Your Business Is Not Ready for AI And It's Costing You

I think people perhaps made the conclusion that all problems uh in software development become abstracted away and solved by AI which is not true but

you must have seen some radical changes over the last few years things have really shifted in the software world what have you witnessed Joe

yeah you can whether you see this in case studies or even you know conversations I've had with friends that are at you know so fortune 5 type companies they all have big initiatives going on they're going top down not bottom up they're talking about AI readiness they're talking about business process mapping.

They're talking about AI governance and they're talking about how this is going to lead to not just the augmentation of workflows but the total transformation of sort of re-imagining of what even workflows exist and how they look. The biggest players are doing this AI is it hype or help? Good day and welcome to the show. I'm your host Pete Cooper and our topic is AI hype or help. Today's guest is Joe Lewinsky from Nuvo Technologies. Joe is a longtime software developer. He came out of Silicon Valley, spent time over there and has seen the change from pre-AI to where we are now and using AI, particularly how it relates to unfolding the potential of AI in businesses. So, let's get into it.

Good day, Joe. How you doing? Hey, doing great. How are you?

Yeah, good. Welcome to our show, our podcast. So, Joe Lewinsky from Nuvo Code Technologies. Joe, tell us a little bit about your business to start things off.

Yeah, absolutely. It's um it's uh it's an interesting question because it's been a a bit of a moving target over the time since I established Novo Code originally. Uh you know, so my background I've spent um about 12 years a little bit more in software engineering.

Uh and I used to work in New York and San Francisco uh you know mostly in the startup ecosystems there and uh you know back uh in um see about 2025 or so I was mainly I had gone independent I had founded new book code technologies uh with the intention of having something of you know just a more you know garden variety uh agency.

Um the thing was of course is that that was not destined to be because the advent of AI and really the advent of AI coming into um you know a broader set of companies and simply being the domain of technology that that was co-occurring as I was uh in the midst of my agency work and so what this really led to was a situation where that um even in the middle of last year you know there was it was kind of more business as usual for software development projects uh and AI was sort of being murmured about. There was different sort of applications for it that were finding their way in small ways into projects.

And then I would say about Q4 of last year and certainly at the beginning of you know the current year uh it just exploded and all AI was that anybody ever wanted to talk about AI was the central piece of every project that you know that I was you know discussing with potential clients and and and so the the focus really became um uh to to focus around sort of like enabling these AI implementations. Um, and I I guess maybe you know sort of foreshadowing a bit. I think a lot of my current work um came has come out of sort of trying to help companies sort of understand what good AI looks like. How do you actually make it work for you?

Because I think that you know everyone by this point has heard sort of AI horror stories or just um sort of uninspiring outcomes that uh that fail to deliver. And um and I had noticed these myself and I realized that there's a sort of um not just an an implementation sort of challenge that needs to occur but there's also an educational uh sort of challenge and organizational challenge that pertains to AI that a lot of my work focuses around now.

Okay. Interesting. So enabling AI implementations I think was the key thing that you said in there.

Now, if I was to break that down a little bit, um there's a need to write some software or to use AI to to write some software which brings about um some change to a business or uh some some I mean you focus mainly on businesses, right? Helping businesses transform with AI. Okay. Very close to my heart that idea uh and that enablement. So take us through what if you have any recent example just take us through without naming names what that really looks like so we can sort of ground that down for our audience.

Yeah exactly. So the my world typically historically until roughly this year was really focused around implementation work.

However, AI is a little bit different because the the patterns and the system itself, it's it's very fundamentally different from a lot of the software that that had existed before. A lot of software is it's ones and zeros. It's discreet. It's deterministic. AI is inherently probabilistic. Um, you know, though it's not quite that simple, but it's it's easy to think about these things in these uh in these two ways. And that implies that there's a whole bunch of different paradigms and design patterns uh that that need to uh that need to exist for it. And of course it depends on consuming a great amount of data.

And so what what happens here is that you know because it's a probabilistic system we have the the infamous hallucinations. Um, now hallucinations while they are inherent to AI systems, there are ways to mitigate them and there are ways to catch them when they're doing so in ways that that allow you to uh do so quickly enough to actually have, you know, return on investment. But um, in order to even get to that point, there's also something that we talk about is like AI readiness. And that's a bit of a domain unto its own. It pertains to things like, uh, data hygiene. You know, how how clean is the data with, you know, um within your your sort of your data ecosystem within your organization.

There's sort of like data unification. How well is all this data able to be brought together uh through one uh sort of layer and understood mutually across all those layers. Uh and then um there's also a a really important step which I see I I I think is a little bit of the frontier right now which is the business process mapping uh that has to be done um around any given automation

and because that is going to distill down into what we call the AI governance layer um because it's one thing for AI to do things but it's equally as important I'd probably say even more important about how that AI you agent or pipeline does what it does.

Um you know who the things like um what access does it have? When should it uh when should it take a particular action? When should it uh you know check back and do something else you know it's like all these things are you know being able to determine what those things are are different from case to case and they all distill down from that business process mapping you know.

So

if you had an engagement with a client u you would do some element of discovery first to see where perhaps AI could give them some advantage right because I think we were alluding to something which has become clear to me is that over the last few years we've seen a lot of uh I guess promise from AI like it it seems to be able to do amazing things Um, and then I also see um, everybody being an expert. And what they're being an expert at is going to ChatGPT, getting a kangaroo's head on a rabbit's body, and now that was the morning. In the afternoon, they're down the pub being an expert, right?

Not particularly useful, fun, shows capability, but when it comes to uh going into businesses, you have to affect their revenues, their um costs or risks or some kind of future mitigation, right? So, um it needs to be analyzed really what the business is doing first, right? So, is that part of what you do?

Yes, absolutely. I I would argue that, you know, this is at least from my from where I stand as, you know, someone that's been in technology for a very long time. I would argue that the that that sort of that AI readiness step, some of that preparation, that preparatory work that goes into the laying the groundwork for uh you know, building some particular automation.

I would argue that's more important even than the the implementation step. While that's important, I would say that if you get the the preparatory steps right, all the AI readiness stuff correct, the system design, maybe not go so far as to say it's trivial, but it's it's at this point um there is an understanding around that and how to achieve it. And it's fairly um from someone in my position, it's fairly straightforward to distill that into a system which can, you know, perform as uh you know, as the client would like it to perform.

So if we were to take maybe a recent example of a particular company you've worked with, how long does it take to get that preparatory work done and how does that what does that actually look like? It you know it's it really varies case to case because the the thing that I'm I'm really finding and that um that I found very challenging about the space right now is that different companies you know at different scales at uh different uh levels of sort of a you know um uh their appetite for you know technology and um are really at different places in the conversation. Um there's uh one I a prospective client that I'm speaking to right now uh in the cyber security space.

They already are well down the road. You know they they they can almost start at a place where you're talking almost at a system design level. On the other hand um another client that I happen to be working with uh they have much of their uh their business very well defined. A lot of their process is exquisitly defined. However, the actual um the mapping of how that then will translate into a system is a little bit less defined and that work still needs to be per performed.

So, um, then if you move further back, I've I've had conversations with, um, some, uh, colleagues in sort of mid-market finance where the firms they're part of are much more conservative and there is just more, you know, they're still in the stage where we bought a bunch of co-pilot seats, we gave everybody co-pilot and they're using it in a very bottom-up fashion to enable their own sort of uh, or augment their own daily workflows. But there's not a broader vision. There's they sort of like they're not even really clear where to start. Um and so they're even pre um that AI readiness conversation.

It's more or less just almost like educating about like sort of the problem space that they're going to face next.

Um so the the conversation really really differs uh widely. Um and I think that'll collapse down in the next uh coming years as everybody the information and the understanding starts to propagate more widely. But right now it's it's it's a very heterogeneous sort of environment in that respect.

So um I guess there's there's a sort of fundamentally assumed level of knowledge, right? Um

I know some people in business who have have not even turned on an LLM, right? They they have no interest really in it. They don't really use a computer much.

Their business is done perhaps through you know meetings through um their presence um driving to various sites and networking and it's very much people focused they know AI is out there obviously but they really don't know the leverage that it can give and I would say that's a I sort of go from zero to five I would say it's a level zero someone who hasn't turned on an LLM is at is is at a level zero their awareness is probably so low that they don't even realize that there's a potential to do anything with their business. But once you start going up the levels and people start to say, "Oh, well, I'm using an LLM now and it's helping me do my reports. It's helping me do my research.

It's giving me time advantage. Oh, well, if it's helping me, maybe I should roll it out across the organization. " So, you know, next thing you know, there's everybody else using the LLM, I think you said, from the bottom up. And they're all using it. I would sort that that level one level, but there's nothing p pulling anything together to really leverage it. And we know that AI has the capacity to do that. Would you agree with my sort of level zero and level one there?

Yeah, absolutely. Um I I think that's very very well characterized.

Um, and I would kind of highlight there there's something maybe kind of hidden in there and that's sort of that that uh that levelwise description is what a challenging information environment I think this is for people to make sense because their their what they perceive can be done with AI. It really maps to the level they're at. But the they're what they believe can be done with AI, it doesn't move in linear fashion as their understanding grows. They might be at level one and be like like, "Oh my god, this it's going to change everything. " But then I find that people at like level two and three get to this point where it's like this is all smoke and mirrors. There's nothing here.

This is, you know, I can't use this for anything. It's like this, you know, this just, you know, uh it it made factual errors in a report I tried to generate with it even after I gave it a perfectly clean data sheet. It's like this will never work. But you know, so it's the complete opposite opinion at that level three. But then as you move beyond that, then you start to see that you're like, wait, no, no, no. It's there's a there there. And there's ways to actually to solve those problems that were encountered at sort of that level three understanding.

And anyway, it's so it's it's very interesting, but I I have a lot of sympathy for people uh that are sort of ascending that curve and trying to make sense of it because it's not widespread in terms of the like the the truth of it all. And why it it is the truth. Um, but nevertheless, yeah, it is the case. I think what the way you describe it is accurate.

So, I like where we're going with this and I think there's a fundamental piece that we all need to understand about what an LLM actually is. It's essentially a massive patent recognition machine, right? It's it's just taken enormous amounts of data and it's looking for patents. Um, and as you said, it's probability.

So, and it's it's finishing the end of a sentence. Say the cat sat on the you know mat or the cat sat on the dog or and it will say most likely it's the cat and then it does the next word and the next word to if you kind of synthesize all that up to me we're it's kind of pushing everything to a kind of mean point like a distributed mean point where everything is migrating towards a common point but nobody is Mr. No business is perfectly average.

Yes.

So, there's always going to be this misfit between, oh, well, I'm not a normal business. Nobody is. I'm somewhere down here. And if you happen to be right out on the extremes, you're going to be a long way from the average.

And the you're going to get a worse result than if you're closer to the middle. Um, do you think that makes sense?

Yeah. Yeah. And and you know, this is actually I'm I'm glad you touched on this because I think this is sort of one of when I think about some of the the deeper philosophical um points that are that get raised with sort of the advent of AI. I think it's this exactly what you pointed to with sort of this that you that AI it's a it's an average path finder that it finds its way through nodes of probability. I mean maybe perhaps not strictly the average but the average within the context of where it is.

And it's interesting because I think that yes, it does inherently then push us towards averages. And I think um that there are one of the challenges, you know, we we think about what is AI not good for. And one of the things that I think that it is is specifically not is making a direct prediction about what lays over the horizon. And there's there's I think it was a Harvard study that came out that kind of talked about some of the the uh um perhaps it's an MIT study. Either way, they talked about sort of the the strategic bias along certain axes that it gets pushed. But what I think this is expressing really is that um it's really pushing towards averages.

And in my estimation, I I think that that genius is fundamentally a tail phenomenon.

And that you know it's like that when you want to break that you know where you're going to find and where you know advantage is going to be found is going to be have to be in these areas where everybody else isn't looking right because that's that's an information environment that the the value has been extracted from and so like somewhere along some some axis uh the tail of that that around that distribution is going to be where the advantage is going to be found and so I I find myself thinking very a lot about like how

where where role where does the human come back in in this new era and I think it's sort of being able to like to exist outside of like sort of like you know what the AI

and kind of where it handles sort of the the the middle and and surfaces a lot of insights and can do so you know very wonderfully and expeditiously. So um yeah but no it's just an interesting point you raised it's something that I find myself thinking a lot about. Well, there's a I think there's an insight here in that AI can only work on what's currently there in the system. Let's call it the system or the internet. Um, what it can't do is look at it can't look down a microscope. It can't look down a telescope.

And it was those things that happened several hundred years ago with Galileo and others where they shaped the lens and they looked down a microscope and all of a sudden they realized there was a whole world that they didn't understand. You know, back before microscopes, people thought that this that the thinking was done in your heart, right? But, you know, the advent of the the microscope enabled us to understand cells and what cells do. And of course, we looked out through um through we knew the world was round, but we thought that the sun went around the earth. And it was through looking through the telescopes that we're able to realize that's the other way around.

And people think, well, so what? You know that was there's the relevance to that was very relevant to that because if you understand how the universe works you can understand how GPS works and therefore you can use your phone to get from A to B very very practical can't do it without general relativity and a deep understanding of the science that all goes behind that and when it comes to looking down the microscope so many things associated with your health come as a direct result of the fact that we understand how the body works at that level so these are not abstract pieces of science these have direct bearings on our lives.

Um, AI is not able, what I'm trying to say is AI is not able to necessarily go out and find new science in the world, right? It's always going to be whatever is already in the digital space.

Yeah. And and I think and this there's there's a bit of like nuance here because I think that I I I can already hear, you know, that there would be somebody out there saying it's like, whoa, whoa, whoa, hold on now.

It's like but there's all these incredible like these AI videos and there's also you know there's you we asked Opus uh you know to compose a joke and it was hilarious you know you're think like it's clear AI is doing creative work and and I think it it draws a bit of attent some attention to like where the difference is here in that I think that AI has demonstrated something like maybe I won't go so far as to call it sham creativity but creativity in a recombinant sort of form where the existing element elements, the elements that we know are there, it can recombine them in interesting and even perhaps uh novel ways, ways that have craft attached to them.

But the transformational, the revolutionary uh where fundamentally new ground is broken, you know, something akin to the level of like a new art movement uh or a new, you know, sort of a fundamentally new branch of science that you know that discovers or explores things that were fundamentally not understood to exist in the meaningful way beforehand.

That does not seem to be a capability of it and it is and now while I'm certain that somebody that there are probably teams of people you know uh very smart working on this exact problem it is not a foregone conclusion that this is possible with the you know what AI can do you know

yeah that's right I mean I think there is this generative component to it too isn't there where it it does have a sense of of creating some random randomization um that they've built into it so it's not directly going to the mean every single time otherwise it just wouldn't be able to function properly.

Um, and I think that does I mean I don't know the science that well maybe you do but it seems to me that you're going to trade off hallucinations with the the randomness which drives you away from the mean. So if you if you're if you if it's too much generation there's too much randomness you're going to start having more hallucinations

or you're going to have everything's going to migrate to the mean across everything and everyone's going to be doing exactly the same thing. And I guess philosophically we we become a hive mind like we're all thinking the same thing because that's all we know and it's just fed back to us again and again.

Yes.

Yeah.

That's um it's funny and that process I think was already kind of underway a bit even before the advent of AI I think with things like you know the global social media and everything.

Yeah. Yeah. Cycle after cycle of uh of mean average um sort of content that we've consumed and then created new content.

100%. I mean, you you and I both in the US, we've seen the the polarization of politics without taking sides. It's clear that there's two sides. And those we all like the side we're on.

Honestly, I try to stay in the middle.

It's hard to stay in the middle, but you know, it's it's it's very comforting to know that there's a clan of people that think the same way I do on this side and the same on this thing. It's tribal. It's inbuilt. Uh it's part of our monkey mind that takes us down that path of feeling comfort from hearing other people share our views and going and getting other views is uncomfortable so we don't tend to do that so it just drives us apart.

Yeah.

Yes. Yeah. It it's I mean that's very well sketched.

I think that that describes sort of the mechanics of sort of how we wound up here and why we find it such a difficult process to extricate ourselves uh from this because it's just it's operating at such a fundamental level that is tough to override.

It is, isn't it? And I think

you know as we go into AI and we probably come back to some of the practical implementations, but as we go into AI, I think it's not human. Someone said this to me the other day and I did a short segment on it. An AI is not a person. An AI is a tool.

Um, and uh, it's a very very powerful tool and maybe it's more than a tool or you can frame it other ways, but it's definitely not a human and I can think of some really good examples. You can't send an AI to jail. There's no accountability, right? You can't find them. You can't punish them, but you also, you know, you can't really reward them. So, it doesn't have that. Not out in the physical world here either, you know? And an instance in AI is just an instance. It it literally the word instance says it's there and it's gone. I mean it's it's it's hard to have any kind of uh accountability. I'm repeating that word, but any kind of accountability uh that a human has.

We have long-term accountability. We have social connections. We have trust. We have a career. We have, you know, all these things that we build up to. Um AI has none of those. You can't swap out the two, you know.

No, absolutely not. And it's it's interesting too that you know there's a big question you know we as we look at kind of the societal level of that everybody I think is has thought about and it's probably you know maybe lost a few hours of sleep over is thinking about like well what's where do humans fit in in this future that's you know that's kind of unfolding rapidly and in totally uh in disccernible fashion.

Uh and I actually think that sort of that that accountability uh and that the ability to be culpable will end up at in some way some shape or form end up being humans moat a little bit I think in all this

um that it's like that at the end of the day you do need someone who can be held accountable um you know against whatever happens and when you have a machine that is you know that's probabistic in nature and has is phenomenologically uh inert it's like you can't hold that accountable. Um, so yeah, so there's something in there.

Yeah,

that's I think that's the the issue I think we're having with with the self-driving cars is it's not that it can't be done and I believe in certain circumstance I had a podcast guest on who was very very close to this like he's been 30 years in camera development. He said out on the highway and a a car that's that's um that's auto driving you know with AI is safer than a human. That's statistically proven. He said, "If you could if you could replace every car with the best AI, we would drop thousands of deaths on the road every year. " Um, but you can't get around the legal issues, right? Who's accountable for driving the car? Is it the instance?

Is it the owner who was asleep or is it the person who who is it the engineer working for one of the companies? Or is it the company itself? I mean, it's now currently if I make an error in in the road and I pay the price for it, whether it's a fine or or worse, you know, I'm accountable. Um, the real world has accountabilities. AI seems to skip over all that.

Yeah.

And it's that example you gave is quite interesting in that um it it struck me how how having a human operator it it collapses all the you know all the accountability and culpability down to like a discrete understandable like entity in one point in that entire chain of like you know of that of the car's existence and the other people that you know say were injured. But in the case when the human is taken away from it, all that responsibilility diffuses sort of ambiguously that there's no obvious note in the chain where it can be applied anymore. Um, and so yeah, it's unclear what to do about that, but yeah, it's just interesting point. Yeah.

Well, let's go back to some of your stuff.

So, you you are you still based out in California?

Uh, you know, I'm in Minneapolis uh now. Yeah. Yeah. Yeah.

But you were in California. You said you were part of the uh the startup world for 12 years. Is that right?

Uh yeah. So I so my I had kind of I started out in uh California um in 2014, went to New York, I lived there for eight years uh and uh then was part of a then sort of like the the last sort of like W2 type work that I did was with a startup out of San Francisco. Um and

W2 that's for overseas people that's full-time staff employment.

Uh yes, exact. Yeah, precisely. Yeah. Where I was

took me about three years after I moved here before I figured that out.

Yes.

Yeah. I have to be careful with my uh my legal nomenclature that uh you know throw around but yes that exactly yeah when I was working for someone um that was uh that was out in San Francisco and it was after that that I went and started uh Nuvo code um and uh yeah so but that that's really it was all for the most part all that time was really in the startup space and it uh it was really kind of a blessing I think to uh to be in that world because it was just certainly in certainly during that time period and still is um there's such a premium on uh you know being moving moving quickly uh doing so with you know craft.

It's it's kind of like a nice uh marrying of of you know quick delivery but alongside it's on the rails of like high standards and uh you know sort of a technical um technical excellence I guess we can say. Um so that that set a really good foundation I think for coming into this world um of AI where you know that you know development you know now that it's so it can be done so quickly and um you can do so much with it that when it rests on sort of I think that that fundamental understanding of sort of you know the sort of like the the way of how to do software well and the sort of scope of concerns.

You are you are a trained software engineer software developer right?

That's right.

Yeah.

Right. Okay. So you must have seen some radical changes over the last few years, right? From from preAI, which is I guess before 2022, even though I've been in AI since 2016, but on more the deterministic vision system sides of things, but when the LLM started to come in in 2022, and now it's happened in the last couple of years, things have really shifted in the software world. What have you witnessed, Joe?

Oh gosh. Yeah. Where to begin? I guess uh well obviously so my work just starting with my my own workflow has changed um absolutely dramatically.

I mean I think from back in 2023 I was using uh sort of a there was a like a a nice tabbed autocomplete that was powered by like a GitHub co-pilot uh

inside the uh the the the IDE that I was using when that and that was like that that made so many things so much more tractable and was in and of itself a an incredible you know speed and productivity boost. But now looking back on that that's nothing when you compare it to um the world of you know say cloud

little thing if people aren't software developers um if if you and I did some software I never sort of expanded that side of things.

I went to other areas but what I think one of the things that makes software hard is you have to get the syntax right. In fact I went right back to the days when we were writing an assembler code and every single you know register change had to be described with exact syntax. If you made the slightest slightest error, spelling error or punctuation error, the whole thing wouldn't work. It'd be full of bugs. The autocomplete really helped that, right? So, it was a big step forward. It

it is huge. Um I mean I I think myself and every software engineer has um memories of many days spent where you're staring at a terminal screen trying and something is uh is broke.

The comp it will not compile and you have tried every last thing you can possibly find. You're you're searching through Stack Overflow uh you know endlessly um and it it might take a day or more you know to sort of fix this one thing and it turned out that it was you know that you you had a comma or you know perhaps where there was supposed to be uh was supposed to be a period or you had misnamed a variable and or just a certain like language convent or um a convention to the piece of software you're using that was you know kind it was in documentation but maybe not perfectly explicated.

It's just little things like that and any number of th those things could could break it and and you're exactly right. All virtually all that is is gone um is removed uh from by by AI and now it's a whole you know now I think people perhaps made the conclusion that all problems uh in software development become abstracted away and solved by AI which is not true but um it uh some of it just means that there's other new uh more sort of like deep and architectural and um sort of system design type uh you know problems that need to be thought about So as I mean I've done this myself and I guess a lot of our audience will have too. Just ask Claude to or or even Chachi Peter do something for you.

I have this example where wheel of fortune game you know was being I was on a networking thing and the guy started this wheel of fortune game which is where you know you guess the this the phrase by picking the letters. You get so many vowels and consonants and so on. He started the game and I thought well I'll write the code as he's doing that. I had the code written, tested, deployed on my computer before he finished the game and which is like three or four minutes and I didn't write a single line of code. Admittedly, it was all in HTML. It wasn't on a server or anything like that. It wasn't very sophisticated, but I think to me it demonstrated the power and speed of something.

Now, to write that would have taken me days, uh probably even longer because I'm not familiar with a code anymore. But I could describe it well enough and and tune it well enough to get a result really quickly. So are you agreeing with me that perhaps those simple kind of things like that AI is really good at, but then you alluded to some areas that you know AI is not good at in the software development world. Can you expand on that?

Yeah, absolutely. Um so yes, I would definitely agree with that. Um simple things, demos, even um I would say even things that maybe in a lot of ways aren't so simple.

I've been um impressed with, you know, uh Claude's ability to uh to tackle uh sort of what they call like oneshot um projects or, you know, first versions uh of things that when they are sufficiently well described and speced out that it can get quite far, you know, with um reasonably like it's it's not just a simple to-do app, I guess, is what I'm saying.

What's a what's a oneshot project? A oneshot is where you give it just a one simple instruction and then it spits it out and it's correct. It works. It does everything. You didn't have to correct it. You there was no bug fixing. There was no

anything else. It was just one you give one command, one prompt and uh and that was it.

Okay.

Um yeah, so that's uh that that can happen and has certainly happened and I think it the the more uh well- definfined the spec is the more likely and the further you can go with that. That being said, um there are more complex workflows.

There are things that that need to that can only be uh defined and described through a process of say like engaging with uh uh you know with a an end user um and and understanding sort of you know if they have a workflow that they do that is you know totally broken or inefficient and touches five different systems and you want to bring that all into one place

you it's you're going to have a you're you're going to have an unsuccessful experience if you try to go off the top of your head and think about how that user could or should use your solution.

Um the the only way to arrive at a good place is through the sort of iterative back and forth and in a lot of cases that is going to lead to a level of sort of like complexity either on the you know the server and the data side of things or even in the user interface that is going to be beyond what an LLM can just you know kind of churn out. Um and um and it's still I would still say that there's left to its own devices. There's still code quality issues.

It's certainly getting better, but um if you want to ensure that certain design patterns in the code are enforced, um a lot of that still requires a lot of sort of um um upfront awareness and sort of top- down management um through your own attention, through rules uh that you can establish for,

you know, for an IDE. Um but things of that nature. So that's that's kind of what I think the landscape sort of looks like right now in that regard.

Yeah, I think couple of I wrote down a couple of things as we went through. So you mentioned I guess I don't know whether you use the exact word. Yes, you did. You use exact word multiple systems. So I certainly use multiple systems.

I have three screens here and I'm running between apps all the time. Now some of that could be coded up I guess. But then there was this unstructured workflows.

I would say I think you said it I don't know whether you said exactly those words but that's what I picked up an unstructured workflow is something that you can't necessarily describe like if I was to follow if I was to get someone to follow me around for a day they could say oh these things are structured but Pete all this stuff over here I I can't seem to find the structure in that

I think it would and I think that's the piece where AI is going to really struggle if you can't If you can't structure it and write it down and it's okay this is step A B C T or whatever AI can't come in and help you right

and a lot of people who work in in the real world they have a lot of unstructured

elements to their work

yeah and that's you know it kind of goes back to your what we were talking about earlier with uh you know sort of the nature of hallucinations and the probabilistic nature of LLMs and that it's like if you leave open space for it, space that's not well defined and everything, it's going to come up with its best guess about what that ought to be. And if you haven't defined that space, well, it's not only is the guess probably going to feel wrong to you, but you might in all likelihood not even have a clear idea of what right is.

You only know it's right because whenever you happen to do it, you have a a particular goal in mind and then you can say, "Okay, well, that's good enough. " Um

yes

but that that is a core issue uh there that it's the the success of automating with AI and this goes back to that AI readiness thing is is that you in that business process mapping is that part of that is that you are defining clear endpoints clear goals and then you just you're you're really structuring or maybe dstructuring down to its its fundamental roots those workflows and if you can't do that um that's a first step.

Yeah. Yeah.

So I have a little example that I think might illustrate this to the audience is if you go to say check GPT and you you get it to do something complex.

Um you know say uh you know help me define you know through all these documents help me write an email or something and then you go and read it and you go ah it's not good enough and you go back and you say you can literally say to it it's not good enough you've missed some things try again and it will literally try again and do and it'll do a better job the second time and you think well why didn't it do that the first time you know why didn't it actually give me the better result the first time and I think some people have been pointing that out to me and I've been testing that with this and I've just fascinated that um if I if I'm more demanding towards it, it seems to step up.

It doesn't automatically step up. And so back to your point about the outcomes, you need I need to know if I'm sending an email, I need to know the standards that that I'm setting. And sometimes they can be very low because I'm just a casual email to a friend and sometimes they're very high because it's a high stakes negotiation and I'm gonna have to gauge, you know, where I sit on that consequences thing, right?

And I think that's again where the human comes back in, understands the consequences of the situation that I'm in as a human and can then steer it to um towards the outcomes in the context of the real world, not in the context of the AI chat because because of course it's very limited, you know, has a limited amount of memory, a limited amount of scope to which what it can operate in. I guess if I was expanding on that idea, imagine I had a portal here and I I had somebody a super smart person just through that window there, but I only gave them a little bits and pieces of information, you know, and they came back with me with the best that they could.

They couldn't really understand what I'm dealing with over here, right? What what is happening because the context is so much broader. It would take me a long time to get there. They'd literally have to follow me around uh to see and live my world. But I mean, I've worked with business partners and that can be really helpful. But there's still it's a little bit like the analogy of the boxer in the ring is the one that really feels the punches and really um knows how tired and how they feel and whether they can win or lose the fight. Everybody in the audience has their opinion, but not one of them is in the ring feeling it, you know.

And I think this is again it's this abstract this there's just this abstraction that comes with AI and you can't sort of just pull them in. Hope I didn't go too far off track, but

No, no, no.

I think that I mean this a very good distinction to draw because it's like you a lot of times we talk about the limitations of LLMs or AI and everything and there's a lot to be said about that but I think the when you the sort of the the analogy of like the the very intelligent person on the other side of a window that you have to manually pass context to I think reframes a lot of those limitations is realizing that it's like that it's yes there's things that are uh relevant to AI specifically but a lot of It is is this challenge of sort of like you know total global holistic context transfer transfer to another entity whatever that entity may be you know because we see that that challenge

in our just our daily work when we have to try to you know kind of um sync between teams between marketing and sales between you know uh you know sales and engineering uh trying to get everybody talking the same language and pushing towards the same goal is kind of another version of that same problem. Yeah. Well, do you have any advice for for companies that are looking to adopt AI?

I do. Um the first thing is to recognize is that you know uh you know quick easy try to like quick hits uh you know just simple chat bots just buying everyone uh co-pilot seat it those things are not likely to deliver the kind of ROI that you're hoping for. Mhm.

Two is to probably think big about the what you hope to achieve with AI automation.

Um you know think you know assume that what is possible is maybe larger than you think it is. So kind of just even as like a a thought u you know generation sort of tool you think in those terms.

Three is to then you know start on these today because the reality is uh these there's a whole you know sort of a preamble of like planning and a of that all that AI readiness we talked about it cannot be done quickly it cannot be done in a quarter it'll probably take you know multiple quarters of time um it will probably in all likelihood depending on what scale of uh you know company you happen to be and the the talent you have on your team you might need outside resources you might need to develop um you know skill sets or hire um outside help uh just to get your mind wrapped around this u so you want to begin sooner rather than later um and I guess uh you know would there be anything

else I guess from that point then it's like just uh try to um try to ensure that sort of uh if you are doing anything that's missionritical with it ensure that you're building in verifiability ility, auditability or explanability uh as a first class design principle into your systems because this gets to really like sort of the core of uh how you actually get ROI from AI. You know, if you have to go back and check everything it does, even if it's 99. 999% correct, if you are in a 100% type of business, the you know, all that means is that you are going to you still have to go back and check everything because one error can cost you dearly.

So the way you solve this is by building in that ability to you know to understand exactly what the AI did and what data it used at every step in the chain. Be able to inspect that very quickly in sort of the the human operator's interface and everything. So they don't have to go and look in five different systems to check something because that's where the value falls apart. Being able to do that quickly is where you know you can uh you can collapse that that time down to where you actually get uh value. Yeah.

So I mean you you said start today. What what compels a business to start today? Why do they need to start today?

Uh because if you look at the you know um some of the big players you know companies with real scale um you can whether you see this in case studies or even you know conversations I've had with um you know friends that are at you know so fortune five type companies in you know up in leadership you know and they're they're all talking they all have big initiatives going on. They're going top down, not bottom up. They're they're they're they're all talking in uh you know terms about they're talking about AI readiness. They're talking about business process mapping.

They're talking about AI governance and they're talking about how this is going to lead to not just the augmentation of workflows but the total uh the total transformation of and you know and sort of re-imagining of what even workflows exist and how they look.

Um the biggest players are doing this and they have the resources to pull it off on a shorter time scale than most um you know so whatever industry you happen to be in there's going to be a gap that opens up very very quickly between you and them and it's I think I predict that it's going to be all too easy to look very uh legacy very archaic very dinosaur in a much shorter period of time than most people think and because you can't move faster than you and then it's really possible in all this. You need to be moving today because if you find yourself behind, panicking is not going to get you there faster. You're going to have to move at the same speed anyways.

You know, there's not going to be any accelerating of it. Yeah.

So, I suppose those those companies that are most vulnerable are those that are operating in a sphere where the bigger companies could come and over overrun them, right? Um you know, as they push themselves down. Um I I can think of the counter example.

I can't think of

well actually I can think of perhaps it's these the big bookkeeping and the big the you know the accounting and that sort of thing where where these big companies have been used to operating you know with the corporates now they can find with AI they can operate so much more cheaply they can offer service offerings further down and that's going to come effectively eat the lunch

of you know the smaller companies because these guys are so far ahead.

Yeah. Yeah. It's um it that's exactly correct. You know, it's it's it's a it's making I think wider segments of the market, you know, more accessible.

Um, and it's it's it's an interesting it's a bit of a barbell thing that I see going on here that it's um the probably the companies that seem most well positioned to uh take advantage of the current era are yeah like the the big operators that are well resourced um you know typically we think of them as moving slower and in in a lot of ways they still do but they they are plugged in enough to recognize the challenge and they are able to dedicate resources. On the other end, there are sort of very small but very techforward companies.

Um there there's some some great folks I talked to uh um that are you know small operators in the the data unification territory, but they they sell a they're selling a product. Um uh which you know essentially that solves one layer of that AI readiness conversation and they they're doing great. Um but everybody kind of in the middle and especially if there is um not a really tech forward um sort of philosophy and everything those are probably the most um in danger because they don't move that fast and they they don't have the technological positioning to sort of like challenge up market or move down market.

Yeah. Yeah. Well, that's that's fascinating.

So any particular industries you think are are the first to be affected by the move to AI? It's a great question. I, you know, I mean, everybody's kind of getting it at once. I mean, because we, everybody's pursuing it, um, so assiduously. Um, however, I I think that, uh, where we've seen a little bit of, uh, in initially maybe a little bit of, um, being slow out the gates uh, in the areas of say finance, accounting, I actually think that there's going to be some pretty big breakthroughs there pretty soon.

Uh because the everybody's speaking the lang that I see is now speaking the language of like you know AI governance and auditability in these things and it's only a matter of time before the understanding of how to distill that down into system design happens. I don't think it'll be more than another quarter or two. And once that happens, uh I think it's it's going to be like that. There's going to be a lot of racing ahead. Um because there's there's a lot of decision-m that can be improved. Uh whether it's loan approvals or you know, like fraud detection.

Um you know sort of u market insights and uh you know investment recommendations that when you have uh a strong system of auditability and verifiability behind it um are now just you know in there's incredible value that's to be there and the the payoff is ends up being obvious. Yeah.

Yeah. Yeah. Well, this has been fantastic. So, we're running out of time here. Um, Joe and, uh, we could go deeper and I could talk all day about this is such an interesting, fascinating topic and developing fast. Do you have any any questions for me or anything you want to talk about that we haven't already?

Gosh, you know, I mean, I I'm always just kind of curious to hear where, you know, because I I come from a you I'm steeped in the technological side of things and I find that I often misapply my own sort of understanding of the space and where things seem um from sort of an over overly technologized sort of perspective. And I guess just kind of like um sort of like from where you sit, I guess like what do you seem to find to be like sort of like the dominant conversations or you know um more perhaps even some of the more interesting wrinkles um around you know AI that uh people you talk to are happen to be saying.

Yeah, I think there's been a realization that um that there has been a little bit of overreaching of what AI can do. Um, and that's driven by a narrative from the big players, right? They want to get subscribers, they want to get people on, and I would call that the consumer retail, like using the LLMs on a day-to-day basis. I consider that to be sort of retail consumer. Um, where it hasn't really had this the impact that I think it will, and it's coming now, is in in its ability to change businesses and change the way businesses operate. And I think the it's taking that um that uh glowing vision of what it could be and then a big reality check on what it really can do.

So we're going to have this I think we're going for it right now. There's this reset. Okay, that there's the vision, but really what what you can do is down here. Now everyone's starting to work down here and we're starting to see people really having an impact. Some of the stuff we've done has had an impact on businesses and now it's starting to get um real meaningful um use cases that are really affecting companies on a day-to-day basis. That that's kind of the macro view that I've got. Um I mean of course there's a lot of stuff going on in you know the race I guess we got these the model races you know between the big models and you know will there be one? Will there be two?

Will there be only one? I mean gosh they're racing like crazy. And of course we benefit from that because we get uh I guess subsidized access to uh to these powerful models um which is primarily propped up by um by uh you know the the the investors but as they race to own the market. Um and then of course I think we're starting to see things hitting us like higher power prices, high costs of of hardware. I heard that um um what was this? Steam machine uh is a new AI, not AI, it's a new game game console. It's it's it's almost 50% higher price than was expected because they can't get access to the GPUs at the prices that you could a few years back.

So, we're starting to see some really big macro effects um that are affecting us in I guess that's a small way, but I think it's going to be quite pervasive. Yeah.

Yeah. That's um that I think is the looming conversation. I think everybody knows is coming, but I think maybe they the scale is yet unappreciated. And I think it's it's completely unclear to me where that's going to lead, but I think it's going to be important.

Yeah. Well, Joe, one more question for you. AI, is it hype or help?

Uh simple answer, it's a help.

You know, uh the the nuanced answer is I think in uh at the you know, the be the hangover from the early AI hype era kind of has set in and I think that's um led people to um erroneously conclude that it it was just hype. Um but I think as kind of we've kind of gone into in our discussion, I think that the reality is is that it's it's definitely a help. Um and that it's going to be revealed to, you know, exactly how that plays out, you know, more and more, you know, over the next year or so. Well, this has been a real pleasure, Joe. Stick around after the podcast. I want to talk to you a little bit.

We'll have to have you back on because I think we've got more to explore, but uh thanks very much for coming on.

Yeah, absolutely. It's my pleasure, Pete. Thanks for having me.

All right. Thanks. Well, that was fascinating talking to Joe. He had lots of insights about how to implement AI into businesses. Um he comes from the software world and he's adapted his business uh to AI implementations. One of the key areas that we noted was the need for smaller businesses to act now and not become dinosaurs. And this is because the bigger companies are investing very heavily in AI and the trickle down effect is going to displace many small businesses who cannot compete.

So move with the times and act now. The second thing is we rolled out a bunch of actual tips specifically for small businesses. And one I'll leave you with here is if you think that rolling out a chat across your organization is going to work and give you a return on investment, you're probably going to be disappointed. And let me talked about some of the metrics like data hygiene, data uniformity and those sorts of things that are needed to be set up early on when you roll out a an AI implementation. Well, if you want to know more about Joe and his business, follow the links in the description. If you want to know more about Skillin and what we do, do the same. I hope that helps.

And thanks for watching.