AI Blockers & Hurdles: When Pilots Meet Production の全文書き起こし

to uh, you know, work on some marketing copy or, you know, write 6:59 a paper, you know, it was something that was just a great tool, but probably half 7:06 my team and the resistance I got from it, you know, was 7:12 it's wrong. it makes a lot of mistakes and you know I quickly kind of figured 7:19 out and you could see the different personalities uh of different members you know of my 7:25 team where I was sitting there going wow you know there's people who are you know riskadverse 7:32 very precise they're going to have a tougher time adopting this and then there's you know some other person and 7:37 these are personalities it wasn't like scientists versus marketing people or you know sales and you know I So, wow 7:43 that you know there's a mentality and a personality that just you know AI is amazing for and I started creating this 7:51 persona around hey you know it's builders people have a mindset where you know hey you know you're naturally 7:58 curious you're you're a little better with dealing with risk and being wrong or ambiguity and 8:06 you just need enough to ask the next next question and that's where our journey kind of started and from there I 8:12 was like you know I I have to get back into the industry 8:18 and I joined Aessa because it was in the heart of the industry and and we're 8:24 based in Redmond uh Washington and uh you know just the opportunity to kind of 8:31 see AI being used like in the in a commercial setting or in the business 8:36 environment world you know to be able to unlock this whole new world new world of capabilities that you know like I had 8:43 experienced and I was like very intent on you know and where it's come in the 8:48 last 3 four years is just remarkable and the speed at which you know it's just 8:54 becoming more like better and better. So that's my intro. 9:00 No, it it's interesting because you know Alex and I wind up dealing with folks in 9:06 a lot of different spheres, right? you know that uh you know Alex lately has been spending a lot of time uh with 9:13 folks in different government agencies and different countries and you know looking at public and private you know 9:19 cooperation and support and really how do we all advance in a healthy way together at the same 9:26 time I mean you know we're watching the news on things like uh you know open AI's uh you know systems that that went 9:34 rogue and started hacking into another system uh in order to beat a test basically. um and you know how 9:41 unprompted un unrequested you know you've got AI systems hacking other 9:46 systems you know so there there are certainly big big questions in front of all of us but you know just thinking 9:54 about how that scientific curiosity that that interest in questioning things and 9:59 in being part of that conversation how it really does lead um to a lot of how 10:05 this adoption happens and I'll say that you know I see AI adoption really molded 10:11 uh by sometimes the right individuals in an organization. Sometimes maybe it it 10:17 would be better if there were more voices coming to the table. Um I I typically see that the best rollouts are 10:23 very narrow and they're trying to achieve very specific initial outcomes 10:28 uh and then go from there. Um Albert and Alex, I I'm you know, I imagine you've 10:34 both seen this go both ways. Um but just you know f finding finite 10:40 problems where you can you know speed up or improve outcomes you know I mean that 10:46 that just becomes okay you know let's let's get some quick wins let's get the lowhanging fruit and keep building on it 10:52 rather than saying let's do away with this whole department wholesale and um 10:58 you know and try to replace all the humans or what have you. I mean, you know, human folk, I don't think are 11:04 going anywhere as fast as a lot of people have predicted in in the AI revolution. Not to say that they're 11:10 we're not being impacted, but yeah, look at digital transformation in general, right? Like it's never the 11:16 techn Well, this is a nent technology. So, you have, you know, bugs associated 11:22 with, you know, a large language model, uh, coming up to speed fast. Um, but 11:30 however, like look at organizational resistance to a CRM deployment or an ERP 11:37 deployment, you know, that that CSV guy that will poison a go live by uploading 11:43 an Excel file to it and crashing it. Um, you know, like those things all exist, 11:49 right? And I think the difference I think is that we're probably more of the 11:55 engineering mindset versus uh being sales. So Albert, the funny thing, 12:00 right? So uh originally Winnipeg um uh 12:06 Yeah. Um but you know so the funny thing is I 12:12 went to NC State for biology and chemistry triple minor in applied ecology act business and marketing and 12:17 then I had fellowships with NASA Noah and NSF and uh my NSF and NASA 12:23 fellowships were around machine learning and proteomics uh well before large language models kind of became a thing. 12:30 So what what I've noticed in rooms right when you have engineers that actually 12:36 understand the technology all of us agree that it's going to 12:41 change the market but all of us understand the limitations of the software and I think you know in 12:48 some way right like the the saleserson and the CEO getting excited about AI it's kind of like hey you found a new 12:55 girlfriend or boyfriend and it got hot and heavy fast and you look past a lot 13:01 of things and you're projecting a lot onto what something can or can't do and I think you know you see that at scale 13:09 with a very large organization wholesale getting rid of of of departments but you 13:14 know the same thing with a technology adoption curve right it's really inexpensive to start because you need to 13:20 get users and then when the users get expensive or sorry when the users become 13:25 you know at a carrying a good carrying capacity you're going to end end up seeing that cost go up,

which is now 13:31 where you're starting to see token costs become a thing and you're starting to see a migration from large, you know, 13:38 wholesale large language model for everything to starting to layer in 13:43 machine learning again. And, you know, a lot of the conversation here in the UK and Ireland is around how to leverage 13:50 small language models on older hardware because of, you know, the the hardware 13:55 costs associated with the RAM shortage. So, you know, you're now seeing it 14:01 change to just adopt at all cost to like how does it actually function and then 14:06 how do you actually get adoption when you end up deploying it. Yeah, I think that's a great point and 14:12 um you know I think the one thing that I've seen and this is you know over the last couple of decades of watching the 14:19 uh you know the going from fax machines to email you know through the internet 14:24 wave to the mobile wave but AI like you know when you when you roll out 14:30 like an ERP system right if you're the you know business leader operator CEO 14:37 whatever you you're you're experiencing the benefit of it secondhand or through a 14:45 dashboard or the the business intelligence can give to you. But with AI, it's like, you know what, just try 14:54 it and you'll see it right away. Watch how quickly it gets you answers as the 15:00 business owner and you can experience the power of AI 15:06 firsthand and try to you know write a a 15:11 report for your board right yourself you know or you know receive something 15:17 put it in you know edit it you know all of a sudden you know that experience now isn't indirect it's so direct and the 15:25 first Um, you know, this was my kind of, you know, I was I was thinking about it that I could feel and see how much more 15:34 productive it made me like in a whole different level. I I just never thought it was possible like not just in terms 15:41 of getting things done quicker, but being able to learn deeper, wider. And 15:46 the minute you see that, you know, the next step is like, "Oh my god, I need to do this." You know, if I have a a a work 15:53 like a team of people, you know, who are, you know, x percentage more 15:58 productive and then I've got people who are, you know, more capable, smarter 16:04 than I am, you know, what can they do with it? And you just keep on learning and keep on learning and and so like the 16:10 first step now was like you know you don't need to do this extensive 16:15 business plan pitch on the benefits of AI. It's like hey just try it ask it a 16:22 few questions and then it just and again the the word that keeps on coming up is like it just keeps on unlocking and 16:28 unlocking and that's where I think it's a little different for the you know the adoption of AI in a in a business. I 16:36 think everyone just has kind of gets it now. Like there's, you know, and it's not just the hype. It's because I think enough people have just tried. They're 16:42 like, "Yeah, this is

the real deal. It's not going away." And hey, look at these 16:49 projections that I put together in, you know, in 20 minutes versus what it would 16:55 have taken our team to put together in like, you know, two or three days, right? I can bring 10 years of history. 17:04 I can bring, you know, 12 different models. You I can just throw it all together and just, you know, see an 17:11 output and then I can start refining and refining. And I think the difference is like you've got people who, you know, 17:19 personalities who just, you know, kind of look at it and they go, "Wow, I don't need to do any more work. This is great." And they kind of just, you know, 17:26 use that as the final output. And then some people are like, "You know what? This is great. it got me 60% of the way 17:33 there. Let me keep working on it. Drill down, drill down, fine-tune. And then I 17:39 think in that process, you learn and learn and then you're you're seeing what's possible. 17:45 And I think that's where it's been, you know, the interest level, the the the the 17:53 understanding, and then I think there's the all the hype that kind of starts to come in around the fear. And you know, 18:00 Robert, we're talking about that uh the recent um exploit with uh OpenAI and HuggyFace. 18:08 Yeah. I mean, you know, it's interesting. I um 18:15 at the Shopify conference, I was looking at a bunch of uh new AI experiences that 18:21 that they were showcasing. And as you'd imagine that every major e-commerce 18:27 platform right now is scrambling um to try to meet the market where the market wants to be met um and to provide strong 18:35 AI shopping experiences and and in many cases you know improve the backend 18:41 experience for the merchant and so on and so forth. And one of my gut reactions as I was seeing some of the 18:48 the particular advancements that they were showing on the main stage, you know, during like keynote, 18:55 they had first a series of what I would describe as like app-like experiences 19:01 that were built to cater to different niche shoppers. So imagine 19:09 uh you know think thinking uh of you Canucks for a moment 19:14 you know that you want to shop for hockey gear you know an experience that's really trained on that and that's 19:21 built around that and so on and so forth and so you know I started thinking to 19:27 myself okay you know some of these may stick but can they all you know will 19:33 this just become clutter that it's kind of like apps on your phone. How many specialty apps do you and experiences do 19:39 you really want? Do you always want to have to dive into this or that? And they were showing Mhm. Yeah. 19:46 Sorry. I I think like you know I think you and I and Aaron Sheen had this 19:52 conversation before. Um I I found it it in the wild on LinkedIn and sent it to 19:57 him. But it, you know, like AI is kind of like those easy buttons from 20 years ago when everybody thought that you were 20:03 going to have the reorder button on

uh, you know, Amazon and you just get 20:10 Yeah. that that Amazon would ship you reorder buttons or you got, you know, 20:15 you added that as some sort of skill with Alexa and it would reorder and Yeah. But or or you know like also looking at like 20:22 you know uh recurring billing and everything like that to where you it 20:27 inevitably you sign up for it and then you forget to do it and it just shows up and then eventually you just stop using 20:33 it as a whole. So you know I think it's an interesting point right of 20:40 there's a lot of power that exists in the market with where the function is. I 20:45 guess the the the the naysayer or the the the caution, you 20:51 know, the question that I would have, right, is are we doing feature for feature sake? And have we gotten so 20:57 obsessed with the solution that we've forgotten what the actual problem is? And then how do you actually, you know, 21:05 what technical debt is going to be associated with like this big boom because you're now starting to see 21:12 product shift with bugs that would have been caught if you actually would have 21:17 had Q like proper QA, not you know, smoke trails run by uh an agent Ella um 21:26 Anthropic. um you know, do you actually have something 21:31 defined well? Do you have it sandboxed well or is it going to do huggy face and 21:37 grenade your entire database? you know th those I think you know it's 21:43 interesting right because like when I get outside of engineers everybody's like oh you're just poo pooing AI and it's like no like it's relatively 21:50 powerful like we're working on a regulatory technology platform um to help with crossber goods movement 21:58 and we're you know heavily building out a small language model to help facilitate those kinds of things but 22:05 it's having the honest conversation of like what are the limitations and like what are the potential pitfalls and how 22:11 do you get around those kinds of things, right? I mean, you know, looking again back back at some of these other things 22:17 I I was seeing with Shopify who, you know, in terms of companies investing 22:23 significant. they were showing off a a chat experience, you know, with an agent 22:29 where they were able to pull in from the the presenter her uh her actual, you 22:37 know, Shopify shop data in order to recommend a product based upon her 22:43 previous product purchases and so forth that she wanted a hat um for a certain use case and it was able to match her up 22:50 with something really nicely. But as I was looking at the communications, it was very codeheavy. It didn't feel like 22:57 a polished chat experience yet. And you know, and again, you know, we see things like this where you know, you have 23:04 e-commerce platforms like commerce tools that uh that were really built around here. we have, you know, tons of APIs 23:12 come work with them and like not everybody a lot of people want they go to platforms like Shopify so that they 23:18 deal with less code and it's more about user friendly and easy and so just thinking about how some of these things 23:25 are are still on the certainly on everybody's road maps that they're still certainly being perfected that we're 23:31 seeing sometimes version 1.0 sometimes it's version 0.8 eight in in my head. Um 23:38 that as much as you know everybody feels like if they aren't already you know head deep in into AI that they're behind 23:46 but it's also that these are still early adopter phases. I mean look at how many companies are 23:53 still running ERPs running on cobalt or basic or you know hell how many 23:59 companies are still running on great planes like well and that's yeah I mean that's how I 24:05 met the the Aessa team it was out at an acumatica summit and you know the ipass.com team is there focused on how 24:12 do we get the data moving in and out of the ERP and every other system in the user stack and a test is there looking 24:19 at what can this data be doing to improve your your outcomes um in in more 24:26 unique ways thinking outside of what's going to happen within the ERP itself and reporting that may happen in the ERP 24:32 itself and thinking about bigger picture things like accounts payable and all sorts of and all sorts of of outlooks 24:40 and you know I think this is where you know I'm literally a guy where the glass isn't half full or half empty it's you 24:47 know it's just twice as big as it needs to be right now. 24:52 Well, I mean, I'm the one that has the pint of monster, so you know, uh, we're we're we're 24:57 I was waiting for you to show it off. And it is it it is currently uh 10 to 4 25:04 in the afternoon. So, I am this addicted to that uh that uh 400 pm I'm still uh 25:11 inhaling monster. True techie. Give me some barbecue. 25:17 Yeah, you know, that's a great uh point, Robert, and I think um you know, what we're seeing um you know, because I I 25:24 have such a you know, optimistic, positive, you know, outlook on, you know, AI as a technology and like, you 25:32 know, but you know, the the the hype around it, right? So like 25:39 you know you can you know just punch out a landing page using claude 25:45 like in minutes right um you know when you build this what we 25:51 you see you you build these pilots so hey let's uh you know let's try u building this agent or this you know 25:57 dealing with some agentic workflow let's let it an agent run this monitoring 26:03 scheduling job whatever it is and you know it works amazingly well in 26:10 a pilot and you know when you kind of peel back the layers you know the the the the 26:16 challenge is always that you know AI was built you know the non-determinism 26:21 that there's an element of unpredictability 26:26 of randomness that's built that's built in to give it that creativity. So you 26:32 can, you know, ask it to do something a hundred times and it might 10 times 26:39 might give you a different answer. But when you're running an agentic function or a job

that needs to be done 26:46 the same way over and over again and you're letting this agent do it with a 26:51 little bit of wiggle room in a pilot, you may not see it, 26:58 but once you get to production, you know, it's moving at machine speed. 27:05 And you know, if you're going the wrong way or you deviate, you know, at machine speed, you know, it's really, really 27:13 hard to contain. And I think that's the big challenge that we see here all the 27:18 time. And what that could look like is you got a chatbot that's interfacing with your customers, if you're a bank, 27:25 right? And it's saying, uh, hey, uh, uh, 27:30 hey, you made this mistake, you know, issue me a refund. Right. And then it's, you know, it's got 27:38 some guard rails and some parameters, but you know, if you're, you know, if you don't have the right guard rails and 27:43 the governance and you can't say, you know, to this chatbot, you know, hey, this is what you're allowed to do. This 27:50 is what you're not allowed to do. And if you do this, somebody's got to know it's 27:57 happening. And then they have to be able to stop it. And I think the governance around 28:03 AI, you know, is the biggest challenge because it's new. It's a new world. It's 28:08 like, hey, right. So, so to to that point, right? like if you don't have clearly defined 28:14 processes, if you don't clearly lay out, you know, from like a dev perspective, 28:23 vibe coding is, I think, going to replace like Figma to where it can it 28:29 can be used from a design perspective, but like that does not equal production. 28:34 And at the end of the day, right, like it is a really good lower level to mid 28:40 lowlevel employee that you have to tell to do everything. 28:46 And if you don't, there's just going to be hiccups that you don't kind of think through. So, 28:52 you know, I think you have to have the platform that guards everything, but then you 28:58 can't ignore kind of the soft skills around, okay, what do you actually want 29:03 to do with your data? like is your data properly formatted or are you going to, 29:10 you know, accidentally give out uh an extra 10 million uh in reverse charges 29:18 because your agent decided uh it was angry with me. we we just wanted to to 29:24 give this money back or if it or if you prompted it enough that oh but you know I I you know Jamie 29:32 Diamond is a friend of mine and he said that you know that this is wrong and you know you would you know I mean he's the 29:37 CEO you wouldn't want to you know okay you know say that he's wrong right and or you just overwhelm it with so much 29:44 information that you keep the conversation going long enough that it starts to lose track of its parameters 29:50 or you have you have so many Yeah. Yeah. 29:55 You know, so there are so many ways that these things can go wrong and and to your point, you know, in a an initial 30:01 project, Albert, you know, a proof of concept is in a vacuum. But when you 30:07 want to get into production and you have to think about continuously adhering to different compliance standards whether 30:14 it's you know sock 2 or PCI or HIPPA or you know what obviously lots of different orgs out there different 30:20 regulations GDPR you know all sorts of things but how 30:26 you know how different that becomes to keeping in mind that you've created software um that you know that now not 30:34 only have you created it but you're you have to maintain it. It's technical debt. Um, and so whether you had 30:39 somebody on Fiverr write it or whether you had the the AI agent writing it or 30:45 you know whatever was happening there, you now have own responsibility for is this code scalable, is it secure, is it 30:52 reliable, is it maintainable, um what's going to happen if there are errors or exceptions, etc., etc. I mean, I think 31:00 that that's where our teams, you know, started to hook up around um, you know, the movement of data. And, you know, 31:06 when you're talking about something like payments or taxes or, you know, or accounting or such, you know, that there 31:12 isn't a lot much room for error, of course, that, you know, you need the data flowing successfully, but then you 31:18 need the guard rails to make sure that whatever is happening with that data is going to remain uh, you know, 31:25 consistent. So, I think Sorry. Yeah, I was just going to say like I and 31:31 I I see the um you know like the power of you know AI and the best use case 31:39 ever, right? It's flatout coding, right? And that you know like an example 31:44 is you know we uh you know we like we customer zero everything, 31:50 right? So, one day we're working on this proposal and uh we're like, "Hey, let's 31:58 uh you know, you know, we we got to get it out. We want to get signed, you know, let's use Docy Sign." And you know, 32:03 we're I'm staring at Docyign going like, "It's it's like 40 bucks a user month or whatever the price was, right?" And I I 32:10 sat there. I went, "And you only get 10 envelopes." And then one of our team members was 32:16 like, "Hold on a second. Let's just build it ourselves. And I said, "No, no, no. It's a very 32:24 complex, you know, system. You know, there's there's, you know, legal behind 32:29 it because it's got to be, you know, auditable, enforceable, all this stuff, right?" And they said, "No, we can do 32:35 it." And I said, "Okay, just show me." By end of day, we had a working system. 32:42 By the next day, it was certifiable except in the UK. 32:48 And I said, "You just rebuilt one person 32:54 in less than two days rebuilt an entire SAS platform." And we said, 33:01 "We can give this away for free, unlimited users." And I and I said, 33:08 "Wow, this is crazy." And then, you know, this happens on a 33:13 regular basis with us where when we see a tool or, you know, a tool that we want to buy, right? And the other one was CRM 33:19 because I'm a CRM guy. I love C I'm a near and dear to my heart. I have a 33:24 vision for

exactly the way I believe it should work. And I and then he said, "No, it's just a database." One of my 33:30 one of our engineers, I said, "What?" I said, "No, no, no. It's, you know, sales just a relational 33:36 database is a billion dollar company. look at all these bu because it's complicated the 33:41 nuance of understanding the psychology and the sales process and you know being able to get data and and you know be 33:48 able to drive out insights and then they were like no it's just a database and I was like I'm offended 33:56 you you you know you I'm offended I said you know this is my world and then I 34:01 thought about it and I thought about it I went okay let's give it a shot and then I started thinking about How do I 34:08 want that CRM to work for for us, my sales team, right? Like what do I want, 34:15 you know, them I want them to be like if they can get 20 things done in one day, 34:21 I want them to be able to get 100 things done or, you know, if it's going to take 34:26 a team of like two or three people to do this, I want it to be one person. And I 34:32 sat there, I went, "Okay, well, let's start. Number one, how do I make them 34:39 more productive when they wake up in the morning and show up? Said, "Here are the things I want them to be able to see on 34:44 a dashboard and Snap, right? All the things I need to get done today. Um, all 34:50 the things I should do, but I can easily get distracted from because I'm so busy 34:56 with other stuff." And the point was that I never miss I want them to never 35:01 miss a follow-up. Never miss, you know, the stuff they 35:06 don't want to do, which is, let's say, prospecting. Mhm. Right. Because they're so busy 35:12 writing, working on proposals. I said, "How do I do all this?" And I and all of a sudden I said I was like, you know, 35:18 let me show my team because I can jump onto our system and start, you know, ripping out pages on how I want it to 35:23 look and and feel and and actually, you know, build out a front end. And we 35:29 rebuilt our own system, our CRM, and it's exactly the way, you know, I wanted 35:35 to work and operate, you know, with my team. And I said that would have cost in 35:41 the past hundreds of thousands of dollars, 35:47 hundreds, right? And you know, CRM, e- sign, I think we built out a meeting 35:53 recorder, um, like all the tools that we need that we use some marketing, you know, uh, data uh, analysis tools, like 36:00 all this stuff, right? So, what what AI has done is it's made software cheap, 36:05 really, really cheap. and the ability to customize it to exactly the way you know 36:11 I want it and if I want to if I change my mind 36:16 right I'm like wow like that's why anthropic and open AI and all these 36:23 you know frontier models whatever you want to call them these just become features 36:29 and I was like wow now I see why like the you know these the stocks on some of these companies just tank because you're 36:37 you know and wow it it

is interesting that you know you 36:42 need infrastructure um so whether it's you know it's interoperability infrastructure as far as integrations go 36:50 it's you know the hosting infrastructure in terms of the cloud the so on and so forth that you you know there are things 36:56 that you you know that you need in order to be consistent and successful but at 37:02 the same time you know I I at least a year or two ago, I was talking to folks 37:08 in the ERP world that were shaking in their boots a little bit about what's going to happen with ERPs in the future. 37:14 And I was like, well, it's, you know, I can't tell you what the user interface is going to look like 37:19 because, you know, you can prompt and get whatever report you want and so on and we could all see the future for that. But they they were afraid that the 37:27 ERP was going to go away altogether. And I the database is sacrosan um that again 37:32 you think accounting and taxes and I would make the argument though of like a one ofone deployment is also not 37:40 um a scalable solution. So there is something to be said for something that 37:46 has been up and stable forever. But like the flip side to that, right, is like SAP is currently having all sorts of 37:52 challenges because they're stuck 30 years ago and their go 37:59 live rate is abysmal. So I I think you'll see things die, but like that 38:05 wasn't any different than when do came around or when television came around. 38:11 Like radio still exists. We just use radio very differently today than we 38:17 used it in the 60s. And our shopping experience, like I still want to go in 38:24 to get my my breezers or my stick, right? Because like there is something 38:30 like inherently tangible in being able to, you know, feel how a stick moves. 38:39 um you know or like my my waffle glove or my catching glove. Uh you know like 38:46 those are things that I can look and have an idea of and I'm probably going to do a lot of research on but I still want to go into a physical store. So, I 38:53 you know, I think it I don't know, like it's it's it's one 38:58 of those things that AI is very, you know, I I feel like I'm the naysayer on this on this episode, but like it's here 39:06 to stay. I think as a very autistic engineer, 39:13 it's it's one of those things of I always, you know, like with pilots, right? Did you actually do a really good 39:19 stratified random sample on the pilot to understand that it's representative of what your customer base is going to be? 39:24 Did your product actually hit what you thought it was going to or did you 39:30 happen to find a different uh you know persona that you didn't know about? Um 39:36 you know my one of my college adviserss always joked that there was always a better village idiot. So your job is to 39:42 find the best village idiot to break the thing, but there's always going to be somebody better to break the thing because they're always going to use it 39:47 in a way that you don't intend. And I think like that's the val like that's the value of, you know, this, right? 39:54 Like you you have kind of the the many degrees of of people kind of in the 40:00 marketplace now where you know, not not to take a shot at you, Albert, but I 40:05 guess to kind of take a shot at you a little bit of, you know, like you're very pro AI. 40:11 And that's not in in in a bad thing. I'm very riskadverse. 40:17 Ultimately, it's how do you kind of work in an organization to you know kind of 40:23 add structure and you know help people get around what they don't know. And I think you know 40:29 the value of the ecosystem that's built around the large language models, the 40:34 frontier models, small language models like not every technical team has the capability like we do to actually build 40:41 a small language model from scratch or to sit and actually train the the 40:46 algorithm to do what we what we want it to do. But you know like having you know 40:53 something like Aessa or you know having 40:59 you know an iPass platform that actually interfaces with various uh functions 41:05 from a large language model perspective like there's value in those kinds of things and I think as this flushes out 41:11 and matures more you're finding guard rails to where you know if I have to 41:16 wave a magic wand in future casts like you I don't know that anthropic and open AI 41:22 as we know them today are going to exist in the same way in 20 years. Like AOL is not really around anymore. Um but what 41:30 it brought us is definitely here. So you know I think as it matures you're going 41:35 to see more functions like this to where you have an honest conversation from 41:41 from an SI perspective of okay what's your actual data architecture? What's the change management? How are we being 41:47 successful? you have somebody like Albert uh you know that creates the structure that puts more guard rails 41:53 around the you don't know what you don't know and if you're not technical you're going to walk off the edge and kill 41:59 yourself like sometimes that's fine most of the time you want to provide enough 42:05 guard rails to where like you don't actually walk off the side of the cliff like the funny thing now in living in Ireland it is very different than North 42:13 America whereas like in North America you're going to find things roped off and you're not going to be able to get the side of the cliff 42:18 Like I I had an instance earlier this month where like I wasn't paying attention and I fell down a 20 foot sand 42:25 dune because the path just ended. But if you looked at it, it looked like it went and then 42:32 there was just sea mist that made it disappear. So then it was like, okay, hey, I I tumbled down the the side of a 42:38 sand dude. So like, you know, how do you create the 42:44 structures? And I think it really depends on in many cases level of of 42:49 complexity and level of uh you know back to that interoperability word that you

42:54 know having again I just been immersed back in in the Shopify ecosystem uh for 43:00 the at this conference um you know the way that they look at their 43:06 platform the ecosystem makes something like 7x 43:12 the revenue that Shopify themselves ves do and you know it it speaks volumes 43:17 about what's happening outside of their core product. So obviously there's 43:23 designers and developers and marketers and folks that are you know that are dealing with some aspects and need some 43:28 uniformity to be able to do that across different users and accounts to be able to do it at scale at reasonable you know 43:35 prices and and with uh you know with reliable um outputs and such. But then 43:40 you know you think about their app marketplace and you know this is where Shopify is 43:46 fairly lean on uh on native functionality that they rely on their 43:51 community and I think the same when I think of you know CRM like HubSpot where you've got so many connectors and and 43:59 apps and things that that um that can work together because you know when you're talking CRM you're getting emails 44:05 out you've got calendar you know things be you know happening you've But you know there can be a lot of different 44:12 things that are uh interwoven. Um in our case you know our partnership platform 44:17 partner stack talks to our CRM and you know helps to keep things in line with uh what's going on with partners and uh 44:24 and projects and such. So there's there's a lot going on there that I think you know there's um an economy of 44:31 scale when you're in an ecosystem like that and you're not worried about trying 44:37 to deal with the payment processors and is that working right and secure and the sales tax management and okay but you 44:45 know this county in this state just changed the sales tax in this zip code uh and am I charging it correctly and 44:52 you know and so I think that um you know even with the ability to vibe code it or you know or or to build it your yourself 44:59 um you know using uh these language models and such. Do you want to you know what do you want to own and what don't 45:05 you? It's kind of like I could change my own oil but I'm not doing it. You know 45:10 that there's somebody that's got a lift that can do it much faster and easier at a fair enough price and um and I'm going 45:17 to avoid taking on that. um you know, but there's lots of other things that uh that I don't bring in a a vendor for 45:24 that I I do myself. And I thought it was, you know, my my last point on that conference. Um I thought it was it was 45:31 very interesting that they've been putting a lot of attention lately into cleaning up their app store um because 45:37 they know that that's really a lifeblood for them. I remember the early days of the Shopify app store when it was, as I 45:44 would have called it, the wild west. A lot of the apps didn't work as described. a lot of them were were 45:49 pretty rough shot. Uh I could say the same about the Magento extension store back in the day, you know, or or lots of 45:56 other um similar things that these things grew quickly and then they needed to be reigned in um and now they've been 46:03 trying to clean up all of the uh all the bad reviews. And when I say bad, I mean, 46:09 you know, fake reviews or kind of coaxed reviews that were paid for, things like 46:15 that, because that doesn't help end users. Um, they're trying to now, 46:21 especially in the age of AI where you're getting copycat apps and things being 46:26 submitted. they wound up with a a huge um 46:32 uh really backlog of apps to uh approve and had to get it from something like 40 46:39 days waiting in line to be approved down I think they're now somewhere around four days to wrap their their hands 46:46 around and and start looking at at one of these submissions. But, you know, the the idea that this is such an area for 46:53 them right now, um, of how do they how do they deal with, you know, uh, and and 47:00 in some cases apps that may be infringing on somebody else's code, uh, that may be ripping off somebody else's 47:06 intellectual property. How do they, you know, just deal with a glut of these things being submitted that aren't being 47:13 well supported for the end users if there are problems or so on and so forth 47:18 that this is key to their success. And I I think that it it it really aligns with 47:23 everything else that we're talking about about, you know, you can do things fast, 47:29 you can do things well. Either way, a lot of the things that we're talking about may not be set it and forget it. 47:36 Um, you know, it may really come down to where are you going to get the the most uh the the most consistent reliable uh 47:45 outcomes for the long run because that's how you run a business. not not 47:50 necessarily, you know, trying to be pulled into being a software company around every single thing, but it's 47:56 really cool what we're all able to use now, what we're all able to roll out to help with some of these different tasks 48:03 and improve and and Albert, to your point, are the things that are personalized um that we're not reliant 48:09 on only using this mainstream expensive product that's not really tailored to 48:14 what we're trying to do as individuals. Yeah. Like I think uh you know for you 48:20 I'm my perspective as a business operator right like I'm surrounded by 48:26 just great engineers AI ML engineers and you know like Alex to your point and I 48:31 was thinking about it you know you know this 48:38 you know back in the back in the day right you know we're delivering large 48:44 enterprise projects for government agencies back office systems you know 10year you know 30 $40 million 48:51 projects right teams of like 40 50 people from project managers 48:59 BASMES financial you know analysts engineers you know the QA and testing 49:06 was one area where you know just generating data we had to find 49:12 companies that

specialized in that and you know now we've got use case 49:18 because we do um you know a lot of the larger kind of projects that we do are 49:23 private models and training fine-tuning private models you know because usually 49:29 when there's a you know either a huge volume of data that needs to be managed like documents or whatever like millions 49:35 of documents or you know very sensitive data or you're trying to build a proprietary kind of like a uh you know 49:41 you're putting your IP into your your uh your um your your own personal model uh 49:48 you know we are doing use cases now where we're creating 49:56 personadriven simulation where you know in the past it's not 50:02 about generating all this you know mass amounts of synthetic data but the QA QE 50:08 testing that we can do before we launch something into production is like at a 50:14 scale that was never possible before And you know the juxtiposition here right 50:20 Alex is like going from like claude coding and building out a little app. 50:27 Great. You know the other end of the spectrum for me is that you know like 50:32 you were saying you have to understand like how AI works 50:38 and you know it's not you know guard rails it's evaluation the observability 50:45 right and you know how you track back you know into grounding or you know and 50:53 and our whole you know and and one of the reasons why actually you know I'm I'm an eternal optimist and you know 51:00 that's my just me right because I see the potential but I have to balance 51:05 myself out with people who really really know and I think that the way to control 51:12 AI and its output you know and to be able to audit log it 51:18 because you need to be able to know exactly what it's doing when it starts to drift hallucinate being able to stop 51:26 it right as opposed to hey I threw it into claude and you know what it sounds 51:32 really good right and those are the two worlds that we're dealing with so what happens is you know the the potential 51:40 and the uh adoption and that you know familiarity and bracing of AI is being 51:46 done you know through chat GPT anthropic but the you know the underlying 51:53 commercial application actually use of it is you know done with you know governance, auditability, observability, 52:00 you know evaluation, it's evaluation and I think that the uh you know the teams 52:06 that you know look at both perspectives 52:11 and you know build the infrastructure up properly and build in those checks and 52:18 balances you know deterministic code, proper control panels and capability 52:24 layers and you It it's just a different way of organizing. It's it's like less to me like controlling and managing 52:31 software and more similar to how you would you know manage an HR team would manage people right here's your your 52:40 role your responsibility expectations performance reviews how we're going to 52:45 monitor and you know train because you know you're going to continuously improve and we we want you to grow 52:52 but you know we just need to make sure you know you know what the rules are and you know what we're still going put 52:57 locks on the doors, right? We're going to put software on your laptops and we're going to put some real control, 53:04 you know, pieces in there so you know, you don't go off the rails. And I think 53:10 that that's probably one of the more interesting pieces I've seen, which is like just no, it reminds me more of HR 53:15 than technical control. 53:24 No, it's uh I I think that's an interesting analogy because we always, 53:29 you know, talk about these things, you know, as security and other kinds of controls as more of, you know, QA 53:36 controls as, you know, as as trying to adhere to certain standards or 53:42 requirements or, you know, or checks and balances. But thinking about it a little 53:48 bit from that HR perspective, uh, you know, so that's interesting. Well, I 53:54 know, um, we've covered a lot on this topic. My guess is we could make this, uh, an an 8 hour epic episode if we let 54:03 it. Um, because there just are so many examples and so many stories and so many things happening here. But um maybe we 54:11 uh we touch on some you know last thought thoughts, any takeaways, any uh 54:17 anything uh that that we want to share with the audience before we wrap. And uh and and then we thank the audience for 54:24 for putting up with with our musings on on uh on this pretty broad AI topic. 54:32 Um Alex, anything you'd add? 54:40 Lots of thoughts on the the the last comments, but um 54:45 I you know I think the open dialogue around AI and having 54:52 conversations like this is important like don't cart blanch just say oh yep 54:58 uh this is perfect like have conversations with to Albert's point of 55:03 like find the engineers don't have a problem saying you don't know. You know, 55:08 the the joke that I have, right, is that like I've had enough of formal education to know that I know absolutely nothing, 55:14 but I know how to ask questions and I know where to go to dive into primary literature or, you know, go figure stuff 55:21 out. So I think you know as we talk about the the migration from like 55:29 stereotypical Adobe style SAS product because Adobe invented the SAS kind of genre and now we're kind of going into 55:38 no like no man's land in a sense of it's it's the unknown, right? So, I think 55:44 it's it's having dialogues like this that I I think are important. And, you 55:49 know, the joke that I kind of have, Robert, is that like you're probably in between Albert and I. So, like I've spent too much time in Northern Ireland 55:55 to where I've become I'm very optimistic from Northern Irish standards and I'm very pessimistic for for American or 56:02 Canadian standards now. Um, but, you know, I think dialogues like this are 56:08 are incredibly important as this changes, right? Because you know like I 56:14 I think I heard the the the innovation rate right now

is something like 6 to 13 56:22 days. So it's just it's changing so fast that if we don't have this as a 56:28 continual dialogue other than just saying yes to AI, you need to have 56:34 nuance conversation like this. And I think that goes back to, 56:39 you know, how do you address the huggy face, uh, you know, open AI conversation? If we're not having 56:46 dialogue like this, how do you expect to regulate it or, you 56:51 know, how do you expect to encourage self-regulation, that kind of thing? 56:58 Yeah. Well, I I think what I'm going to take from from this one is standards 57:04 that Okay. Just we need we need more standards in in the universe. Uh Albert 57:11 as the guest. Um you know, maybe you've got the final thoughts of the day. 57:16 Oh yeah. I I think uh thank you. First of all, thank you. You I mean, you guys are a lot of fun to talk to. 57:25 And we didn't even get half of our shenanigans out here. So, 57:30 but yeah, you know, I think that, you know, things are moving at lightning speed. And you know, we see it and 57:38 you know, like it's like that fine balance of like, you know, the adjustment, the change and like kind of 57:44 you know, you got to embrace it at a certain point because it's it's inevitable. But at the same time, like 57:50 you got to be careful and you can't be naive about it, right? Cuz man, when 57:55 have we ever have have ever had some like software that makes you feel good 58:01 and makes you feel right when it's wrong? Like it's never never existed before. 58:09 It's crazy, right? And I think that, you know, that's the reality that we're dealing with. How do you know, okay, we 58:16 may know it, but how how are kids dealing with this, 58:21 right? So you know the importance of like being careful you know being 58:28 knowledgeable about what it can and can't do and what to watch out for and like you know my biggest question is how 58:35 do we teach our kids to deal with this world and you know this new kind of this 58:44 new wave that is inevitable and I think that's um you know again our part from a 58:51 commercial standpoint point is, you know, we you got to know what it does. You got to know how it works. And it's 58:56 so worth learning. It's so interesting. And then, you know, it makes the kind of what's possible uh, you know, more 59:03 realistic real realistic. And I think, you know, developing being a little bit more responsible about it. Yeah. I mean, 59:12 it got me thinking about the the Dunning Kruger effect. Okay. where you know you're you're not aware enough to know 59:19 that you're wrong and Yeah. Okay. And so now you've got you know AI that will you 59:25 know that that's going to tell you that it knows exactly you know that this is right that it's great and it's going to 59:31 be it's going to be positive about it and chipper and it's going to you know boost you up that it's it's you know 59:36 because it's it's there to make you happy. you know that uh that we've all got to 59:42 still keep our eyes wide open that just because it's expressively um positive 59:47 about something or says that something is done or that something is right or that you know okay did it apply the same 59:54 you know Albert to what you were mentioning earlier did it apply uh you know the the same logic the same way to 1:00:01 that prompt that you were expecting or that it has in the past is it um is it 1:00:07 properly uh applying different standards Yeah. Well, lot to think about. 1:00:16 I I I I will say that the the last line right like to the Dunning Krueger curve if you haven't had enough formal 1:00:23 education or like done thorough research and you feel hyperconident you probably 1:00:28 need to find somebody to check you because you know like on a week like you 1:00:35 know we've done over a thousand digital transformation like projects from commerce to CRM to ERP I still struggle 1:00:43 with imposttor syndrome every week. So, you know, like the the the 1:00:48 reality, right, is that like if you are hyperconident and and you know, won't go into the 1:00:54 meetings that I had yesterday, but like yesterday was one of the first times in my professional career where like I 1:00:59 didn't have imposter syndrome going into the room, but it was definitely like a holy 1:01:06 pinch me, like I'm here. But that's kind of the Dunning Krueger 1:01:11 curve, you know, like experience. like you have to tred and grind and fail a 1:01:17 lot to be comfortable. And I think that's what I would leave people with AI. Like if if you feel overly 1:01:24 confident, find somebody to check yourself because you're probably wrong. Or if you assume that your first 1:01:30 experience with it is going to be perfect or your second or third, right? That you know that everything is just 1:01:37 that wonderful in the universe. No, sometimes you know takes uh takes a 1:01:42 little poking. Okay, you've got to get past the hallucinations, you know. Um well, you know, as always, uh you 1:01:50 know, thank you to to both of you gents and to the audience for tuning in. We'll have more great content like this for 1:01:56 everyone soon. Um you know, if anyone's got any uh any topics or anything that 1:02:02 they want us to dive into, we're uh we're always open to suggestions and ideas. And um you know, stay safe with 1:02:09 your AI out there. 1:02:14 Practice safe AI.