S5 E10 - Doug Finke – From Mainframes to Machine Learning

In this episode, I have a conversation with Doug Finke. A 16-time Microsoft MVP and long-time technologist, Doug’s experience ranges from mainframe assembly programming to pioneering AI integrations in software development. We reminisce about the early days of system programming and explore how those foundational skills have enabled technologists like Doug to excel in the rapidly evolving world of AI.

Doug shares his journey from COBOL and assembler to becoming a PowerShell innovator, highlighting how his early exposure to structured programming and domain-specific languages now serves as a superpower in working with large language models (LLMs). Our discussion pivots to the game-changing potential of AI, specifically the rise of function calling, modular design, and the Model-Context-Protocol (MCP). Doug explains how these patterns transform AI from simple copilots into powerful collaborators capable of orchestrating entire software systems.

We also look at the implications of AI-driven software development for enterprises, examining whether vertical SaaS solutions may soon be disrupted by in-house, AI-built alternatives. Doug emphasizes how AI accelerates both code creation and maintenance, challenging long-held assumptions about whether to buy or build software.

Our conversation concludes with a philosophical lens on teaching AI, the evolving role of junior developers, and the importance of naming, structure, and design patterns in crafting software that AI can effectively understand and extend. Whether you're deep into AI tooling or just starting to explore, Doug's perspective offers a clear bridge between classical computing and today's transformative technologies.

Transcript

John Willis: [00:00:00] Hey, this is John Willis. I got another great guest today and we're gonna have a lot of fun I think this afternoon. Doug Fink, you want to go ahead and introduce yourself? 

Doug Finke: Sure. My name's Doug Fink. Been in the industry for, since mainframes and came across neural nets years ago. I thought it was too much mathematics.

So back in 21 as chat, GPT and Bert and all that, you know, I've always focused, I always like looking at and adopting our tech technology and then all of a sudden this thing called copilot hit. It was like, wow, this thing is crazy. And then it mushroomed from there over the last three, four years. So I'm a software developer.

My focus, I wrote a book about PowerShell for developers back in 2012. So yeah, I love tech and it's really great to be here today with John. And like we said earlier, we have similar backgrounds that we play with similar toys back in the decades. 

John Willis: Yeah, no, I think , obviously your presence on LinkedIn is is really good, right?

Like, in other words, you just. Bubble up and I see your posts and all, and I, I think I made some reference to like some really early programming [00:01:00] languages that were more cis admin. Like things like, you know, if you go back generation, like, you know. Decades of generations, you know, pre PowerShell, you know, if I want to be, you know, we call 'em C list and, and stuff and back in our day.

Because it looks like we started about the same time in this career. Exactly. And like you were either gonna program, in fact, back then you had a limited set of languages. And when we started, right. I started 

Doug Finke: as a I had to go to, I graduated high school. I didn't, I went to, I went in the front door and out the back door of college.

Okay. Said, I worked at Depository Trust Company, midnight to eight as a clerk. Oh. And I just started saying, yeah, I wanna be a programmer. I don't even know why I said that. Sure. And one of the folks on day shift said, well, if you go to either NYU or Chubb Institute and take a six month class and do well, we'll con, we'll think, we'll see if we can hire as a programmer.

It was three months COBOL. Three months assembler. Okay. Yeah. Yeah. And then when I went for interviews, all the COBOL people were like, you don't know anything about cobol. What'd you learn in class? I'm like, I don't know. And then they hired me in the systems group, which was all [00:02:00] assembler. Sure. And I was, and I hated assembler.

John Willis: Oh, you hated it? Oh, I loved it. It 

Doug Finke: was good. I mean, I did my job to a degree, but, and then I came across C List and TSO, and I was like, oh my God, this opens up. And then every other, every guy, it was like 10, 15 in the group. They hated me. Because I could knock out things faster. That's, 

John Willis: that was the thing that was, yeah, I think it was like, I was sort of joking about how, like, you, you young kids, you don't even know how we had to do variables that could steel us.

Right. Ampersand word, you know, name Ampers amp. Ampersand, ampersand. Name ampersand. City Amp. Like, like insane symbolic competition. But I, I'm sort of similar to you in that, but I. I sort of got routed as an ops person, but like back in the day, I, I, same thing. I, I kind of, I love that going in the back door.

Came out the front door of college. I just very similar, but I just got thrown in at Exxon and they were growing so fast down in Texas with all the geophysical processing. I mean, before [00:03:00]you know, if anybody even blinked about coding, they, they gave you here, tried writing this, assemble a program, you know, and so I was, you know, I spent my first at least five years just heads down writing operating system code mainframe assembly, yeah, yeah.

Mainframe, IBM mainframe literally. I mean, like, you'll, you'll like, I think we're gonna, we'll be the only two listening to this one, but you know, we, we had all these problems 'cause we were doing geophysical processing and like, so every time a new operating system come in, immediately we'd have to patch like block sizes, like, you know, 30 2K block sizes.

I mean, we needed like 4 million block sizes 'cause we're doing reels of, of geophysical data. So they, the, the young kid, like, I'll do it. You know, and like I was patching operating systems like, you know. Wow. Right outta high school and

Doug Finke: wow. That's basically what, that's basically what I was, I was like, I was building mainframes with nine inch reel tapes and I had no, they're like, mount the tape over there, we're gonna do this and that.

And then I'd learned how to do it. And then we were all IBM. And it's funny you mentioned like Exxon, because we, [00:04:00] I was the youngest guy. I was 20 when I started doing it. 

John Willis: Yeah. And 

Doug Finke: most of the guys in the group were 15 to 20 years older. And they were from like JPL and from Texas Instrument. Yeah, yeah, yeah.

Because they wanted to make money on Wall Street. And they weren't traders or anything, but they made more money doing this kind of stuff. But yeah, make building operating systems and all that kind of stuff. Fascinating. 

John Willis: Yeah, no, it was, it was a good time to get sort of baptized into, I mean, I can, I think one of the things, you know, I want to get into your thoughts about AI and obviously to be a bulk of this, but as we just have a little fun here, reminisce and you know, I, I think that's, like, I feel blessed that like, I literally had to patch operating systems when I was 20. You know, same kind of age as you you know, and, and, and write systems. I, I, so I get this little weird bragging, but it's actually pretty trivial. I wrote the first chargeback system for a Cray because it, but it wasn't really, you know, it wasn't really, wow.

It was like a 10 line program that, that captured start and ends. And then ex Exxon needed a billing system. For like, you know, pre early [00:05:00] BSD Linux or whatever. Cray ran and, and they just had to have, if somebody was gonna run a job, they had to have a billing mechanism. So I literally wrote a, a system that captured all the starts and ends.

'cause there were no accounting like mechanisms. But, 

Doug Finke: but yeah, I mean, just being, and just the last piece of nostalgia on that is. I was, you know, I cut my teeth on IAMV, A M and Bs a Oh, sure. 

John Willis: Yeah. Yeah. 

Doug Finke: And then, then one day IBM walked in and said, we had this new thing called DB two and we're like, bad.

And they were like, oh, you gotta learn sql. And all the old timers were, were, were raking 'em over to calls, like, why'd you do this? And the partitions, and it's slow and it's that, and I hated SQL and I hated DB two because of that. Yeah, no, 

John Willis: I, I, you know, like I, you know, you just, a word I haven't heard in years is Visa and virtual systems access method.

I think it was, it was clever stuff. I mean, it was high level indexes. I mean it, you know, a lot of the world that probably still runs on that stuff maybe. 

Doug Finke: Absolutely. When we got the first PDAs back in the late nineties, it was basically a BS a M type [00:06:00] interface to save the data. So I used to plug in my little wire link and I would upload my PDA calendars and whatnot, and I would program against it.

'cause I understood the, I knew, yeah, yeah, no, it's this format, it's it all set into here. Grab this. And people are like, oh wow. And I was like, you should write a product. And I'm like, ah, 

John Willis: we're definitely gonna have no listeners here. But I also wrote a full screen editor for vsam. I, yeah, yeah. It was with a couple of people, but like the idea that you, 'cause you, you know, the, they remember they, they first you had to do like command, like we, we had in IBM land back in sort of early eighties what would look like vi but like it was, it was sort of a command line thing, but very much you sort of line edited and, and then they came out with a full screen editor.

And you know, remember ISPF and SPF and all that stuff, right? And Yep. But you can only edit like sort of text files or sort of basic files, right? But, but you couldn't really edit a v sem file. So like me and a couple people, wow. We tried to turn it into a business that didn't go anywhere, but interesting.[00:07:00]

So, yeah. All right. The other thing, I guess the, the other thing I wanna point out is I didn't realize the significance of your background until I did a research you know, correct me if I'm wrong or whatever, but a 16 time Microsoft MVP, according to chat, GPT is a very significant and accomplished because it looks like it has to be 16 years in a row.

Doug Finke: Yep. 

John Willis: 16, you don't automatically get the, you know, the next year. And there's few you have to consistently 

Doug Finke: contribute. Yes. 

John Willis: And there's fewer than a hundred global like that have that type of stature. 

Doug Finke: Abso absolutely very small group. Who, who lasted that? Yeah. And that was all basically 'cause of the work I started doing in PowerShell early on, because the only way you could write functions in PowerShell in version one was the uc Sharp.

Wow. Okay. So I was able to Okay. Step over both worlds. And then just dialing back on that, people used to, I was a visual basic consultant for years in Manhattan at in Wall Street because I could outdo c plus plus folks. And people were like, [00:08:00] oh, why weren't you an MVP then? I'm like. You could be an MVP in the nineties, like Absolutely.

Okay. So I could, I was like, wow, I could have been done one for 30 years. I'm like, nah, that's a, that's not ANU number. I want to hear. Yeah, yeah, yeah, yeah. 16 years. So yeah, it's good stuff. 

John Willis: Yeah, no, the PowerShell stuff is, was, you know, like pretty cool. But Jeffrey Snow, did you work with Jeffrey Snow at all?

Yeah, yeah, I did. 

Doug Finke: I did. I wouldn't say I worked with him, but he came to my user groups here in Manhattan. Okay. And he would present before he moved on to other stuff? Yeah absolutely. In fact, with the PowerShell, one of the things I did when I came across ai, we'll, we'll finally drizzle some AI in here, is that I ported the open AI Python, SDK.

Power shell. Yeah, 

John Willis: no, I've seen it. You've gone really deep into probably my, one of the more prominent authorities now on AI and PowerShell as far as I can tell. I got one story I'll throw out there and it's sort of like, yeah, just Snow was an interesting guy. I, I did a podcast with him way, way back when, when I used to DevOps Cafe and and, and I, [00:09:00] I, I think when he said out.

When we weren't recording the, the story was that he had gone through, gotten to I to Microsoft to an acquisition. Mm-hmm. And he was put in a pretty high level position and he started working on the original that was sort of like, it was a product project name or something, which was ultimately PowerShell.

And, and he says some high level manager came to him and said. What the hell don't you understand about Windows? Take a secret. I think he told me personally it was Balmer, but you know, like when he puts it on, you know, his videos, he says a high level, but he said he was like the guy that nobody wanted to be nearby the coffee pot, you know, and he take a demotion to finish the project and like it changed Microsoft.

Exactly. You know, like it, like they're not part of the game, I don't think. He doesn't do what he did, you know, getting them in these bandlines and like the world was changing. 

Doug Finke: Absolutely correct. Like 

John Willis: Unix desktop, not Unix, command line, all that stuff. It, it just, 

Doug Finke: he said he, his point [00:10:00] of view, and I used think the original name of it was Monad.

Yeah, monad. They, 

John Willis: sorry. That's right. Yeah. Monad. And then 

Doug Finke: they called it PowerShell. And he came from Deck and he came from IBM m Yeah, that's right. That's right. Yeah. And then that's how he got into, and then he got into Microsoft. I think he was. He used, I'm not gonna say he was in the office of bill Gates, but he was pretty close to it.

And they were, yeah. They would say, why are you doing another command line? Why don't we just take a Unix one? Yeah. This his windows. He's like, and he's like, no. And he's like, I don't wanna do, I don't wanna do text-based parsing. He goes, I'm gonna do this thing called object based. Across the pipeline, which blew my mind.

'cause I came from the mainframe was like, well, yeah, you take the data, you pass a text, and you, you do your, your sad gr and a and then all of a sudden I'm watching him pass objects and now you could talk to an object with a.property. I'm like, oh my God, what is this thing? Anyway, so I got addicted to that.

But yeah, and Jeffrey Stover was the brainstorm behind that. And he put an amazing team together over time. He had to scrap the first one, and then he did the second one. But yet everybody at Microsoft, even to this day. PowerShell is considered kinda like a red redheaded stepchild. Oh, [00:11:00] really? 

John Willis: Oh wow.

Oh, wow. I mean, it's 

Doug Finke: great and, and it's heavily used in the IT space, but it's in the, in the CIS admin space for all types of stuff. But I look at it as a programming language. Like I wrapped, I took Python and ported it to PowerShell so I could do AI in my scripts and at the command line. 

John Willis: Yeah, that's crazy.

Yeah. One last thing. I wanna, and we'll, we'll go, we'll really go into the AI here thing after, I promise everybody. You, it seems like you've had a really long career in banking that sort of fascinates me too. Or the financials, right? I, I, you know, I sort of, in my last. A couple of projects, you know, mostly around writing, but that turned into automated governance.

I got to work more with banks, but banks were just fascinating from an IT perspective, I think. Right. No, 

Doug Finke: Well, it's interesting. It's always interesting. Like I used to look at Microsoft and go, oh, I want to get into the big house. And then I would talk to people and get to know people and, and the stories I was like, oh my God.

And then they would be like, Hey, I've got this new idea to do. Option trading. I'm like, yeah, that'll, that'll never fly. 'cause the trade traders are X, Y, and Z. So I was never really I was, you know, I did [00:12:00] trading systems and whatnot, but I was never really next to the traders and I never really did that kind of work.

Mm-hmm. I would do, like, we would go into a Citibank and we would take a 20-year-old Java system and try to port it into WPF and C and that was what we get contracts for. And that's the kind of work that I did. So, yeah. They were, you know, they were typical. It, like they you know, they, they had a job to do.

They built systems, they had to maintain 'em for decades and they had no time to learn the new stuff to keep moving forward. Kinda like what's happening today with ai. Sure, yeah. But it's a cool, it's, I mean, I live in Manhattan, so it's like, you know, I could walk to many of my contracts in different places.

I 

John Willis: did. Yeah. That's fun. 

Doug Finke: So, but yeah, it's an interesting, it's an interesting, interesting space. A lot of unusual people. 

John Willis: This AI thing, 

You know, I saw some of folks, I hear it's, I hear 

Doug Finke: it's a fad, 

John Willis: yeah. Something's going on here. I don't know, like, help me out here. What, what is, I'm an old timer like, yeah.

No, I mean, I think you, I was looking at something I wanted to kind of drill down to like even some of the crazy stuff. Like, but, but at a [00:13:00] high level, right? Like, like if, if like we're beyond the Kool-Aid, right? And like this is real. And, and like wake up people that like, I mean, what I think you said had something that we're gonna turn into like, you know, 10 million to a hundred million to billions.

I mean it like the, the, the impact on software development. So you talk a little bit about, like, your thoughts about the impact or like, like for people who are sort of naysaying or like, I mean, we're seeing less of the naysayers. I mean, like at the end of last year, they're like, God, that cursor that, you know, the, the co-pilot's terrible stuff.

And now we're, it seems like there's fair amount of people on board on like that, like, and you know, companies are, are saying like, a 40% of our code, whatever. Right. But mm-hmm. What's the reality check from your perspective? 

Doug Finke: It, it again, I'm gonna say it's hard for me to do a reality check because I'm completely addicted and I don't have to, you know, I'm not reporting into somebody who says, Hey, I need this by Tuesday.

[00:14:00] And, and by the way, you gotta do your regular job. But the a hundred million so Chamath on the All In podcast, I, I think you, you reposted one of the things that I, yeah, I, it wasn't, I dunno if that was a specific one, but he makes the point that. In his company, I think it's called 80 90, they're building a set of software that is it's not just about copilot, where it's like an individual contributor can do amazing things.

It's like they're doing the entire the entire SDLC, right? Mm-hmm. From PRD to fine tuning to QA to UAT to deployment. And at every step he's saying is you see a significant, increase of capability and throughput. Right. And I think he might even use the number like 70% and that it compounds over the entire project.

That's what I'm seeing. I, I sit here. You know, if, if I took a script, like I don't write scripts, like one big script, like a hundred lines of code and it does a bunch of stuff and it has global variables, I've never developed like that. I always develop things in functions and [00:15:00] modules so I can plug and play and mm-hmm.

That, that's my background, so I do that. But AI works with that unbelievably well. Because it says, oh, I can make, but if I gave it a hundred to 200 lines of nonsense or a bunch of stuff that's doing this and that and the other thing, it gets lost. It's like it doesn't, what are you trying to do? So it's not unlike a human.

If I gave that to an intern or I gave it to a mid-level software guy, they would be, you know, I don't understand what this is doing. I have to figure it all out. If I touch this, I get a side effect, all this other kind of stuff. So I approach it that way and when Chamath talks about it. I go, oh yeah, if you do, and like we've, we've, we follow how do you say his name?

Ru Cohen. Yeah. Reuven. 

John Willis: Yeah. Reuven. Yeah. 

Doug Finke: Reuven. Right. And I get what he's doing right, i's like, yeah. If you, if you take typical AI really responds to well structured code. Most of the world is not well structured code. 

John Willis: Right, right, right. 

Doug Finke: If we're gonna sit back and we can say, Hey, you know what? Let's soup to nuts.

We can start rebuilding this because we have these tools. We can have a team that was supporting a [00:16:00] hundred million dollars worth of throughput to do it at a $10 million cost now. Right? It's not gonna, and then you have that order of compounding capability, I 

John Willis: think. Yeah. That's what like, I mean, you know, as I'm thinking this through, right, like one of the things I sort of went.

Reasonably deep. Early on I was working with a client where, you know, function calls were very important. Right. And, and you know, that the, the whole, I think it's almost like as I hear you talk through it, it's like the inference capability of AI or sort of lms, whatever we want to call the latest, let's call, say consistent our discussion.

They're really, they work really well. With well-defined functions. Right. Even on the inference level. Right. Because, you know, we'll get into MCP in a minute, but, or a little, a few minutes, but, but this is, 

Doug Finke: this is, this is the starting point function call. So go ahead. Yes, that's right. That's 

John Willis: right. And, and, and what I think a lot of people, maybe a lot of people do understand at this point, but the idea that, that, that that whole system works really well is if you're, if you create the right context for the [00:17:00] definition of the function as opposed to like, I have to go to a, b, C 105.

Right. Or, I mean, that would be a terrible naming and by the way, there are a lot of those, it wouldn't even matter in, in sort of the inference model because the, the way it would find it, it wouldn't be like, I have to look for an index. I gotta find what the function is.

All the sort of like bad or quasi good architecture decisions we've made over the years. Not really. You just need to define the function and be into the sort of just right, the goldilock zone description of it, and AI finds it. I need 

Doug Finke: to talk to you more in the future because I use Goldilock Zone a lot, but you're exactly right.

And one of the cool things about function calling. And I ported that back in 2023, and I started seeing Unbel like, like Cohen says, I'm seeing this unbelievable stuff and I wasn't doing swarms or anything of that. Right, right, right, right. But if you define the name correct correctly, and if you define the parameters, and if you define it's the types of parameters, and you give it, if you have help.

For your function and you include that, it becomes part of the [00:18:00] context to the model, to the LLM. Right? And then it goes, it can then begin to figure things out. But if you're gonna call it function A, function B, and it has parameter, A one and a two, and it's, yeah, it's, and that's what people have done over the years, over the years.

Right? Back in the day, we used to say, if you can't name your parameters and your variables, you don't know what you're building. I took that to heart. A hundred years ago, I was like, oh yeah. And then they're at and cradle in the grave naming. Right. You don't call it like first name here and then person at the end.

Right? Right. But if you can, and I took that to heart a hundred years ago. You give that to LLMs and they go to town. 

John Willis: You know, it's interesting too, like stepping back a little about our early discussion. I think one of the reasons that maybe that worked really well for you and I, I don't do as much in anger coding, right?

I have coded quite a bit in my career. Every once in a while I dip into an interesting project. I'm doing a little more coding now over the last five years than I've done probably the last 20 years. But one of the things we learned early on was to see importance of naming. And I think procedural languages, you know, sort of things like, you know, CLI or the, you know, scripts and, you know, [00:19:00] precursors to scripts and PowerShell, right?

Like you could easily see the power of those becoming architecturally. Strength Strong because of the way you defined it, as opposed to just using these sloppy methods used with high level language. So I think there was something there 

Doug Finke: absolutely. 

John Willis: In putting us and when 

Doug Finke: and when, like I grew up on Strauss struck when he invented c plus plus and then later on Eric Gamma, when he came out with design patterns and I think was a Kent Beck.

He wrote a Forward to one of your books. 

John Willis: No, no, it wasn't Kent. No, no. 

Although I, I do have a great 

story about Kent Beck, I met him for the first time. Nice. Last year it was a, I can't remember it.

Maybe it was early. No, it was early this year, and I was giving people copies of my AI book and but, but I always bring a couple of my Deming book back with me, and so he comes up to me, he goes, do you mind? Yeah, this is Kent Beck saying this. He, do you mind if I can get a signed copy of your book? I'm like, nice.

Whatcha kidding. Beck just asked me for a sign, and then I went to pull out the AI book and he goes, no, no, no, no. [00:20:00] The other one, the Deming one. 

Doug Finke: Really? 

John Willis: Yeah. Like that was like a moment, you know, like you put that one on you, you somehow store that in those moments.

Doug Finke: And 

John Willis: it, 

Doug Finke: it's, 

John Willis: and, but it's, 

Doug Finke: and 

John Willis: it's 

Doug Finke: those and it's those folks.

Yeah. Yeah. That impressed me. Yeah. Yeah. Back in the nineties and then I used to write programs doing that and people go, people you know, that knew me in the software space, they were like, wow, your stuff is easy to read. I'm like. Yeah. And then, you know, I come from the Donald Newth or Knuth, Donald Newth you when he use literate, literate programming, right?

Like you wanna write it so you don't have to write documentation. And I used to sh that was like. That was my goal, that was like, oh, this is gonna be fun. I'm gonna write it like Donald New said. Right? Yeah. And and then it's like, holy mackerel, LLMs, they, they're gonna, they're gonna eat this up. Yeah, yeah.

John Willis: No, the whole, and I, I think I, my first project where I really understood, I've done a bunch of hackathons for like MongoDB and different clients and 

Doug Finke: mm-hmm. 

John Willis: We were doing a sort of a sponsored hackathon between Google and and MongoDB, and they really wanted me to come up [00:21:00] to speed on vertex for the, the.

And the way they did their function call was like, not crazy unique, but it was like, it really better than even the way I learned open AI's functions or, or lang chain tools. It was, there was, it's the clarity of how this all worked with maybe just me and the way I think, but, but it was Vertex that really sort of like, wow.

Now I, I kind of absolutely get the whole thing. I mean, 

Doug Finke: and then they, we also talk about it, you know, we should talk more about function calling, but that the you know. When you actually build out your API, right. At some point when you're using it in your code, it should look like a domain specific language.

Yeah. Yeah. And domain specific languages for me were, you know, I'm not a, again. I'm not a classically trained software developer. I just learned that stuff and picked up little bits in here. Yeah. Yeah. Here and here. But DSLs man, it's like, and then again, you give that stuff to an AI and it goes, I gotcha.

Yeah, 

John Willis: these kits very powerful. Right. And so, so then all right, so a couple more things and then I want to circle back to the function calls to MCP. But before we get there. [00:22:00] The other thing I saw you talk about, and I, I think this is, I think, and again, it's probably part of that podcast that you, they all in podcast, but we're talking about like the the, I think like this idea, like could we, or are we, and I think.

Like the die is cast start thinking about replacing all the vertical software out there. And like, I, I think I, I say this all the time. I get in arguments with vc. There is no moat right now. You know, I mean, you look at what Reuven's doing, right? Like with that stuff, like, like, you know, I, I think about the, the, and then you, you guys really honed in on what people don't realize is.

If I could rewrite some project, I, I, I've, I'll say this, I don't know specifically, but I've heard a story where a large organization is using their in-house sort of developers that they were using, you know, and there's a, let's take a few of 'em. Let's try to replace Workday. Right. I, I think like, because people don't realize, or they do.

Like the bloat, the [00:23:00] cost and the people cost, especially when you have a lot of vertical solutions. And then there's the integration cost. Exactly like the people, the nasty guys like us who build all that, you know, scaffolding for these companies. You call

Doug Finke: it, you call it scaffolding. It's not, 

John Willis: I have called it scaffolding in the past.

Yeah. But look, there's a significant cost. And I'll say one more point and I wanna hear your thoughts about, is that the other thing is. One of the arguments for using vertical software or sort of added package, like don't build it yourself, you know, was the maintenance and keeping up well. I mean, like, I've been following some of your posts and, and I've been obviously seeing a lot of posts about how much AI not only helps us build software.

How it may, it equally is strong in maintaining and updating our software. So like that, that problem of like, I don't want to be in the business of, because I don't wanna support software. I think that's going away. Absolutely. So I think there's a, there's just a [00:24:00] whole you know, I think the, the economy of vertical software is just gonna change dramatically.

Doug Finke: I, which is one of the reasons I started putting out what Chamath was saying. 'cause he really succinctly and, and crystallizes that, at least for me. Yeah, I mean with, with using ai, not only is it good at creating stuff you need to have, it's a different mindset. Right. And I don't sit there.

You know, like, like, like Cohen talks about like when he builds his code, it's doing checkpoints into, into version control. I've been doing that for a year plus, right? Like I was vibe coding before it was called that, and not that I'm. Whatever, but like, oh wow, if I do this, and then I go, oh, it works. And I ask you to do tests first and then write the software like we used to learn from Uncle Bob Martin.

Right. If we do it that, and we tell LLM to do that, oh my goodness, it does really ama and then it's, if it's working with green tests, I check it in and then I go to the next step. So when people go, oh, it went off the rails, I'm like, well, did you check it in? Oh, I don't use version control. Well, you're logged then you're, you're [00:25:00] not even a software developer, in my opinion.

That's right. But now if I'm gonna look at some old code, new code, whatever it is, I can let, I can go to the LLM and say, look at this code and I wanna do A, B, C, and D. Don't update anything, let's just chat about it. Mm-hmm. Mm-hmm. It can go through all the code and begin to pick out different pieces I wouldn't even be aware of.

Right. Whether it's my code or somebody else's. So in terms of maintenance, it's off the charts. SaaS is gone. I mean, and it made sense back in the day. It was a good argument. Yeah. Yeah. No, I, I buy before you build, you know, buy before you build, but not anymore. 

John Willis: I, I, I think there, you know, I think that, like I said, you know, my sort of my favorite saying these days is, there is no moat.

You know?

Doug Finke: The only thing I wanted to add on that is like, I think there's, you know, the people that I get a lot of pushback, even if I just come out and say, Hey, this looks great.

Then people come out and say, oh, it's just all hype. It's gonna fail, and then you're gonna fall on your face when it fails. I'm like, I'm just telling you, they think it's cool. Yeah. And like, check out what you know and say, oh no, you're cherry picking. I'm like, I just can't have these conversations anymore.

It's like, 

John Willis: yeah, no, no. They're, they're, they're a waste of [00:26:00] oxygen actually. Yeah, I mean, like, I periodically I try to sit down on Reuven's sort of Thursday hack, you know, AI hack. 

Doug Finke: Yep. 

John Willis: Hack Day or hackathons. And it was the last one I was in, it was pretty interesting 'cause he was talking about Claude Flow and, and it was one of the, like, he has a couple of guys that are just like.

Like really the, the freight train is rolling. These guys are like fixing and there like, there's four or five of 'em at least that I can see that are just, isn't just Reuven is he is. And you know, the, the one thing I keep hearing about people saying, well, well one of the things they say about his code is, well, it doesn't work on my machine.

Well, tough. I mean, I guess you're not really a coder, right? Like, because you know, it didn't work on your machine when you screen scraped it off the, you know, like, you know, stack overflow, right? Like, you know, so yeah, like figure out how it works. And, but the other thing, every time somebody asks a question, one of his, like, top chief gurus, like in part of this, sort of this army that he's creating, he kept saying, well, I just, you know, like, well, how would you turn that off and turn this?

I just asked [00:27:00] cloud flow. I just asked like, in other words like stop complaining about like, the code might not be right or the code might not work, or the code that Reuven Cohen wrote, you know, may not work on your machine. Like, you know, like just go ask AI to fix it and just go ask guys explain to why it doesn't work and stop fretting over this like, nonsense boogeyman, you know that Exactly.

Doug Finke: Or like, oh, I have to know, whatever little thing, you know. It took me a while, like, you know. I would sit there in 2022 as I was building this out, stuff out and trying things. And then, you know, first hearing things like the word prompt engineering or hallucination, like what the heck is this? 

John Willis: Yeah, sure.

But 

Doug Finke: I would think about something and then I would start doing stuff and then I'd watch copilot just pop up and start helping me. And then I was like. I would go, it took me a while to go, oh, let me ask AI first. Right? And this AI first, and then people start talking about AI first mindset. It's a tough habit to, to create because it's like, oh, I just want to go.

I don't, I wasn't even using Google and, and Stack Overflow by the end [00:28:00] of 22, 23. I would, I would go to ai, but to think about saying, well, how would I build this? Or, Hey, go build it for me. That's a whole different set of. How it feels and whatnot. Yeah. And then like I think I mentioned when, when you posted about what he was saying and getting those questions, I, for a while I would, I still do, I would go on, you know, X or one of the social media platform, people would ask a question, how do I do blah, blah, blah.

I would copy their post. And I would drop. Yeah, I saw that ChatGPT and I would put the answer back and they go, yeah, I saw that. How did you do that so fast? That's great. I don't like, I used ai. Yeah. And then, and then people would get angry, 

John Willis: you're cheating. Yeah. No, no. Like, but that's like, dude, like you're sort of, it's like, you're right.

It is sort of that like. Even like stuff, when I'm doing, I do more, I do probably more research because I'm just, I'm like the, the cent, the the classic dog squirrel, you know, squirrel. Squirrel. You know, like I, like any given day I might be like, I'm gonna figure out quantum computing today. You know?

Exactly. And and I've had a, I've had a career, you know, blessed career that I can do those kind of things. So I, I find, but it took me [00:29:00] a while to, to like stop going to like even Google. Like, why am I doing that? You know, why? Why am I not just asking AI first? Just stay there, you know, 

Doug Finke: and have a conversation about it.

John Willis: But same thing with coding too, like that, you know, listening to that gentleman just say, you know, every time somebody asked, and these were all pretty hip people, there was like, you know, a hundred people on here. And the people were asking questions, were asking very hard questions about, you know, what, if I'm using this and I've got this model and I got this turned off and, and I'm, you know, and I'm using Tea Mark or whatever, right?

And, and and every time they'd ask the guy, say. Just ask Cloud flow. I had that problem too. Absolutely. I ask Cloud Flow. Absolutely. 

Doug Finke: And that's why I also find it interesting, like, I don't understand how you can teach people this. There's, we're at a different level of, you know, back in the day I remember somebody saying, you know.

Give up math and give up computer science. This is all about rhetoric. And I pulled out my quadrivium in my trivium book and I'm like, oh, well, what the heck was Socrates and Plato talking about? And it's [00:30:00] like, we're eliciting responses. We're having a conversation with something. And it's like, you know, that's so different than like, I, I talk to people and they look at me like I'm telling them to talk to their toaster.

And it's like, no, this is, yeah, 

John Willis: yeah. Yeah. 

Doug Finke: I, I have conversations, I brainstorm. I ask it to implement. I look at some stuff or I, I just try it, copy paste errors back in, and then I go, well, how would you do this? Can I do that? Can I do the, what would you do? I ask it. What would you do? Yeah, 

John Willis: yeah. No, no. Yeah, yeah, 

Doug Finke: yeah.

It's crazy. But that's like, that's not how Wes. It's a whole different world, so yeah. Yeah. We're not 

John Willis: used to doing that with computers. Right? Like that's not the way we've always been. Tell the computer what to do. Now we're, we're, 

Doug Finke: we're I hate to say it, but I, I start thinking we're syntax monkeys.

Oh, you gotta have a function. You gotta set some parameters, you gotta check a return code. You gotta take the value from the function and put it over here. No, that's not the way it is, isn. That what we 

John Willis: would tried to get to in the first place. Really? I wasn't that the point of a compiler, you know. 

Doug Finke: Don't, don't start me with that.

'cause I was there when the first C compiler [00:31:00] came out and I watched all the mainframe assembly guys laugh at it 'cause it couldn't, you know, generate code correctly. Yeah. Or like, oh look, it, it's so inefficient. Totally. And then I would get the new compiler, like from I, IBM m like two weeks later. I'd go, Hey, look at this.

We disassemble it, it fixed that. Look at now it's doing it. Like what are you guys gonna do now? Yeah. It's crazy. I don't understand it. So I'm just the old guy yelling at people. 

John Willis: Do you think, I mean, you just mentioned something really quick, but I, I, I struggle on this and there, there's a lot of things about creativity.

I think there's a couple of threads here, but, we'll, at MCP and, and context protocols or whatever we want call, but the I guess, you know, one of the things you said, I, I, I sort of wonder, when people ask me what should my. You know, you probably get this a lot. I'm given a presentation. A parent will come up to me.

My son's in high school, my daughter's in high school. What should they focus on? I still sort of fall the default, like, you know, all things being equal. Get your math, but, but I don't know. You know, 

Doug Finke: I totally agree with the math [00:32:00] engineering software programming. Because it gives you a different understanding and it, it helps you critically think, it helps you set right, okay.

It can help you start thinking about how to build these things, but that's not how you interact with the machine anymore. But that kind of, that kind of mindset is, is is funda. You know, you can't, if you gave you know, what was the old thing? If you gave an AI to a junior coder, you know, you probably get a bunch of slop and it take longer.

If you give it to a mid-level coder, it might be a little better. But if you gave it to a season coder, and I'm not trying to. You say, season coders are great, but if you get give AI to a season coder, they might come up with a masterpiece in one third at a time. And that to me is the, is the Goldilocks tie in.

Right. And I, it's unfortunate that's the way I, I Where does that put 

John Willis: our industry then? Like what exactly, what, what do you think? I mean, what does that mean? Is there sort of a golden path where like it isn't as ugly as the senior people or the men, 

Doug Finke: you know? You know, you know the name Steve yy, right?

John Willis: Yeah. Yeah. Oh yeah. 

Doug Finke: And he came out [00:33:00] with one of his great posts early on this year, and he said The death of the junior coder. Yeah, yeah, yeah. 

John Willis: Yeah. 

Doug Finke: I read that and I was like, that's absolutely correct. And and he makes the point that in, even in the, in the legal space, they can have AI grip through briefs and whatnot.

They don't need paralegals. They're slower, they get things more wrong. You always need three people, seniors to go through the stuff. And I started saying to people, Hey, this. This makes sense to me. We're going to, and like, well, if we don't have any junior coders, what are we gonna do? I'm like, that's beyond my pay grade.

I don't know. But I don't know. Maybe they, they have to learn. AI 

John Willis: is, you know, one of the things I was gonna say earlier, the idea that we sort of like, I don't know, it definitely wasn't by design. I don't think that it actually, the things like functions and ultimately now MCP. Those actually sort of force you into good architectural decisions.

And I don't know that somebody sat back, you know, eight years ago and said, Hey, and let's do this and it'll make everybody a good programmer. But the way inferences become is that there [00:34:00]are a path where maybe it some point, the language translation just gets turned into good architectural code and we don't have to worry about senior.

I mean, I don't know. I mean, I think. I 

Doug Finke: don't think we, nobody here has a crystal ball. And I think, you know, we're, we're on we're, I was watching my Microsoft talk about their copilot stuff the other day and they were like asking questions. They go, well, what about this with copilot? Like a similar process.

And they basically said we're just learning workflow. We don't know how this is gonna, and that's the people who are building it. They're saying that, and I'm like, I totally agree, but will it get translated into it? Yeah. Do you need to still guide it? Absolutely. I think I took myself off the train of thought of what you asked, but 

John Willis: Well, as a senior, like is there a world where like we just collapse into senior programmers and what happens to mentorship and apprenticeships and 

Doug Finke: I think, but it's, I.

I don't really know, but I think it's, I think the, I think the mid and the, I think the junior, mid and senior people who are dragging their feet are [00:35:00] gonna really miss the boat because you're not gonna get, okay, 

John Willis: now that's like, no doubt version 

Doug Finke: six is version six. You're not gonna drop in and go, oh, I just hit a button.

No, this is the 

John Willis: O of any technology. If you're not sort of constantly adopting a new technology, you, you. You and even more now than ever, you're gonna make yourself obsolete. There's no question about that. So, 

Doug Finke: but as far as where is it gonna be to senior, I've been chatting with AI and I said, you know, it seems to me like LLMs were built by elite software developers.

Which I think we all agree, right? It was built by Google DeepMind Andre Carpathy and open AI folks. It was built by elite folks. For elite folks, right? They wrote, right codec. If you look at o open AI's Codex and include code, they run on Linux. They don't run on Windows yet. They're gonna, they're all porting to Windows, right?

But they were written to help solve their problem. And these are elite thinkers. These are guys that know how to ask questions of ai, to, to do benchmarks. I haven't, that's like, wow. Yeah. So it was written by them for them. When I look at [00:36:00] this stuff, I'm like, oh, no wonder why it works with design patterns.

If you use design patterns. Well, I, you're off to the races. I'm 

John Willis: like you. I mean, I'm just sort of guessing, but I think the biggest contribution of those people is they understood math and they coded like mathematicians, and I think that's what we're seeing is the fallout. 

Doug Finke: Good point. 

John Willis: Good point. Yeah. I mean, again, I, I don't know for certain, but I mean, I, I do, what I do know is I just wrote a book about history of AI where I literally documented like LeCun Hinton all the way back to the original neural network guys, you know, that were, they were all, you know, pretty heavy, you know, mathematicians.

Doug Finke: So, and that's a good point, but I think also at the same time, we're also changing how we're, you know, it's, it's not that anymore. And, and will it collapse into this whole process? But was it an accident that we got the function calling, I mean cps, right? Model context. Protocol by Anthropic they drew heavily from Visual Studio Codes, LSP language service provider.

John Willis: Oh, really? Okay. Alright. Let's, let's, let's do the deep dive. So there's, I mean, a lot of people are like calling this a [00:37:00] GenX, right? But, you know, and again, I don't want to get into sort of like, well, gensy is an M cv, but like, but we could start with there, like we learned how to use Copilots and Cursor and these tools to help us.

And then we had sort of the vibe coding, like you said, which was not just say, build me a big old program, you know, add sort of the tests, be iterative on the processing, right? Like, and then to your point, I think the better outcomes were very function based. And then. You know, and then people were talking about like chaining tools and stuff like that.

But, but I think Agentics explodes with MCP. Is that a fair assessment? 

Doug Finke: I'm, I, I, I definitely see how MCP came because of function calling, and I have my guesses why function calling started, but, now there's some people are saying, well, you know what? We just need basic Unix, CLI capabilities. I heard 

John Willis: that.

I've heard that. Yeah. Yeah. And I'm like 

Doug Finke: that. And I'm like, Ooh. And I, and then my work with CPS and function calling, I'm like, there's a lot, you know, get into [00:38:00] mcps. It's fantastic stuff that you can do with it. But there is like this abstraction that gets in the way, and it's sometimes this and sometimes that.

And you have to be. Have to do some things that are different and weird. So now I'm gonna go down and look at the pure CLI, like, what does that mean? How do you plug this into, you know 'cause 

John Willis: if those are just the functions right then and we know how to get, I mean, so give, can you give us an overview?

I, you know, I know I speak very well, but like for people who sort of listen, like, Hey, I've never really heard it from a programmer. Can you give us sort of an overview of what your explanation of what MCP is? 

Doug Finke: So model, context, protocol, the way it's basically defined is it's, you can write a piece of code, right?

And they've got SDKs for all languages, whether it's TypeScript go, so on and so forth. And what you do is you have a basic wrapper around it, and it goes back to like almost interface definition language back in a day. Similar concept saying, okay. You can call, here's a list of the functions that I wrote.

Here's the signatures of all of them, parameters and whatnot. And now I have a [00:39:00] USB that I can plug into any kind of machine. Doesn't matter if it's a Mac, a Linux box, or a Windows box, whatever it is. And now the MCP is pluggable and it's a server, and then people who write VS code and cursor and whatnot.

Those are MCP clients. So it knows that when this USB is plugged in, it says, what do you got? I, here's all my functions. And now that becomes part of the context when you prompt in these environments, right? And those things become part of the prompt. And then AI goes, oh, I think these, this, and this should be called and tells the client, oh, tell it to call this.

And then your code gets said, Hey, you need to call this with these parameters, and off it goes and returns a response. And that becomes this weird soup that we play in. I dunno if that that's a, a reasonable Yeah, no, that's good. I mean, 

John Willis: I think, you know, my Langchain, you know, they, they, they started us with the tools, right?

That was the whole point of the tools. Like, you know, hey, I wanted to, I want to do a math function. I want to do this and I wanna do this, and I want to check the weather. Bang. Right? So, and exactly. And MCP is sort of our client [00:40:00] server. Abstraction for that. So it's What about like these other protocols that are, and again, I'm, I'm just, not to put you on the spot, but like, one of the things like, you know, for like, again from Reuven and just a little bit of experience, the MCP seems to be sort of client server or horizontal and some of these other sort of more mesh based, like the Google, like at least Google's A to A and all that stuff.

And in the end I would think, you know, not not having built. I do wanna do my first swarm. I'm like, okay, come on. Like, I've been problem, I've been talking about I gotta do a Reuven Cohen Swarm, you know, I'm just looking for the right application. And then I'm just gonna sort of like dive in right here sometime this summer hopefully the next couple weeks.

But because I know that's this whole conversation on steroids, but but, but what about like these, like in instinctively, again, going back to sort of our experience that. A mesh based version of this stuff might work better. Or maybe even, like you said, maybe, maybe it's a mesh based version of, of, you know.

Command line. 

Doug Finke: Absolutely. And that's, you know, that's, I've, I think MCP [00:41:00] came out, I, when it first came out in last November, I was like, eh, it looks like function calling to me. What magic sauce did they give? And then in a couple of months it's like, oh, this is, this is magic. And now the A to A stuff, I think it's good, but I'm, for me, so don't.

Don't take me on this. I'm, I'm like, this stuff is moving so quickly. Yeah. , I don't have a favorite LLMI don't have a favorite. IDEI don't have which one's doing what. And guess what? Next week they're all gonna pull from each other. Is it a mesh? Is it a swarm? I love the way Cohen talks about it.

He's like the swarm versus a hive mind. I'm like, oh, these are the next level. That's pretty freaky. Yeah. 

John Willis: Yeah, yeah. 

Doug Finke: And then he's, yeah. And the way he's doing it. It's, I know that inside the big houses, right, anthropic and open ai, they're applying this, right? 'cause they, they're seeing these cones on their radar.

The rest of the folks are doing this in Google, Google, CLI. Unbelievable what you can do with that thing. Yeah. And it's not just for coding. So this stuff is evolving, but, so it's, you know, you gotta stay on top of, you [00:42:00] gotta stay on top of it. You, you gotta have it brush over your mind or 

John Willis: wash. So, you know, I, I made that post 'cause I, I'm getting tired of people like that whole thing.

Like, you know, like, like either try it. You're not a coder and complain about it, or you are a coder, and try it and learn. Right? Like stop, like criticizing what I mean, real code's gonna, like, there's gonna be a path for him to make a gazillion dollars. And he knows that and, and he's, you know, he's always been an entrepreneur.

But right now, like I said right now, like take advantage of what he's doing and stop complaining. But there was some naysayers and I know like through sort of some like, I don't like no, but I know, but I don't know. Like there are really important people grabbing some of his calendar time right now and spending those hours with him and paying, I was gonna mention 

Doug Finke: to you like, you know the name Adrian Cockcroft.

Oh, I know Adrian 

John Willis: really well. Yeah. 

Doug Finke: Oh yeah. But he just posted Netflix. He's, I didn't realize that he goes back as far as Cohen does back in the cloud [00:43:00] before the clouds started. So 

John Willis: Adrian, just give you a back, sorry, Adrian. Yeah, please. He he basically, I've known Adrian for, you know, for, for forever. And we, you know, we've done some, we've done a bunch of stuff together, but not nothing significant.

But he, he was at Netflix, so he goes back to Sun, like he was one of the early Sun gurus, right? But then he was early on in architecture and Netflix, and he is the guy that took Netflix to the cloud. 

Doug Finke: Yeah. 

John Willis: Yeah. I mean, so, you know, and then went over to Amazon and became like, you know, the head, head of cloud and then he ended his career as a sustainability guy.

Yeah, no, what he just built, it was a great example. That's the kind of how I like, okay, John, shut up. Adrian's doing it like your friends are doing it, you know you know, guys that come to me for advice, they're doing it like. Just let's do one. Really. Let's get one out like Adrian might. So yeah, no, he did a whole home management system Exactly 

Doug Finke: this morning.

And, and to the point about like, are these meshes, are they gonna be whatever? I don't know. They might be swarms and hives. Yeah. Who knows? I mean, I watching, I was watching open ai, one of the key developers there, we start, he built a little swarm product process before. Cohen was doing his stuff [00:44:00] not to.

Okay. Yeah. But he, they were so, they're all, it's all in the pot and they're all thinking about it. I don't know. I, again, I sit back and I look, I go, where are we gonna be in six, in six weeks? It might, yeah. No, I 

John Willis: mean, I think you're right. Six weeks is the marker. You know, it's not even a month. It's not quite a month.

And it's not like three months. It's not a quarter. So the other thing I wanted to talk to you about, I think is a little driving me nuts is this windsurf thing. I, you know, I think that this to me. I don't know that this, some people turn this into a positive story and I, I, I don't, I think it's, in one way it's positive in that it just tells you there is no moat.

Right? Like, it, it glares with neon signs. There is no moat. There is no moat, but it's sort of a negative for investors and like, I mean like, we've been doing this a long time, right? Like the Windsurf company was gonna sell the OpenAI. There was some IP conflicts then all of a sudden, and because it's open source, right?

The this, this is what happens. I've [00:45:00] seen this before with other projects, you know, even after the sale where the whole team leaves. 

Doug Finke: Mm-hmm. 

John Willis: After the sale, because they only, like the three or four founders get all the money and everybody else gets table stakes and, and then. The found, the, the open source team that runs it says, Hey, why don't we just support the project?

So like, this sounds like what happened here, where, you know, the, the deal went south. The, the, the, the founders and some of the architecture team went over to Google, got crazy money,

I think, you know, and then, you know, now the Devon guys, which I, I old Devon story too is like, I don't wanna waste too much time.

A week after Devin came out, there were like five better versions of it that were open source. So the Devon Folk bought what was left of Windsurf, which is just an open source project and Devon's proprietary. I, I don't think this is a good story for anybody other than there is no moat. 

Doug Finke: I, I a hundred percent agree as far as a good or bad story.

I mean, this reminds me of back in the day, even when Andrews Heidelberg was poached [00:46:00]from Borland after he wrote Delphi to come over to Microsoft. Now you go way back, 

John Willis: but yeah, yeah, yeah, yeah, 

Doug Finke: yeah. Let's go back, man. You know, nostalgia, if we have enough of, it'll kill us, but, so I don't, but there's been a lot.

This same, this is the same drama that's played out. It's just at a faster and a larger scale. In fact, I think what Cursor just hired the guy who built Claude Code for away from it. Oh, really? Okay. Oh wow. So this is, this is gonna be, this is gonna be the play, right? And then you know, scale ai, I mean, I knew some folks I was talking to folks on by Pure, by accident on, on X back in 22.

They, they came out of things like they worked at like, not at Cupid, right? They were data scientists at Cupid so they can do things and match people or whatever. Okay. And they were the first, first folks that were jumping on GPT stuff going, Hey, you can do this and try that. And I'm like, wow. And then they actually talked to me and I would ask them the, the dumbest questions.

Uhhuh like, well, this, and then they got picked up by scale, AI and Wang, who, who headed that up was at the, you know, at the White House talking about the next [00:47:00] level of ai. And for Billions, he got picked up by Zuckerberg. Zuckerberg and he is now part of that big. So this, this whole thing, there's trillions of dollars on, on the line here.

Right. And you know, again, we like, to your point, I just wanna like, 'cause I just like talking about AI deniers, but to your point about, I can tell immediately that you just want to complain about this new stuff. Yeah, yeah, yeah. Oh, it's, we need governance, we need responsible, we need ethics, we need, it doesn't work for this.

I can't count the number of RS in strawberry. I'm like. Oh, I, I immediately 

John Willis: do math probably like, like, I mean, you were talking about, you had a background in, in sort of a neural network. I mean, part of my research to write the book, I, I went way deeper than what I understood into, I had to explain, you know, to my mother-in-law, you know, that my, my target is I either have a expert and my mother-in-law.

Sweet. Well, they both enjoyed the book and I had to explain sort of the, the Lacoon and mis en mis I've heard of that. Yeah. So the handwriting exercise [00:48:00] in to my mother-in-law and it worked, right. So, but but yeah, no.

Doug Finke: So you can pick up on people immediately who just don't, you know have never, they don't try it or they try it.

I typed in a question and it didn't gimme the answer, so I, I tossed it out. Gimme the wrong 

John Willis: answer. Or like, yeah, no, no. Yeah. No, it's, yeah. Like, again, like going back to my math thing, like it does do math. Well, it, because it's like, it actually does do math. It actually does geometry. Exactly. 

Doug Finke: Yeah. So, and the fun, fun part that ties in the function calling, what's some of the fun things?

I have fun demoing is I'll go up chat GPT and I'll say multiply the number. 8 7 6, 7 0.1 5 7 6, 5 times 8, 2 3. Do it'll. And what it'll do is it'll try to do it in its head. And it doesn't do it right, but if I give it a function that does math, that has adding subtraction work, sure, yeah, yeah, of course.

Boom, I'm done. To me that solidifies it has to use tools and that's the, that's the way to go forward. Tone has proven all that stuff. That's good. It's a fascinating, I wish there were more [00:49:00]people like you, you know, that we can sit down and just, what are we gonna build, right? Like, just, just jump into the swarm and hive mine.

Doesn't matter what the app is. Build, do I think the hackathon is 

John Willis: key? I think we, we need to get, I, you know, I, I've ran one for for companies. I, I really want to start like, you know, so getting. Like really just at most conferences, try to run a hackathon, because I think what I've seen people do in hackathons is just mm-hmm.

You know, you, you, you start at like 9:00 AM and by 7:00 PM you're judging 'em. And they're, and some of the weekend ones, but the ones I work on are usually systems oriented and you know, sort of DevOps or whatever. And what people produce in a day is just, that's pre hive, pre sort of swarm hive stuff, you know?

So. Good point. Yeah. So I think hands on, 

Doug Finke: hands on experience. I haven't run hackathons. But yeah, I don't, I'm not, I'm still in the camp. Like, I don't know how, you know, to teach this. I mean, I know how to teach it. I mean, I had to talk about it. And there are a lot of people that come back to my channel, like I'm doing another one on [00:50:00] Friday.

Yeah. Yeah. People want to come and listen and they talk, Hey, I got this result and that result. But doing a hackathon, I don't, I don't 

John Willis: know. Hackathons are beautiful. No, they, they literally, because it, it's sort of, you get people that are sort of like, and then you, anyway, they work in groups and sort of like, it's, it's, I've loved hackathons for years because you see this group mentality where you.

People in, I've actually, where I've managed, like I'll go around to different groups and say, Hey Joe, I need you to come over to this group 'cause you have expertise. I haven't done that with the AI stuff, but I used to do it with infrastructure as code hackathons where I'd somebody'd have a particular expertise in something and I knew this team was lacking, so I'll swap people out.

Right. Nice. But, but just the idea that you have five or six people on a team and everybody has sort of, you know, different. Angles and inputs, and you just watch it because you can, the, the feedback loops on this stuff are just like you, you know, you've pointed out throughout this whole podcast are just incredible.

Well, I had fun. I don't know about you. 

Doug Finke: Absolutely ab this is fantastic. [00:51:00] Absolutely. Yeah, I would love to do more and take more of your time at some point. Yeah, this is a, it's, it's a whole new world. We have no idea where we're going. And it's really good to talk to somebody who's like, you know what?

I don't know. I know it can't count ours, but I bet you there's something else it can do that's can be, yeah, we do a 

John Willis: whole podcast. It's a nonsense of people. So, you know, I like, I think Ethan Mollick wrote a great book. I mean, he's sort of my competition now, but, but like. You know, like how much energy is being wasted by people.

Every time there's a new model where they're going out and running all these things, oh, it can't answer these questions like that. You're missing the point here, like testing how many r it counts every time a new model could. It's just wasting energy. 

Doug Finke: Exactly. You could be, instead of spending your time, you could be investing and learning how to leverage this thing.

John Willis: That's right. Yeah, yeah, yeah. Like back when 

Doug Finke: PowerShell came out, people would go, well, yeah, would you write a trading system in it? Absolutely not. But you know what I would do? I would take the trading system I wrote in, in c plus plus, and I would automate it so I can make sure things [00:52:00] worked and I was monitoring and I could deploy it like instantaneously.

And that's why I love when DevOps came out. Yeah, we were kind of doing that. We, you, a lot of people get to that point and then they go, oh look, somebody codified and articulated it. Alright, let's do those steps. And then all of a sudden your systems are, are running much, so much better. Yeah. I wouldn't write, but don't write c plus plus to, to do all that stuff.

Everything's tuned for something differently. Right. Counting. Ours are not counting, ours is not an indication of much. 

John Willis: It is and you? Or can you do math properly or, I mean, even benchmarks I think to a certain extent, like are, are, are somewhat waste of energy. But I agree. Alright, well we will do this again for sure.

Because I think Awesome. And I maybe you can come on mind and Well, I would love to anytime. I mean, you know, I, I, one of the things I have to do is start getting on a podcast to sort of promote my AI book. 'cause that's a, you know, I want to get that. And I think, and you, you say you go from. And you go from Aristotle to ai?

Well yeah. You know, one of the, you know, as I was using AI to critique it, you know, so I, I didn't use AI to write it. In fact, that's a dilemma I'm [00:53:00] having right now. Okay. You 

Doug Finke: don't have to apologize. It's okay if you did. 

John Willis: I know. No, I, that's the point. I think I'm hardly, I mean, I've got enough feedback from people, like, it's okay to write a book with ai, but not to date.

I haven't, but I did use it. For analysis. So it was amazing how I was able to throw in, and one of the first complaints was you really didn't spend enough time on aristo. So the truth of the matter is there's two arcs. You know, one is neural network, one is expert systems. Neural network starts in 19, what, 43 with McClellan Pitts, who create the first mathematical paper.

But I do go back to nice. You know, I, I start with like bull and babbage 'cause you gotta understand Bull and logic and then Ada Lovelace, you know, absolutely first programing and, and then Turing. And then you got McCull Pitts, and then beautiful, then all the way through. And then a lot of unlikely characters like Grace Hopper.

Like I've exactly 25 books on the history of Eye and not one of 'em mentions Bert. Well, she's the first one to put a natural language interface on top bits and bytes. Right, exactly. Like we [00:54:00]don't, well, we probably don't get really as fast to, you know you know, NLP. Without having and, and, and 

Doug Finke: what she did, what the way they had a program back in the day, so to speak, WW you know, they didn't even have subroutines and functions and they had to do things and rear rearrange boards and whatnot.

John Willis: Yeah. 

Doug Finke: I even go back to Leni 'cause Leni created the bullying system for us. Yeah. We got 

John Willis: a little bit of Leni in there and you know, like, again, you can go sort of nuts horizontally, but we carry a pretty good thread. And then, you know, then you get into sort of the modern AI of these like, you know, LeCun and Hinton and you know, Alex net.

Mm-hmm. And work our way up to, you know, the sort of what went on. So it really isn't, it's about how did you get to ChatGPT so, yep. Alright, my friend. Thank you. How do people get to you? Is they're looking for you? 

Doug Finke: You can find me on LinkedIn, Doug Fink, D-O-U-G-F-I-N-K-E. And I'm on all the other social platforms.

Yeah, 

John Willis: no. So you're a great person to follow honestly. 'cause I, and again, then when I found out we had similar background, so cool. Well good. We will do this again. We'll get that all up in the show notes and, and all the good [00:55:00] stuff, man. Thanks Doug. 

Doug Finke: Thank you. Talk to you later. Have a good one .