel: it.
David Q Chen: do what you think it's doing right because you think that okay I I wanted to translate the code into a certain thing but I find myself over time I have to constantly try to read the code to make sure it's not
 
 

01:07:32

 
Mike Wicks: We'll
David Q Chen: it's not you know writing code that in a weird way that I know is going to lead to problem downstream and does that a lot of times right it's uh so there's a lot of um back and forth so I read a lot of code I don't write as But I I make sure like, okay, this part is modification, but I don't want to manually edit it. I want it to be
Hal Casteel: Right. Right. But you but with your background, you're able to do that.
David Q Chen: able.
Mike Wicks: take you to the
Hal Casteel: If you look at across this particular group,
Mike Wicks: house.
Hal Casteel: I would probably put you in the category of being able to do that most effectively. Right.
David Q Chen: Oh, thanks.
Hal Casteel: Right.
David Q Chen: Thanks.
Hal Casteel: But the danger is is particularly as we're dealing with agentic systems now, I can be running five parallel processes with multiple agents running in the same you know five different sessions running up to you know 20 30 agents in a single session.
 
 

01:08:35

 
Hal Casteel: The output of that many agents over five sessions ex is beyond any human ability to you know basically fly by wire. We're generating so much more product than we've ever been able to generate uh individually before and we haven't set up the right work structures yet that are able to evaluate to David's point the quality of what the output is. You know there's the qu qu qu qu qu qu qu qu qu qu qu qu qu qu qu qu qu qu qu qu qualitative quantity and quality are not you know quantity does not equal quality and there is not a direct relationship between the two. Um however in aentic systems we now have what we can call u mixture of experts. I don't know if you work with mixture of experts uh at all David in the way you're working but you know let's say we set up two different sets of mixture of experts. Now a mixture of experts might be a system architect you know responsible for you know a product manager who might be defi you know responsible for defining the product uh the system architect for defining you know kind of the overall architecture of you know how that's going to be deployed and the overall systems that that may make up and then there'll be you know an actual programmer you know an expert in the language that you're programming in And there might be a tester and there may be a QA uh
 
 

01:10:13

 
Hal Casteel: assessment. So that would be you know six or seven agents that are all more on the front end. You know once you have your specification everything you know is kind of designed and uh and then the agents work together or independently depending on how you set them up. uh and then you get the work product and then I use what's also called a mixture of experts who are judges and they may have a totally different set of skills.
Mike Smith: I think that's awesome.
Hal Casteel: I'm not employing a lot of people. I'm employing a lot of agents but I use that mixture of experts judges to come back in and do the qualitative assessment and then guide and then send the work back when it doesn't meet a certain standard. So standards are important not just the specification but but creating standards.
Mike Wicks: Back.
Hal Casteel: So that was one of the things I also started building in the code of tech was the idea that you know agents must work to certain standards be it documents coding whatever whatever standard.
 
 

01:11:20

 
Hal Casteel: So you've got the regulatory framework, you've got your your standards, you then you have your framework of the actual uh architecture of the multi-agentic system. And what I'm finding is that when you work that way, you actually get a a pretty good result uh you know that's functional, buildable, um understandable, explainable, and then the last thing that I've really focused on is what's called context. In the context from a human point of view, we have short-term memory and long-term memory. Uh, you know, our long-term memory can be triggered by things like smells, experiences. We see something or we remember something because of something. You know, there's all these ways that human memory is, you know, it's a it's an amazing tool, but we're not actually that great at remembering for very long our immediate experience. short-term memory degrades very very quickly. Uh now this is where this and this is something that agentic systems have been uh struggling with from day one uh particularly in the Gen AI world
Mike Wicks: Day one.
Hal Casteel: because the what's happening is you the LLMs are set up to be able to handle X amount of context.
 
 

01:12:44

 
Hal Casteel: Think of about it. Think of it as a bucket of tokens basically once it gets uh to a certain point and it turns out like with anthropic David I've I've actually measured it exactly now you know where it wants to autoco compact is it 76%. If you go beyond 76% of your 200 you know token 200k token for a given it will it'll autoco compat. Now what does that mean? Whatever you put into that context window, that particular framework of what you've been working with, the LLM is collapsing that uh and reducing it arbitrarily without your control unless you really get very explicit about it to what it thinks it should remember to move on into the next session. you know the next you know it refreshes itself to you know basically it compacts and then it what it remembers then gets loaded into its context and then it continues and this is to me the most interesting problem to solve and I really you know for the last year I've been working on that problem uh you know both from a scientific re you know from the computer science research perspective uh personal you know obviously working with LLMs day and night.
 
 

01:14:12

 
Hal Casteel: Um, but the this the computer science research has been really interesting. There's been it turns out that for example, if you if you repeat your prompt twice, you actually get an improved response. That doesn't make any sense to me. I can't explain it, but it turns out to be, you know, it's been, you know, researched adequately. Uh, am I doing that? uh in a sense I'm doing that by using my specification to make sure that my agents
Mike Smith: Mhm.
Hal Casteel: keep coming dial because this the coming back to the specification is what really so I make my definition of the work unit very small does that make sense you want to call it a task you want to really define it and if you define
David Q Chen: I was I was going to ask that was well I was ask that which is there because there's always that problem of you want to write down your context as soon as you can because you don't want it to compact you once it compacts you want to refresh it but the more work you do the more that concept accumulates right and then as you as I say longer and longer eventually your your token will run out much sooner so there's this problem of
 
 

01:15:29

 
Hal Casteel: Well, you don't want to keep reloading it, right? So, if you really think about it,
David Q Chen: like
Hal Casteel: you know, I if I asked you to, you know, uh, shovel a ditch from here to there, I only want you to concentrate on the next 10 feet,
Mike Smith: All
Hal Casteel: right? you know, so there's the there's the bigger picture which is the you know your your design documents that define
Mike Smith: right.
Hal Casteel: the you know the overall expectation and then broken down to the task level and then to your point about context one one thing I developed with CODATECT once I discovered that uh you know anthropic for example auto compacts its context for that particular session at 76% % I export that entire context at 75% in a file JSON NL file which is the way anthropic stores project memory. You you are you familiar with those files a little bit? Yeah. and I and I have a script that parses them into the individual messages which are contained within that context and I database it.
 
 

01:16:47

 
Hal Casteel: So I have a constantly growing database of everything that's been happening over time.
Mike Wicks: Uh let's
Hal Casteel: uh which means you the agents can actually query that when they're working on
Mike Wicks: see.
Hal Casteel: a given task and reinform themsel with the relevant te context that's pertinent to what they're working on. See that's a big that's a big evolution because if you're able to take selectively out of the massive context and compact it only to the relative context to the work unit that you're working on. Then you've got, you know, meaningful information that's being brought back in the context and that, you know, rather than overloading the context, you want the right information.
Mike Smith: Mhm.
Hal Casteel: We do the same thing as we're working with people, right?
David Q Chen: are doing it with ng are doing it with a vector database or doing it with like
Hal Casteel: What's that? Uh actually I locally I'm running a SQLite database
David Q Chen: a
Hal Casteel: locally uh that just you know it captures all the you know it's got a bunch of tables in it that captures I actually index all my agents all my components all my uh tools and all the all the messages within all the context.
 
 

01:18:01

 
David Q Chen: So it's a raw text search or versus like or tokeniz like natural language search versus like
Hal Casteel: It's a net. Well,
David Q Chen: uh I see
Hal Casteel: it's a semant it's I've got a semantic search on top of it. It's it's not a full vector uh rag.
David Q Chen: it
Hal Casteel: Now, I think there is a place for vector rag uh which we can talk about.
David Q Chen: may problems. That's what I'm saying.
Hal Casteel: What's that?
David Q Chen: It's like rag also his own
Mike Wicks: Officers don't
Hal Casteel: Yeah,
David Q Chen: problems.
Mike Wicks: stop.
Hal Casteel: it has its own problems and I actually preferred not to go the rag route uh because I wanted to make sure that my the message I I have I discovered in the anthropic uh system the way they store their context is that there are natural boundaries that are in their in the way that they're uh messaging is is captured. Uh it's not obvious when you look at the files. I mean, I had to it took me quite a while to kind of break that down.
 
 

01:18:52

 
Hal Casteel: I' I'd love to maybe talk with you more on a technical level to kind of look at how you know what my parser is doing to see if you would agree because if we can refine that parser to be very powerful and to really capture the context into you know once again the smallest possible unit that might be associated with any particular activity that's happening in the agentic system. Then we have something that humans don't have, which is real long-term memory. You know, no longer are we taking short-term memory. You know, we're actually capturing long-term
David Q Chen: Yeah,
Hal Casteel: memory.
David Q Chen: I think that's the biggest problem with uh working with this It was it was a big problem I had which was
Hal Casteel: Yeah.
David Q Chen: like as the context is longer and longer be you can just task of managing this itself is a task
Hal Casteel: Well, I created it ultimately as a program to solve it.
David Q Chen: right.
Hal Casteel: I mean that became its own aspect of codec that is solely focused on memory management as a system.
 
 

01:19:56

 
Hal Casteel: uh because I think now what I want to do because I you know once again codate techch's designed to be a multi- tenant meaning multiple you know potentially multiple businesses you know you might have cap gemini working on something and you know your own internal developers working on something right and they may even be working in the same codebase or the certainly the same project at different levels right so this ability that that opens up a whole new see we've been experiencing this individually where we're trying to manage memory. Imagine what it's like at the enterprise level, right? You know, or you know, or healthcare where we're dealing with massive data sets or in finance where we're dealing with very very
Mike Wicks: health care.
Hal Casteel: large complex data sets as well. So, I really wanted to focus on solving that problem in some, you know, very long-term strategic, manageable way because until we solve that smartly, we can't really let AI totally loose, you know, like in a full stack perspective like in
David Q Chen: There's there's a there's an idea of I mean recursively there's an idea of using AI to
 
 

01:21:01

 
Hal Casteel: an
David Q Chen: solve that problem which is you try to use a some kind of a recurrent neural network to decide which is more important. Then you you try to compress that into
Hal Casteel: Well, there's there's a new idea that just I think at MIT in December,
David Q Chen: story.
Hal Casteel: I don't know if you heard it. It's called a recursive recursive language model.
Mike Wicks: highlight.
Hal Casteel: Have you heard of that yet?
David Q Chen: I think I've heard something on
Hal Casteel: Okay. Yeah. Read that paper.
David Q Chen: that.
Hal Casteel: Uh because what I did, I read that paper. One thing I do with uh CODATECT ch, it does scientific research really well. And so I take papers, I extract, you know, you know, really create, you know, from a scientific paper understanding the elements and then I look for, you know, you know, there's kind of like theoretical and basic research and then there's applied research, right? So I'm I'm heavy into the applied research, you know, where I'm taking the scientific research and then looking for good papers like the recursive language model and then applying that into CODATECT ch, you know.
 
 

01:22:10

 
Hal Casteel: So this idea where we could now your version of CODATECT for example if you had particular ways of thinking about things you could create competing models on top of CODATECT ch that would you know work because you you develop your own scripts your own programs your own
David Q Chen: f***.
Hal Casteel: skills and agents within the framework. Do you see what I mean? that this is this this is a way you know like a battle of the bands approach which will allow us to have I mean while I've I've been the de primary developer of CODATECT ch at this point of view now that I'm letting it out into the wild what I'm looking for are for people who will kind of take it use it as a platform build on build on the platform but then find new ways and new solutions and then we collaborate
Mike Wicks: Nobody.
Hal Casteel: and we you know some of them may end up being products they may be solutions they may even end up being businesses Right.
Mike Smith: Yes.
Hal Casteel: And you know I CODATECT ch will never own what other people create, but I I would love to work with a group of people who who would like to cor collaborate because I think you know many times you know we're stronger.
 
 

01:23:14

 
Hal Casteel: You know Jared the the problems you're going to bring to light in the insurance industry. I'm part of a business accelerator at the Florida Atlantic University and one of the guys is an insurance guy, you know, and you know, you're in Canada now. Your regulations are different,
Mike Wicks: f***.
Hal Casteel: but he's facing the same kind of problems you are and I've been using Kodch to kind of help him address some of those issues in the insurance industry. And you know once we find you know you know differences in Canada for example that may be regulatory or even you know enterprise you know where your particular business has rules of its own those need to be those are the overarching rules right there's the legal framework and then there's the actual enterprise policies and procedures that are and the standard operating procedures that are equally important. This is where workflows and really that kind of work comes into play. So I mean if I look at this group see with ERP insurance and and healthc care backgrounds like this group has you know if we collaborate and figure out our own type of problems that we're trying to solve.
 
 

01:24:31

 
Hal Casteel: You know let's share the problems right code's really good with problem solving. you know AI can propose solutions right but then I you know I because I built a framework that can go beyond that do the research and and validate you know the kind of solutions that we're creating right you know I don't want you know humans are more valuable as teams than they are as
Mike Smith: has
Hal Casteel: individuals but because of AI we're actually beginning to in a
Mike Smith: changed.
Hal Casteel: sense dissolve teams in many ways because individuals can do a lot more individually, right? So,
Mike Smith: Yeah.
Hal Casteel: it's kind of an interesting time if we look at what work means in the future. you know, how do we keep humans in the loop, but also the the power of a team that can, you know, kind of, you know, if if you bring interesting uh problems to the table, if you look at this group, we're all, you know, once again, that party analogy, we're all going to have our different points of view that are actually very valuable to help solve those problems, right?
 
 

01:25:41

 
Hal Casteel: Because we all bring our different experiences and tools to bear. So, I don't want to talk too much, but I mean I and I don't want to dominate,
Mike Smith: No, no, it's all good. No,
Hal Casteel: but yeah.
Mike Smith: it's it's it's your your stuff. So, it's uh I'm not sure if you wanted to show a little bit more or like how's
Hal Casteel: Yeah. Uh yeah, I'm happy to. I mean, let me yeah,
Mike Smith: it
Hal Casteel: let me kind of talk about a little more about the memory because I think that's a particularly uh it's probably the to me that's the most impro important problem that we can solve. Uh to my everybody's struggling with this particular uh problem. Okay. Um so you know I talked a little bit about the uh you know basically these these are slash commands and there's hundreds within uh now in the anthropic world which basically helped me kind of think through why their approach was valuable. I mean basically with short commands this particular one is an anthropic command and what it does is it allows us to export a file or copy the context that's in that particular session into memory.
 
 

01:27:14

 
Hal Casteel: And what I do is you know I just save these files um I give them a
Mike Wicks: Trans.
Hal Casteel: random string just so they can be uh saved. I I save the source files of the context files because I if I ever want to rebuild my database I can. So once I've saved that this file here, let me show you what it looks like. See this is this is what anthropic the session that I was working in. You can see it's pretty big, right? There's a lot of data in there and
Mike Wicks: Hold
Hal Casteel: so some of it looks I mean some of it looks interesting some of it's you know understandable some of some of it's not here I was
Mike Wicks: on.
Hal Casteel: updating those uh that risk management framework one thing I do with all my documents I I create a structured
Mike Wicks: Perhaps
Hal Casteel: header that can be data database. So every document has this same structure that allows me to be able to uh you know basically save the document as a databaseable uh searchable uh element and so I I do that with all code and all components that are built into the platform.
 
 

01:28:45

 
Hal Casteel: So that you know that was the last thing I was working on before I did the ex the export. So once I do that um let me close
Mike Smith: So once I close
Hal Casteel: this.
Mike Smith: this, does anybody else have any questions?
Hal Casteel: So then I have a
Mike Smith: Sorry, I'm just going to throw it to the group. Does anybody else have any questions before we kind of move on? I know we're we're at about 8:30 Eastern right now. So, sure if uh and is all this making sense?
Hal Casteel: Yeah.
Mike Smith: Um I just want to kind of kind of take a pick the temperature uh here.
Mike Wicks: Yeah, it's it's making sense. What one thing I was going to ask you how like maybe I misunderstood, but you you would stop at 75% and then
Hal Casteel: Well, I did this one manually just to demonstrate it.
Mike Wicks: export
Hal Casteel: Uh, yeah, let me I'll I'll speak to that in just a minute. Yeah, so I auto I auto export uh at 75%.
 
 

01:29:54

 
Hal Casteel: I just did a manual export. it wasn't at it wasn't approaching that number yet. Um but this CX command is basically the context extraction process where it takes those uh session files and then it it it breaks it it's basically a parser and a datab and then it loads it into the database and then so once it it does its thing uh you know it's
Mike Wicks: Yeah.
Hal Casteel: it's a Python script that you know breaks that those files down into uh databasable units of activity basically and it'll you know it right now I have a you know there's a couple hundred thousand messages since I think it's you know roughly December 5th or something. So, you know, and I I'm I'm generating a lot of data. So, that, you know, the average user is probably not going to generate anywhere near that much.
Mike Wicks: that budget.
Hal Casteel: But do you see this 50%. So,
Mike Wicks: Yeah.
Hal Casteel: each session has its own UYU ID, which is a unique identifier that is automatically assigned to every session.
 
 

01:31:20

 
Hal Casteel: uh and then I I'm able to see visually what the status is of how many tokens I have remaining in that context. Um, but it it'll auto I have another uh it's a Rust binary that runs locally that you know basically watches the the it counts the tokens and does the math and then it auto exports and that
Mike Wicks: Works.
Hal Casteel: and then it and it also auto indexes everything.
Mike Wicks: Works.
Hal Casteel: So you know once once that's done then you're able to you know have I have other commands like this which are you know I can I can search you know that or the agents themselves can run the you know you know it's basically whatever they're working on they could query the database directly independently of me uh directly to pull their own context into what they're working And that's kind of how the system is working right now.
Mike Smith: So I have a question. So what is the input? So again if we had that idea How do we get that,
Hal Casteel: Yeah.
Mike Smith: you know, how do we take that idea, you know, and in conjunction, you know, with, you know, conversations with you, how do we get that into into here so things can actually start
 
 

01:32:48

 
Hal Casteel: Yeah, that's a good that's a good question.
Mike Smith: running?
Hal Casteel: So um so I have a couple you can basically copy I I I keep one folder right now that is basically remember everything is broken into subm modules and the CODATECT
Mike Smith: Yeah.
Hal Casteel: core this here I have this one called analyze new artifacts and this is kind of you know I'll I'll open that in the finder just to show Uh so I have a lot of you know when I think of something I create a
Mike Wicks: You know, when I
Hal Casteel: folder in this uh and you know for example YouTube is
Mike Smith: Mhm.
Hal Casteel: a gold mine right? Uh so I I will convert YouTube uh videos into audio I mean not into audio clips but basically into text files that allow me to uh you know I can take a you know this is a a YouTube transcript from a guy that is very smart on AI.
Mike Smith: Mhm.
Hal Casteel: I don't read these things and I actually rarely watch the videos uh because the way I handle that is I automate the export of the YouTube uh
 
 

01:34:09

 
Mike Wicks: part of the YouTube.
Hal Casteel: into these text files and then I run codec has a docu you know basically a YouTube transcript and analyzer set of agents and skills and then that may create something that's interesting and if it looks like an interesting idea that could applied. Here's another one that is uh has anybody
Mike Smith: Okay.
Hal Casteel: heard of modal? Modal is a a a containerization. You want agents, they could run wild on your operating system, but that's pretty risky. So,
Mike Smith: Mhm.
Hal Casteel: good idea is to run them within a container. And Modal is a company that does it very well, but it's very expensive.
Mike Wicks: company.
Hal Casteel: So I took everything I learned about modal and I did a bunch of research of my own and then I created a modal like uh containerization platform of my own that I was able to deploy onto uh Google cloud workstations where I can have multiple containers within a single workstation for multi-tenant you you know, multi-user, uh, much cheaper than I would ever be paying uh, modal, for example.
 
 

01:35:34

 
Hal Casteel: So, you know, I'll use things like this.
Mike Wicks: Stay
Hal Casteel: So, you know, lots of different I'm interested in lots of different things. So, you you want to take, you know, you would run code tech locally and then have,
Mike Wicks: here.
Hal Casteel: you know,
Mike Smith: Almost.
Hal Casteel: you drop things. They could be ideas, they could be images. Uh, you can prompt also directly. Uh here's another way you can interact with the uh system. Here there's what I call a CR command. And this what this does is it basically allows you to you know prompt directly to the and then you you know you can you can have things like a new uh uh let's create a new project plan for example, right?
Mike Wicks: I I just have a question for you're you're familiar with uh Microsoft
Hal Casteel: Yeah.
Mike Wicks: Fabric.
Hal Casteel: Uh, not no
Mike Wicks: Okay. Uh best way to put it, it's just one one place where they can pull data from different cloud sources and it's all in one spot.
 
 

01:36:48

 
Mike Wicks: So all the developers have access to it. Uh everything's in one place. So um what's happening now is a lot of people are moving from like PowerBI premium to fabric but with all this you know hype
Hal Casteel: Yep.
Mike Wicks: around building agents and and you know co-pilot solutions uh a lot of these advanced analytics companies are just basically building one agent that sits on top of all the fabric data.
Hal Casteel: Right.
Mike Wicks: So it's pulling all the
Hal Casteel: I'm doing something yeah I'm doing something similar in that you know like all these
Mike Wicks: knowledge
Hal Casteel: session files and all the you know basically you know multiple users working within their own projects depending on the uh you know the enterprise structure right you
Mike Wicks: Good.
Hal Casteel: know are you part of a team you know I have a whole user management arbback uh type of system built on top and then that all of the local stuff gets pushed to the cloud into a Postgress uh database that you know basically is
Mike Wicks: Could
Hal Casteel: kind of it indexes all the different uh data basically it's it's not a data lake in the sense
 
 

01:38:04

 
Mike Wicks: some of they could some of them like plug in your solutions to their let's say fabric data and then work from from your
Hal Casteel: yes absolutely yeah it's it can write you know it yes it
Mike Wicks: solution.
Hal Casteel: would because I'm sure fabric has a an API or MCP server,
Mike Wicks: Yeah.
Hal Casteel: right? Yeah.
Mike Wicks: Yeah. I I I'm just thinking of like how I would sell this and
Hal Casteel: Yeah.
Mike Wicks: uh it it would probably start with that, you know, maybe having like a uh giving you a certain amount of access to their fabric data, right?
Hal Casteel: Yeah.
Mike Wicks: And you know,
Hal Casteel: And I don't even need the access because, you know,
Mike Wicks: as
Hal Casteel: it's not me. It would be you know they'd be running kod you know CODATECT ch on their own in their
Mike Wicks: Right.
Hal Casteel: own and uh so what yeah the abstraction of what the tool itself into what people have done do with it you know I'm I'm trying to I'm not I wanted to
 
 

01:39:03

 
Mike Wicks: Yeah.
Hal Casteel: build an extensible platform right because if we can build a community of users right you know just the the power of ex you know what people learn on the platform would be very valuable to improving the platform right or extending the platform for example I have a a guy that I'm working with he's a former army uh documentation specialist in the military they have very strict standards around their their documentation and when he got out of the military
Mike Wicks: It's
Hal Casteel: he started developing curriculums for a combination of uh charter schools, private schools and public schools. And each one of those has slightly different methodologies and requirements in terms of the uh and soc uh designed around you know basically education and curriculum development. So when I saw his use case, um what I did is I gave him kind of a customized version and then showed him how to extend the
Mike Wicks: That's Charlie.
Hal Casteel: u curriculum. I'm I can't remember the name of it. the but basically you know like one of you were talking about content marketing patterns right you're going to have your own patterns but what I've created is a framework where these patterns can then be extended does that make sense so you know my think of it it like a kit of
 
 

01:40:51

 
Mike Wicks: Oh.
Hal Casteel: parts I wanted it like Lego bricks but what you build with it and if you want to build your own Lego bricks go for for it, right?
Mike Smith: One
Hal Casteel: You know, so that that ability for people I don't I I don't know about you guys,
Mike Smith: more.
Hal Casteel: but I don't like platforms that tell me how I have to work, right? You know, that becomes irritating, right?
Mike Smith: Sure.
David Q Chen: Do you do P open source it?
Hal Casteel: I I don't I I I want to keep it closed source until we build up a sufficiently a community. Uh I think it's two there's two things. The reason I don't one is this this if it's let loose in the wild could fuel massive unemployment all by itself. Uh because it will it does replace humans very quickly. Um so what I wanted to do initially was work with businesses who really understand what that potential means as we kind of reinvent what work does look like in the future.
Mike Wicks: That's like that.
 
 

01:42:02

 
David Q Chen: Is
Hal Casteel: Uh and then once society kind of adapts to that,
David Q Chen: it
Hal Casteel: yeah, open sourcing it or parts of it would make sense. Uh but I don't think society's ready for massive unemployment
David Q Chen: is so is it capable of deciding which scale or which sub
Hal Casteel: yet.
David Q Chen: agent would fit a particular task?
Hal Casteel: Yeah.
David Q Chen: So like are you able to say
Hal Casteel: Yeah, it is. Here, I'll show you how that works. I have a command called witch and you know so if I go like mixture of expert agents for example right and then I give it a task um you know let's say u let's just say assess the the quality since I was working on the uh of the riskmanagement framework, right? I I don't even need to give it this much information. I'm just kind of using it as an example. Um, what it'll do now is it the witch command basically searches the database for all of the agents, skills, and components.
 
 

01:43:23

 
Hal Casteel: See, it's pulling the component indexer and it it's figured out that, you know, it's like a quality review and audit for compliance documentation. So, it's figured out the use case. Now it's going through all the uh agents and it it's figuring out which which agents. So it looks like a docu documentation quality agent. Uh we'll see what it's it's going to come up with a you know basic it's it's planning and going to recommend a set of agents basically. You see how that works? So you have your because
David Q Chen: So, so right now if you give it,
Hal Casteel: it's
David Q Chen: so if you give it a document to say here's a task going accomplished,
Mike Wicks: Please
David Q Chen: is it capable of saying like I'll plan out all these tasks.
Mike Wicks: go.
David Q Chen: I'll decide these agent will do these as a sequence and then perform it completely
Hal Casteel: Yeah,
David Q Chen: automated.
Hal Casteel: completely automated. Yeah. Um,
David Q Chen: It's
Hal Casteel: so now it's giving me a couple options, right?
 
 

01:44:29

 
David Q Chen: cool.
Hal Casteel: Uh, you know, I told you about the mixture of expert judge panel. That would be the best because this is a compliance related. Uh, not only do I want the work done, I want it measured and judged. So, you know, if I wanted kind of a faster or dirty, you know, use a single judge, but if I were really doing this for real, because this is compliance related work, I would I would pull in the judge panel.
Mike Wicks: That's a good girl.
David Q Chen: Essentially,
Hal Casteel: Cool. Don't you think? I mean, it's interesting.
David Q Chen: it's your own workforce. You are able to
Hal Casteel: Yeah, it does. and and it's smart about what which which uh you know which
David Q Chen: basic
Hal Casteel: uh so and here was three a third option which
Mike Wicks: Okay.
Hal Casteel: was you know a documentation quality agent which validates the documents themselves whether they meet the code of tech document standards and then here was a compliance checker agent that is supposed to check it against the NT EU and ISO standards that are embedded in the platform.
 
 

01:45:49

 
Hal Casteel: So that third option actually would be good. I would pro sometimes I will go you know so I might say option
Mike Wicks: It would be Christmas.
Hal Casteel: three and then option two because you know option three would do the assessment of you know and give the feedback and then I can pass it to the judges that they they'll that'll then validate whether option the option three output was was appropriate. it. So you're not locked in. You can make, you know, you can you can basically string these things together like programs. So not only do you have multiple agents, but you actually the workflow that you're creating that it's proposing, you can string together in multiple ways. To me, that's one of the real interesting things that that I want to work on going forward is kind of, you know, how, you know, we're so used to thinking of enterprise workflows a certain way. What would business look workflows look like in the future with agentic systems because they don't they're not
Mike Smith: Mhm.
Hal Casteel: so they don't need to follow necessarily fully the human workflow, right?
 
 

01:47:01

 
Hal Casteel: They can create new works that are more appropriate maybe because to hire you know imagine you had to do this work right you know look at these are highriced people right you know you know I could I could have five or six agents working on this if you look at the value replacement value of the agents themselves with human workers you know you're one one developer running on CODATECT ch, you know, basically you're getting about a $6 million a year return on the value of that. You know, like a somebody like David who's really, you know, not just a a casual developer, but a real smart developer, you know, running with a platform like this, you know, his his value to the organization's actually multiplied. You know, I mean, let's say he's typically making a a half a million dollars. He's making 10 times that much more value for the corporation working with a tool like
Mike Smith: So that that is an interesting thing because you know one of the thoughts I was having
Hal Casteel: this.
Mike Smith: was in and about the kind of service side of because that's also supposed to
 
 

01:48:17

 
Hal Casteel: Well,
Mike Smith: be
Hal Casteel: it's creating a huge opportunity from my point of view. Uh yeah. So what what happened to me this fall? I was kind of I wanted to launch this as a I mean I didn't raise any money.
Mike Smith: correct.
Hal Casteel: I've done this all on my own nickel. um you know and honestly not that much money invested because even though I'm I'm because I'm working agentically you know so relatively speaking my productivity is very high you know I have I don't know how many millions of lines of code in in invested in this platform you know I have 84 I'm only showing you one the the I only showed you code core I didn't show you the other 84 things it's built Do you see what I mean? See, CODATECT ch core is the brain of the operations.
Mike Wicks: No problem.
Mike Smith: Yes.
Hal Casteel: But now, you know, like I built a workflow analyzer for enterprise, you know, that's a whole, you know, I built, you know,
Mike Smith: Yeah.
Hal Casteel: so I built all these other components because I think, you know, there's what, you know, the verticals within different industries, right?
 
 

01:49:24

 
Hal Casteel: And then if you think about like the horizontal workflow analyzer, what should be identified and what should not and what should be hybrid. So I want to data mine enterprise workflows and look for those horizontal opportunities and build things, right? Horizontal solutions that can work across multiple verticals, right? So there there's a that's a gold mine, right? Picks and shovels. We're so now we have a platform that can basically find gold and then build it, you know, because the cost of building it is relatively low.
Mike Wicks: Clear.
Hal Casteel: Finding the opportunities, it should be looking it it should be a prospector as well as a minor, right? You know, really looking at those to your point a service business.
Mike Smith: Well, I'm I'm thinking because you know because now you can use things like notebook LM you can it can do go
Hal Casteel: So
Mike Smith: through do deep research and they're saying like things like you know companies like McKenzie and stuff like that you know the big uh the big uh consulting companies they're they're struggling now because you know because you can just do this and get that level of stuff pump it out pump out you know you know you know competitive analysis you know across you know different platforms for different companies and spit it out for them all in a nice tidy little
 
 

01:50:51

 
Hal Casteel: Yeah. Well, see Work Helix, I mean, uh,
Mike Smith: thing.
Hal Casteel: James Milan, who's the CEO of Work Helix, you know, with these, you know, basically there, you know, they they went to they are the they are the AI assessment engine for McKenzie. you know,
Mike Smith: Yeah.
Hal Casteel: they so they they they McKenzie pays them $600,000 to run a scan on a Fortune50, right? And then then McKenzie works with Price Waterhouse Cooper to take the output of what work helix outputs to go in with developers and
Mike Wicks: Hey,
Hal Casteel: consultants and charge another millions tens of millions probably to come into enterprise and solve those problems. What I wanted to do was take CODATECT ch almost like a work, you know,
Mike Wicks: Let's
Hal Casteel: a work helix replacement for SMB market and let people run them locally inside the business and use their own resources or, you know, work with local, you know, talent or their own team.
Mike Smith: No, no, time out.
Hal Casteel: No,
Mike Smith: Time out. No, no,
Hal Casteel: no model.
 
 

01:52:03

 
Mike Smith: that's a that's a total business consulting consulting thing that we can
Hal Casteel: Yeah. Yeah.
Mike Smith: actually start to mind. And again, if again, if you kind of think about it and think about, you know, the different people that we got on this call,
Mike Wicks: take
Mike Smith: these are just again as a as a consulting company, this that could be the start, then you make your millions, then you give then you give it away. You don't want to give it away.
Hal Casteel: Well, yeah. That's why I didn't plan on giving it away because what happened to me was when I was in the business accelerator,
Mike Smith: Now,
Hal Casteel: I was in two business accelerators at the same time this fall at Florida Atlantic University. I'm still in it. I mean, it goes on for and I'm getting into some more. I was interested in getting into those accelerators because of the number of businesses it was bringing me in touch with. uh in those in the in the 50 businesses that I was interacting with 15 of them approached me about building their their products basically and and you know offering me equity in their company and and pretty goodized checks and uh we're taking on some of those projects.
 
 

01:53:08

 
Hal Casteel: So what I did because I knew I was going to be overwhelmed very quickly with the amount of work that people were going to be coming at us with, I started working with dev shops in the Orlando area, seven different dev shops to bring them up to speed on CODATECT ch because, you know, we're going to cherrypick opportunities and then hand off those that we don't want and for partnerships so that we have basically can build parallel service arms, you know, as the opportunity grows because you know I don't want to get bogged down runninginess. I don't want to become a cap gemini. I'd rather you know but I want to be able to make a lot of money basically. So we should talk about that Mike how how we should be thinking about this out of you know the different ways of modifying you
Mike Smith: Correct. Because people don't have a McKenzie level of stuff.
Hal Casteel: know.
Mike Smith: So you think of all I mean you think of going into like an industrial area in any city.
 
 

01:54:05

 
Hal Casteel: Yeah,
Mike Smith: They don't have they don't have this. They would probably love it. they would probably love to pay for it if you if you pump the
Hal Casteel: right.
Mike Smith: idea into it into code attacks and then basically get you know get the output out because all the some of the AI tools now can actually just do some just you know just kickass stuff.
Hal Casteel: Well,
Mike Smith: You can just get it
Hal Casteel: and I think also, you know, going back to that 95%,
Mike Smith: out.
Hal Casteel: you know, the mck the uh MIT paper that was in July, the idea that third party providers actually provide more value than the the
Mike Smith: Correct.
Hal Casteel: internal resources do internally.
Mike Smith: Yeah.
Hal Casteel: Um
Mike Smith: And think and think about it in practical terms like Jared uh workflows for in the um
Mike Wicks: Yeah.
Mike Smith: industry uh in insurance industry um you know some of those you know repetitive tasks
Hal Casteel: yeah.
Mike Smith: that you know hey you do it you do it you do it.
 
 

01:54:55

 
Hal Casteel: Yeah.
Mike Smith: Hey, you know and and also not only that but also you know using it as you know lead generation um customer management.
Hal Casteel: Oh yeah. Yeah. I I'm building a CR Yeah.
Mike Smith: So
Jhared Smith: Well,
Hal Casteel: I built a CRM into the platform for lead,
Jhared Smith: I
Mike Smith: again,
Jhared Smith: think
Hal Casteel: you know, lead generation and prospect management. Um I actually I mean I'll show you here real quick. Uh, one other
Mike Smith: yeah, because I was thinking like even like you know itilla like you know taking it and you know starting to you know
Hal Casteel: thing.
Mike Wicks: Oh,
Mike Smith: doc you know get that level of you know documents and and again becoming becoming the uh you know the go-to guy you know in in your particular uh the company that you work for and kind of get that get that stuff
Dunnville Grand Tour: Yeah, definitely. No, that I think that's a goal of mine.
Mike Smith: started and but but again taking
Dunnville Grand Tour: That's a goal of mine for sure.
 
 

01:55:48

 
Hal Casteel: Yeah.
Dunnville Grand Tour: Yeah.
Mike Smith: taking you know I I think in in my mind's eye taking the lessons that we could learn through Ailla to provide that level of documentation for different um you know different companies that's that's a that's
Hal Casteel: It's a huge opportunity.
Mike Smith: printing money service you're just printing
Hal Casteel: A huge opportunity. I mean, if we That's where I wanted to. Yeah.
Mike Smith: money and these are things that and I think these are the things that we
Hal Casteel: Yeah. I'm I'm totally with
Mike Smith: just need to be thinking and again that's why we kind of brought the group together you know as a as a collective and now we've got some other you know some expertise in all these different areas like where where can we just
Mike Wicks: All
Mike Smith: take this and apply it you know and you know hey spin up spin up a consulting
Hal Casteel: Right.
Dunnville Grand Tour: Yep.
Mike Smith: company get get things moving and and you know
Hal Casteel: Right.
Dunnville Grand Tour: 100%.
Mike Smith: um because I think I think service service you know is I mean that that's easy that's like consulting you know hey I'm just going to tell you how to do it you know and and maybe show you how to
 
 

01:56:48

 
Hal Casteel: Yeah.
Mike Smith: do it but you know it's going to be up to you guys to to go and go and do it you know and again I think we've
Mike Wicks: right.
Mike Smith: got enough you know we've got ERP experience um here we've got like I said insurance exper experience we've got you med device experience. How does this apply to, you know, not necessarily the med devices themselves, but but companies that you're dealing with in and around the med devices that need this level of
Jhared Smith: So, H how if you don't mind me bouncing something off you.
Mike Smith: stuff.
Jhared Smith: Um the the I'm interested your take on the it seems to be the arms race in our
Hal Casteel: Yeah.
Jhared Smith: industry right now is who can make a risk analysis and actuarial AI that can do the actuarial work for the insurance companies so they do not have to pay the actuaries.
Hal Casteel: Do you understand the problem?
Jhared Smith: Do you think I do? Yeah.
Hal Casteel: Okay. Well, yeah. Work with me for,
Jhared Smith: I'm curious of your opinion on it.
 
 

01:57:48

 
Hal Casteel: you know, let's create a you know, we'll we'll create a requirement, work together under fully fully shape that out. Uh, CODATECT will build one. I mean, it things go pretty fast. That's one thing I found. Uh if we understand the problem well and we we specify it clearly um you know I mean basically I'm running on top of $200 a month claude max accounts uh and you know it it's buildable. I mean if you so yeah I would love to go to market with something like that because that's that's a high value uh industry solution, right?
Mike Wicks: Right. Yeah. for for myself like I I'm constantly uh there's different case studies that come up where you know clients various industries right with with um with data analytics it's you know so widespread right uh but when I look at for example
Hal Casteel: Yeah.
Mike Wicks: discreet I I always think that you know they're they're leading kind of outpacing batch and and process manufacturing in terms of IoT uh spend Um, so I I feel like for them, you know, I guess Mike, like you you noticed, but like planning, scheduling is kind of the main area of like, you know, high earners, right?
 
 

01:59:13

 
Mike Wicks: Like those guys would definitely need some some assistance with with an AI tool just to predict, you know, planning scheduling, especially with like you even with batch processing, new new ingredients coming in all the time, right? different changes in the market in terms of what people want.
Mike Smith: And then and then I imagine from from that also starting to to you know look at
Mike Wicks: So
Mike Smith: historical stuff and also kind of get forecasts out there so you can actually start you
Hal Casteel: Yeah.
Mike Wicks: yeah.
Hal Casteel: So there's kind of the planning, budgeting,
Mike Smith: know
Mike Wicks: Yeah.
Hal Casteel: forecasting piece and then I think there's a whole supply chain management piece which is you
Mike Smith: correct
Hal Casteel: know what happen you know how do you how do you adapt to changing
Mike Smith: well how how do you feel the gaps in in kind of the supply chain management now and all the different and all the different touch points Because you know especially now
Hal Casteel: Right.
Mike Wicks: Yeah.
Mike Smith: again with tariffs and you know just you know just you know the upheaval of the supply chain management you know can we can you almost put things into codex and then spit out just a you know hey this is this is SEM optimizer optimize
 
 

02:00:26

 
Hal Casteel: Right. Well, think think of think of code techch in two ways.
Mike Smith: your
Hal Casteel: One is as the platform that can create things and model and and measure and do all that. But then, you know, but then, you know, I actually believe there's still a huge value in the, you know, kind of design, you know, use kod to like the architect, right?
Mike Smith: Yeah.
Hal Casteel: And then and you but it doesn't have to all you know I only use LLMs as
Mike Smith: Yes.
Hal Casteel: appropriate. See I think people are overusing LLMs in a way. You know they're trying to use it for everything. See, what I've been I've been doing is I'm starting to pull back and go more into a a neuros symbolic programmic way, which is, you know, basically to break a problem down into what can be managed programmatically, which is actually far cheaper because once we write that program once, we don't, you know, if that program is a program of programs, you know, then you're not using so many tokens and we don't need to pay for tokens.
 
 

02:01:26

 
Hal Casteel: We only use the the LLM for you know the the parts where AI is appropriate. We may want to use machine learning tools in other places, right? We may, you know, it it's like build it up appropriately the way we should be building software in traditionally, right? LLM is just another magic sauce. It is a very powerful one, but it shouldn't be the only solution. That's my point is you know I wanted to build a platform that can think intelligently around those type of problems and then design and build them but at the same time save energy save tokens save you know not hallucinate really get you know this idea of context sometimes you know if we if we try to keep everything in
Mike Smith: Right.
Hal Casteel: memory and then use the LLM to solve that problem well there's there may be 10 different ways to solve that problem programmatically if we break the problem down into small enough work units, right David? I mean, wouldn't you want to solve problems that way if you could, right?
 
 

02:02:29

 
Hal Casteel: Because you run your program, you know what's in your program once you get your program thoroughly debugged. You don't know the second time you run through something through an LLM exactly what you're going to get.
Mike Smith: Yeah,
Hal Casteel: It's like an approximation of the same solution,
Mike Smith: the
Hal Casteel: but it may not, you know,
Mike Smith: same.
Hal Casteel: I they're going to get better, but are they going to be repeatable, reproducible, explainable, auditable? Maybe, maybe not, right? I think we need to build systems that are particularly if we're dealing with business problems like what you guys are talking about,
Mike Wicks: Yeah,
Hal Casteel: you know,
Mike Wicks: I mean aside from ERP,
Hal Casteel: we want to make sure that it's
Mike Wicks: like a lot of the AI projects I've worked on are like for whatever reason they start with HR, right? You know, some simple co-pilot solutions there and building, you know, multi- multi-
Mike Smith: Mhm.
Mike Wicks: aents for for different HR, you know, departments. But it's it's always very basic. And for me, where I've noticed, you know, like Mike said it, but uh reporting, forecasting is a big big thing for most companies, but I've just never seen it in planning and scheduling um where someone's, you know, working side by side with an AI tool.
 
 

02:03:41

 
Mike Wicks: I haven't seen that personally
Mike Smith: Yeah,
Mike Wicks: yet.
Hal Casteel: No,
Mike Smith: because I've been actually I've been actually using it to when I when I do my
Hal Casteel: but I think
Mike Smith: demos and again this is again everything needs to be kept but you know you know I I tap into my uh you know into things like my you know using claw I tap into my demo account and I'm running, you know, project analysis, you know, tell me where I'm going to hit critical mass, you know, in my project where I need to I'm going to run out of I run out of time and I need to look at, you know, acquiring more headcount to to account for that and and it can go through
Mike Wicks: Yeah.
Mike Smith: and it and it pull it out and basically tell you at this point in time, at this date, this is where you're going to hit that critical And I think I think a lot of
Hal Casteel: Yeah.
Mike Smith: companies you know especially if they're kind of project based companies I think you know or um you know you
 
 

02:04:37

 
Hal Casteel: Service businesses. Yeah.
Mike Smith: know businesses engineer you know engineer which again which is more
Hal Casteel: Yeah.
Mike Smith: project based um those are companies that are just they're just they're just eating this
Mike Wicks: Yeah.
Hal Casteel: Yeah. Yeah.
Mike Wicks: Well, a close friend of mine who uh they build transformers for Mad Device Aerospace, right? Um and uh one thing I I've noticed with him is he has historical data in his ERP um of quotes and you know estimates he's he's put together. But with that business, it's always uh you know, they they give you a a bill of material and you have to put the the estimate together and and it takes his team quite a lot of time to get back to the client. Sometimes it takes, you know, an hour or two just to, you know, push out an estimate. Um and uh whereas he could use that historical data, previous you know um quotes they've built out and you know use that towards some a new customer that might you know request something for
 
 

02:05:41

 
Mike Smith: Yeah,
Mike Wicks: them.
Hal Casteel: Yeah.
Mike Smith: something like items or again that kind of gets into kind of configure price quote CPQ type type functionality where you know you know you've got en engineers working in solid works designing designing designing you know,
Hal Casteel: Well,
Mike Smith: but
Hal Casteel: actually, yeah, that brings up an interesting one. Um, Will McKinley, who's one of my co-founders,
Mike Smith: then
Hal Casteel: uh, he he had me integrate into a, uh, it's it's one of the open-source 3D modeling tools. And you know we were able to you know generate on demand virtually any shape uh you know very very fast. I mean it was it was pretty interesting.
Mike Smith: So think of
Hal Casteel: So when we have APIs like that,
Mike Smith: that.
Hal Casteel: those it's really fun actually because it's you know that kind of stuff we haven't been able to you know usually it's you know a a designer having to work through those things but because you can iterate so many times and so rapidly with AI you know you're able to get what you want.
 
 

02:06:49

 
Hal Casteel: My wife's an architect and you know we we were designing a new house recently and she gave me you know basically a napkin. I took a picture of the napkin and fed it fed it into code attack and then used the 3D modeler you know and then you know and then you throw it to nano banana on Gemini and she goes yeah that's it was like it was pretty
Mike Smith: Yeah.
Hal Casteel: cool you
Mike Smith: And that's and that's I
Mike Wicks: Well, that that's that's the thing with with that case is like,
Hal Casteel: know
Mike Smith: think
Mike Wicks: you know, he gets a custom request for a custom transformer, right?
Hal Casteel: Yeah.
Mike Wicks: And they might have built something very similar a year ago,
Hal Casteel: Yeah.
Mike Wicks: but you know,
Mike Smith: correct.
Mike Wicks: his engineers.
Hal Casteel: You've got that historical data and you can find it
Mike Wicks: Yeah. And you know with with time you know of estance you know some of these companies especially ones bigger than
Mike Smith: Yeah.
Hal Casteel: actually.
Mike Wicks: him they'll send out an estimate within an hour right and uh so it's competitive in that sense.
 
 

02:07:44

 
Hal Casteel: Yeah. But I I built
Mike Smith: I think I think speed now in those particular industries is is also now going to be key. I mean if you can if you can deliver you know uh you know a 3D diagram in
Hal Casteel: Yeah.
Mike Smith: in minutes compared to you know two two or three weeks you're you're you're in ahead you're at
Hal Casteel: Yeah.
Mike Smith: the head of game just pure and simple.
Hal Casteel: So what one thing I did built Yeah. I built a you know like these document standards into the platform you know so you can create templates for any type of document you want like you know it could be a contract it could be a invoice it could be a quote could be whatever you want and then you know once you have that as a template then you you know the you design a program that puts all the pieces together and then you know right now I'm working on uh email automation where you know the last mile is to send it right. I mean, unfortunately, I I'm a goo Google AI partner, but they don't have their own mail service.
 
 

02:08:42

 
Mike Wicks: Oh my god.
Hal Casteel: I have to use a third party uh mail service to to send mails. But but it's cheap. I mean, you know, mail services. So I'm going to have in a mail out you know the ability not just to create all that but to automate even this you know review approve and then send should all be
Mike Wicks: H I I I wanted to invite Pete uh
Hal Casteel: but yeah let's think about
Mike Wicks: the uh the owner of Elletron to uh to this call that he was busy in uh driving home from Kitchener uh back to uh to Oakville. But I think you know if if he was a let's say pilot run for that scenario and then kind of use that as uh you know to to branch out to bigger companies than him. Um,
Hal Casteel: Yeah.
Mike Wicks: is that something
Hal Casteel: No, I'm I'm super interested.
Mike Wicks: that
Hal Casteel: I mean,
Mike Smith: Yep.
Hal Casteel: Mike and I have been wanting to work I mean, I loved working with Mike when I was at Oracle, and you we've,
 
 

02:09:35

 
Mike Smith: That's
Hal Casteel: you know, long run, we've been talking forever. And, you know, if this ends up being multiple people doing different things, that's cool, too. You know, what I'd like to do is kind of have a group that, you know, comes together periodically, even if we work separately. But, you know, I'd love to, if you've got ideas,
Mike Wicks: Yeah.
Hal Casteel: I'd love to help you sort through your ideas and then, you know, get some
Mike Wicks: Well, for for me, just being in in sales, like I'm I'm constantly approached with,
Hal Casteel: proto.
Mike Wicks: you know, hey, uh, we want to build out this this co-pilot solution, and it turns out to just be, you know, basic like Power Automate thing, right? Where they don't even need an agent.
Hal Casteel: Yeah.
Mike Wicks: And I'd say like 90% of the projects, you know, tend to be that or, you know, some basic semantic models built on top of what they have. But in general,
Hal Casteel: Right.
Mike Wicks: I haven't seen anything too too technical aside from, you know, discrete manufacturing, chemical manufacturing, pharmaceutical.
 
 

02:10:36

 
Mike Wicks: Um, a client I worked with,
Mike Smith: Think of it.
Mike Wicks: uh, Pharma, like David, from what I remember working with them, is anytime they're they're compounding and, you know, moving from one cost center to the next, obviously there's like a an e E signature that needed to be take place before it moves to the next cost center. But that's that's one thing I I could remember.
Hal Casteel: Well, yeah, that's what that's one of the things I did at Grail was we were, you know, I had to do uh uh ISO 13485 uh certification, which is digital signature on procurement. And uh you know we we had the option of going like with Docusine had a life sciences solution and I went you know they wanted to charge us every time we had a signature and I went no we built our we we built our own internally with AI
Mike Wicks: Yeah.
Hal Casteel: and it worked great. It's been Grail's been using it for three years now. So point solutions like that, little things are great, right? Because anything that saves the business money long long term is I think an appropriate application quite
 
 

02:11:38

 
Mike Smith: Well,
Hal Casteel: honestly,
Mike Smith: yeah.
Hal Casteel: you know.
Mike Smith: And think of think of some of the bigger companies who might already have like a data warehouse where they're storing stuff on a you know SQL server or they're storing stuff on massive
Hal Casteel: Yeah.
Mike Smith: spreadsheets you know combining all of that
Hal Casteel: Oh, this thing is this thing's amazing at you know data mining spreadsheets because you know because then you know you can
Mike Smith: stuff.
Hal Casteel: take all that data and then database it properly and then make you know and then build proper UIs
Mike Smith: Correct.
Hal Casteel: on top of it and you know make you know the other thing that's beginning and I think this is a really interesting space is the idea of dynamic user interfaces where people interact with something and get things the data the way they want it and that's you know the the idea that you don't have to hardcode these rigid UIs necessarily. Have you played with that at all, David, at all? You know, where
David Q Chen: I I actually did I I had had this idea half a year ago and I I actually tried to implement
 
 

02:12:34

 
Hal Casteel: you're
David Q Chen: it. I was like had exactly that idea which is the UI need to change in terms of the with the
Hal Casteel: Yeah. Yeah. This idea of a,
David Q Chen: data.
Hal Casteel: you know, kind of a dynamic No,
David Q Chen: It was it wasn't I think it wasn't successful. I I played around with it.
Hal Casteel: but the timing's better now.
David Q Chen: It was yeah I think this tech was the same and also I find the it
Hal Casteel: The technology is better, you know.
David Q Chen: couldn't achieve the same speed I wanted. I want it to be very responsive but I think the process of recognizing the data type and then generating
Hal Casteel: Yeah. Yeah.
David Q Chen: the UI that that process actually quite the delay is too there's actually a remaining
Hal Casteel: Well, that's why I've been running the databases in the cloud and then running that those type of things in the
David Q Chen: problem.
Hal Casteel: browser, you know, because that's,
 
 

02:13:18

 
David Q Chen: Yeah,
Hal Casteel: you know, that's
David Q Chen: I I think that the answer probably is in that you have a set of patterns of UIs and then you
Hal Casteel: Yeah,
David Q Chen: choose the ones.
Hal Casteel: exactly. Yeah.
David Q Chen: So I was like I want it exactly as
Hal Casteel: Create a bunch of re Yeah. Yeah.
David Q Chen: Yeah.
Hal Casteel: Create a bunch of React objects that are already kind of preconfigured for what you need and then let the
David Q Chen: Yeah. Yeah.
Hal Casteel: UI self assemble based on the use case.
David Q Chen: Yeah.
Hal Casteel: Yeah. I've been playing around with that.
Mike Smith: So,
Hal Casteel: I think there's a lot of opportunity there.
Mike Smith: so I know we're at about quarter after 9. Um, I don't know. How is everybody feeling about this? you want to like um I know that I know the last one we did was back in September. I'm thinking we we maybe u increase the meeting cadence to like almost like a bi-weekly meeting if everybody's okay with that.
 
 

02:14:06

 
Dunnville Grand Tour: I am
Mike Smith: And
Hal Casteel: I am and then and I'm available like Jared if you you know I I can start work
Mike Smith: again,
Dunnville Grand Tour: good.
Hal Casteel: with you immediately to kind of flesh out that requirement around the problem set that you're talking about uh because the most execution is relatively
Mike Smith: Keep keep in mind,
Hal Casteel: easy
Mike Smith: okay? At the end of the day, we want to prioritize it.
Hal Casteel: every no I want to cherrypick ideas that are high value not just to a single oneoff customer that are reproducible scalable and and you know you know I mean this is the 2026 is going to be a big year uh 2027 is going to be even bigger But if we get in now into some of these where we start defining some of these problem sets that haven't been solved yet or solved and productized you know this is you know I want to deal in a product you know I code of tech is more I want a suite basically right you know and the idea that there'll be a a platform and I white labeling is I'm a-ok okay with white labeling what I've built I've already had a Neo Bank in Brazil that's marketing uh finance products and using some of what we've built already in Chile.
 
 

02:15:32

 
Hal Casteel: So I was like, you know, they were like, you know, can we put our name on it? I said, no problem. You know, it's like I want to build a brand, but I, you know, ultimately I want to build products. I'd rather get it out there and scalable and and creating a lot of revenue in terms of product space because that's somebody's going to do everything that we're thinking about and there may be 10 people already working on it. So, you know, time and opportunity costs, you know, you you have our windows of time are getting smaller and smaller basically. So, don't think about it too long. Just get let's get to
Mike Smith: Yeah.
Mike Wicks: Yeah,
Hal Casteel: work.
Mike Wicks: I mean I for me uh I think like initially when we had that first call the the scope was to
Mike Smith: So,
Mike Wicks: target more uh regulatory um types of projects,
Hal Casteel: Well, that's my long-term vision,
Mike Wicks: right?
Hal Casteel: but the bicycles are too bloody long,
Mike Wicks: But yeah.
Mike Smith: This
 
 

02:16:28

 
Hal Casteel: basically.
Mike Wicks: Yeah, that's what I'm thinking. It's it's tough to get in.
Hal Casteel: Yeah.
Mike Wicks: I I think if you know if um if I invite Pu to the next call and let's say we we help him with a you know basic uh basic solution for what he needs and you know he's constantly going to these uh trade shows for med device aerospace
Hal Casteel: Yeah.
Mike Wicks: but you know I don't mind going there with him maybe pitching solutions to other competitors of his um but I I'll try to invite him to the next call for sure.
Hal Casteel: Okay, that sounds
Mike Smith: Yep. Yep. And again,
Hal Casteel: great.
Mike Smith: is two weeks every like bi-weekly every two weeks. Is it is everybody good with that? And again, if you can join, you join.
Hal Casteel: Yeah.
Mike Smith: And if you can't, I I'll send it out. I was having difficult,
Mike Wicks: Yeah.
Mike Smith: you know, sending out through uh Al's um his invite, but I'll I'll do it through mine so that we can kind of just kind of keep it and kind of keep the collective together.
 
 

02:17:22

 
Hal Casteel: Yeah, sounds good. And if this splinters into multiple focus groups focused on different industries,
Dunnville Grand Tour: Perfect.
Hal Casteel: that's fine, too. I mean, I I think however, you know, I think but I what my experience in the do years was I created a company called Lighthouse Communication.
Dunnville Grand Tour: Yep.
Hal Casteel: had a couple economists and what we did is we we sliced the economy vertically and then we I mean we were focused on B2B marketplaces and so you know I look for you know high value B2B opportunity I try to do B2B uh pretty exclusively uh not into consumer stuff so much just because of the the you know but anything that looks
Mike Wicks: I need help.
Hal Casteel: B2B you know count me in because but if you but if you have a consumer idea I've actually got one guy. He he it's he's a he's a guy that got out of the ar you he's a a rich kid that joined the uh army. When he got out he went back to fishing. He's got like an $8 million boat in Fort Lauderdale.
 
 

02:18:24

 
Hal Casteel: And I said, but he had an idea for an app called Hookup. And I go, what is it? He goes, it's Tinder for fishermen. And you know, it's basically SW fishermen like to fish with people who fish for certain type of fish and people with boats want to share their expenses. And I go, "Well, that's pretty cool." You know, and it was interesting because when he said what he was doing, about a half of the people in the room were fishermen and they were all go,
Mike Wicks: What you want?
Hal Casteel: "Yeah, that's cool." And then, you know, I I said I'd help him build it. So, we put it in CODATECT ch and, you know, built an app for him. He went out to Cal he's he sells solar installations for a living. So he's pretty good at sales and he went out to California and was talking to he you know
Mike Wicks: f***
Hal Casteel: he's the people who supply bass boats to the industry.
Mike Wicks: you.
Hal Casteel: One of the owners of one of those big companies was a friend of his that he fishes with you know wrote him a check for 75 grand.
 
 

02:19:21

 
Mike Wicks: All
Hal Casteel: I was like, you know, it was a casual conversation in about 15 minutes of, you know, development work in CODATECT ch and he managed to land and then he then he talked to six more people and raised another half another half a million dollars.
Mike Wicks: right.
Hal Casteel: I I was like, I wouldn't do that. I'm not raising money, but the fact that he could sell it impressed me. So, if you got ideas, you know, I can build it and help you uh sell an idea.
Mike Smith: Mhm.
Hal Casteel: I'm all for that because, you know, people need to see what their what their, you know, their ideas need to come into fruition, right? I mean, it's like can't sell. Um, anyway, Mike, thanks for bringing us all together. I really appreciate it. Sorry you guys have to suffer through the snow up there, but
Mike Smith: Uh, it's okay.
Dunnville Grand Tour: They love the
Mike Wicks: Good for seeing. Oh.
Mike Smith: We get four We get four seasons up here.
Dunnville Grand Tour: snow.
 
 

02:20:20

 
Hal Casteel: Yeah, I'm going to be down here in Brazil until I don't know when,
Mike Smith: So
Hal Casteel: but definitely not. I mean, we had frost in Orlando and I I went to Brazil at Christmas Day, so
Mike Wicks: I uh I did the drive from Orlando all the way to uh Miami a few years back.
Hal Casteel: that's quite a drive.
Mike Wicks: Was a fun fun drive.
Hal Casteel: Yeah.
Mike Wicks: Yeah.
Hal Casteel: Yeah. Yeah.
Mike Wicks: I call uh Orlando the uh the armpit of Florida.
Hal Casteel: Well, it's interesting. It It kind of is. And that's my impression, too. I moved here from Santa Monica, California, which is the arm It's not the armpit of LA, but LA is the a****** of America, quite honestly. Right, if you ever have to deal with the 405 or the traffic there. But, you know, when I when I when I moved to Orlando, I was like, "Holy cow, I've gone into the you know, what what's a tech guy doing in Orlando?" Right? And uh you know, but now that I'm getting to know, you know, like we have SpaceX, we have the Space Coast, we have we have uh over 3,500 medical device companies and the you know, there was a whole lot more on it's the third, you know, the third largest state.
 
 

02:21:34

 
Hal Casteel: Uh we have uh central Florida has become what they call a tech corridor. We're kind of you know we've got a whole lot of tech happening but it's not connected like in the Silicon Valley where which is what I'm used to is you know like San Francisco or places
Mike Wicks: Yeah.
Hal Casteel: like New York where you have a density you know you have to drive and like you were saying that drive from Orlando to Miami this this
Mike Wicks: Yeah. I I I didn't like I didn't realize the Florida showers were like a real thing. Like I thought it was, you know, overrated until it happened to me.
Hal Casteel: Oh, no. You want to get off the freeway when it rains,
Mike Wicks: And Yeah.
Hal Casteel: right? Yeah. I mean, it's dangerous,
Mike Wicks: Yeah,
Hal Casteel: right? Yeah. I mean, it's
Mike Wicks: a lot of expos are are constantly happening in uh in Orlando. A lot of the process
Hal Casteel: Well, yeah. No,
Mike Wicks: exposing
 
 

02:22:29

 
Hal Casteel: I mean, we Yeah. And we have uh something called Neo City now, which is uh when Biden was in power, they they put up two billion dollars into a chip fab. That's a custom chip fab that's, you know, like defense quality. It's the only fab in the country that can do custom chips that meet defense standards. So, there's things happening around that I, you know, it wasn't until like I was doing entrepreneurial work in uh Central Florida. I wasn't raising money, but you know, I've got a young guy that's working with me that's my you know, kind of out, you know, he's he's the legs out in the community and you know, I start I ended up being a a speaker to an economic forum uh for AI and there were hundreds of people showed up. So the and they were all business and development and you know, a lot going on. So there's a lot more. You know, the I think if we look in our own backyards, we're going to find everybody's going to find a lot of opportunity.
 
 

02:23:27

 
Mike Wicks: 9:00.
Hal Casteel: But then what we should be doing though is really looking at how those opportunities play out into verticals and product. I mean that I'm with Mike on that that we want to really look at. We only have so much time and we should really focus on things that are going to re, you know, bring high value. Uh, you know, it's fun to make money and it's fun, you know, I want to work on things that make a difference basically, not just moneywise, but hopefully do some good. Yeah, I'm old school do goodter kind of guy.
Mike Smith: Oh, good. Uh, okay.
Hal Casteel: Well,
Mike Smith: Any any other questions, guys? I know we're kind of went way over, but uh,
Hal Casteel: yeah. Well,
Mike Smith: Attilla, you're you're
Hal Casteel: and individually if feel free to reach out to me at any time,
Mike Smith: good.
Hal Casteel: you know, I'm happy to have a meeting. I try to keep my calendar very free and I'm, you know, I'm two hours east of East Coast down there right now.
 
 

02:24:28

 
Mike Wicks: Yeah, I'd say for me it's mainly if if you can send me some different ways of like it connecting to
Mike Smith: Okay.
Hal Casteel: So,
Mike Wicks: Microsoft solutions because my whole like the last six years I've just been working with Microsoft partners, right? So a lot of the the people I meet are are heavily invested in that platform. So something where they could work work with you as well. Like that that's what I'm kind of
Mike Smith: So maybe I don't know maybe in the next session maybe I can work with Hal how about
Mike Wicks: after.
Mike Smith: we go through just you know just a setup like
Mike Wicks: Yeah.
Mike Smith: how you're if you're okay with that just how how you know
Hal Casteel: No, absolutely. Yeah. And I I'm going to have I'm going to have three different versions.
Mike Smith: just
Hal Casteel: I mean, I right now I have a install script that will install Codec on your local machine just like uh running in cloud code. Um I'm I'm developing a docker that you can run locally and then I'm I'm working with Gemini or not Gemini but uh these Google cloud workstations uh which are you know browserbased uh development.
 
 

02:25:37

 
Mike Smith: So why don't we why don't we do a session where okay this is and kind of just step go through the that you know how do you get it up and running and then and then we can
Hal Casteel: Yeah.
Mike Smith: again circle back and then
Hal Casteel: Is everybody running m Is everybody running max?
Mike Smith: it's
Hal Casteel: What What kind of uh uh hardware are you guys running?
Mike Smith: I think my side everybody they're running probably Windows
Hal Casteel: Windows.
Mike Wicks: Yeah. Yeah.
Hal Casteel: Okay,
Mike Wicks: I'd say primarily wind. Yeah.
Hal Casteel: David, you're running a Mac for sure.
David Q Chen: Uh,
Hal Casteel: I'm sure.
David Q Chen: I have I have all of the systems, but I'm primarily using my Windows these days.
Hal Casteel: Okay.
David Q Chen: I run Windows,
Hal Casteel: All right.
David Q Chen: but I also run the Linux subsystem underneath
Hal Casteel: Yeah. Yeah. I mean, it runs on anything.
David Q Chen: it.
Hal Casteel: I mean, I've run it inside of the WSL on a Windows machine, you know.
 
 

02:26:28

 
Hal Casteel: I mean,
David Q Chen: Yeah.
Hal Casteel: it it's it's all it all works. I mean, okay. Yeah, that Mike, that's a good idea. And uh
Mike Smith: So I can I so let me I can work with you over the next uh say two
Hal Casteel: gives
Mike Smith: weeks and then maybe that next session we just kind of go through okay this is this is what it's going to take this is get it
Hal Casteel: Yeah.
Mike Smith: in this this is what it would be to get it up and running. Um, and uh,
Hal Casteel: Right.
Mike Smith: yeah, then then the feedback loop. Then we just kind of get that feedback loop back to Hal as we're kind of throwing those ideas into it.
Hal Casteel: Yeah. And I think once people start playing with it, you start seeing your own ideas come to light. That's when things get, you know, real fast, basically. So, all right. All right. Well, you have a It's 11:30 here. I'll say good night.
 
 

Transcription ended after 02:27:39

This editable transcript was computer generated and might contain errors. People can also change the text after it was created.