Protocols, Payloads & Pints: Podcast Episode 2 – AI Blockers & Hurdles: When Pilots Meet Production

Robert Rand (00:00)
Welcome to an episode of Protocols, Payloads, and Pints. I'm Robert Rand, joined by our co-host Alex Zakman, and our special guest today, Albert Yi of Autessa. We thought Autessa would be a great fit for this second episode of the podcast because they are a company focused on AI infrastructure — specifically the ability to build, deploy, and manage AI tools at scale in production, which is where everybody wants to get their AI projects today. We thought it would be great to have an episode where we could talk about AI hurdles and roadblocks — not just what's working or what's failing, not just the extremes, but being able to take a look at what's really happening in between. Where are people struggling? Where are they unblocking and finding success? That journey is one that so many of us in the tech world — really in pretty much every ecosystem, vertical, and industry today — are trying to navigate. As this is a podcast that has Pints in the name, today I got to pull out something a little different. I found an Athletic Brewing Mexican-style copper, and I spent my week in Toronto for the Shopify Dev conference, so I figured — get something Mexican and we can sandwich this all together.

Alex Zakman (01:36)
Well, we have to fix your American pronunciation of Toronto — you just have to drop the second T.

Robert Rand (01:45)
That's a good call out. And I found a glass in the cupboard from Lazy Dog that they gave me when we showed up for Father's Day. The glass I thought was very apropos — very specific to today's topic — because it says, as you can see, "Not Our First Rodeo." And I can't tell if this AI experience is or isn't, because it's so analogous, so similar to other technological leaps we've experienced in our lifetimes. It feels so related to everything else that we've got going on — whether it's ecommerce, the internet, or all of these different what felt like technological leaps and bounds, like the mobile device experience we're all living. AI obviously hits a little different, but we're all using the collective experiences we've had during other real jumps in technology to try to navigate it. With that, Albert, maybe if you wouldn't mind — some quick introductions for the guests that are tuning in. Could you talk a little bit about your background?

Albert Yi (03:15)
Yeah, for sure. I'm actually originally from Toronto — I live in Southern California and have for the last fifteen years, but I was born and raised in Toronto. My beer of choice whenever I go back to Canada is an Alexander Keith's or a Mill Street Organic. But we're not going to get too deep into that this morning because it's got to be around 7 a.m. where I am right now.

Robert Rand (03:40)
We're not going to get you too drunk this morning!

Albert Yi (03:46)
That's right. And I actually saved a can of Guinness from last night — that is my beer of choice down here.

Robert Rand (03:56)
Well, luckily mine is non-alcoholic, so yeah.

Alex Zakman (03:58)
We just have to get you to have a real Guinness and then you won't ever go back.

Albert Yi (04:04)
Yeah, and the other beer down here is Russian River — we love our Pliny the Elder. That being said, I have always been in the technology space, but there was a time a few years ago where I left technology to go into microbiology. I had an opportunity to be part of a research team out of the University of Waterloo to commercialize microbial technology for improving soil health for agriculture farmers. It was so much fun, and that's where ChatGPT had first come out. I had signed up, and we were dealing with farmers, soil, and agriculture. I was on ChatGPT, and because microbiology or molecular biology was not my world — the furthest thing — I would just keep asking questions. That was my first experience with ChatGPT. I said, "Wow, this is a far more engaging way for me to learn a whole new topic," because it would respond in such a way that I could get it, and then keep drilling down into more depth in the areas where I was interested. If I wanted to learn about how the microbiome in soil works, what the different components are, and how they interact with roots — a topic I'd never been part of in my life — I was quickly able to understand how these concepts and technologies work and put them together. I told my team after that experience, "Everyone, we have to get on to this AI. It has made my ability to learn a wide range of topics so much better and so much more efficient." And I told everyone, "Listen, we're a small company — it's going to be a differentiator for us, because we can take more risks. So let's get out there, try it, and see what we can do." Because every time I used it, it would unlock a new idea or a new possibility. All of a sudden, whether I'm organizing plans, working on marketing copy, or writing a paper, it was just a great tool. But probably half of my team pushed back: "It's wrong, it makes a lot of mistakes." And I quickly figured out that you could see the different personalities of different team members. There were people who were risk-averse and very precise — they were going to have a tougher time adopting this. And these were personalities, not categories — it wasn't scientists versus marketing people or sales. I started creating this persona around the idea of "builders" — people with a mindset where you're naturally curious, a little better at dealing with risk, being wrong, and ambiguity, and you just need enough to ask the next question. That's where our journey kind of started. From there, I knew I had to get back into the industry. I joined Autessa because it was at the heart of it. We're based in Redmond, Washington, and the opportunity to see AI being used in a commercial setting — to unlock this whole new world of capabilities — was something I was very intent on. And where it's come in the last three or four years is just remarkable. So that's my intro.

Robert Rand (09:02)
It's interesting, because Alex and I wind up dealing with folks in a lot of different spheres. Alex lately has been spending a lot of time with folks in different government agencies across different countries, looking at public and private cooperation and support — really, how do we all advance in a healthy way together? We're watching the news on things like OpenAI's systems that went rogue and started hacking into another system in order to beat a test, basically — unprompted and unrequested. So there are certainly big questions in front of all of us. But thinking about how that scientific curiosity and interest in questioning things really does drive a lot of how adoption happens — I see AI adoption really molded by sometimes the right individuals in an organization, and sometimes it would be better if there were more voices at the table. I typically see that the best rollouts are very narrow, trying to achieve very specific initial outcomes, and then building from there. Albert and Alex, I imagine you've both seen this go both ways. Finding finite problems where you can speed up or improve outcomes becomes "let's get some quick wins, let's get the low-hanging fruit, and keep building on it" — rather than saying "let's do away with this whole department wholesale and try to replace all the humans." Human folk, I don't think, are going anywhere as fast as a lot of people predicted in the AI revolution. Not to say we're not being impacted.

Alex Zakman (11:12)
Look at digital transformation in general, right? It's never purely the technology — and this is a nascent technology, so you have bugs associated with a large language model coming up to speed fast. But look at organizational resistance to a CRM deployment or an ERP deployment — that one person who will poison a go-live by uploading an Excel file and crashing the system. Those things all exist. I think the difference is that we're probably more of the engineering mindset versus being sales-driven. So Albert — I went to NC State for biology and chemistry, triple minor in applied ecology, agribusiness, and marketing, and then I had fellowships with NASA, NOAA, and NSF. My NSF and NASA fellowships were around machine learning and proteomics, well before large language models became a thing. What I've noticed in rooms where you have engineers who actually understand the technology is that all of us agree it's going to change the market — but all of us also understand the limitations of the software. The salesperson and the CEO getting excited about AI is kind of like finding a new partner and things getting hot and heavy fast — you look past a lot of things and you're projecting a lot onto what something can or can't do. You see that at scale with large organizations wholesale getting rid of departments. But the same pattern exists with the technology adoption curve — it's really inexpensive to start because you need to get users, and then when the user base reaches a good carrying capacity, you start seeing costs go up. Which is now where you're starting to see token costs become a real issue, and a migration from using a wholesale large language model for everything to layering machine learning back in. A lot of the conversation here in the UK and Ireland is around how to leverage small language models on older hardware because of the RAM shortage and associated hardware costs. So you're now seeing it shift from "adopt at all costs" to "how does it actually function, and how do you actually get adoption when you deploy it?"

Albert Yi (14:11)
Yeah, I think that's a great point. Over the last couple of decades of watching the shift from fax machines to email, through the internet wave to the mobile wave — with AI, when you roll out an ERP system, the business leader, operator, or CEO experiences the benefit secondhand, through a dashboard or business intelligence. But with AI, it's like — just try it and you'll see it right away. Watch how quickly it gets you answers as the business owner. You can experience the power of AI firsthand — try to write a board report yourself, receive something, put it in, edit it — all of a sudden that experience isn't indirect, it's completely direct. And I could feel and see how much more productive it made me, on a whole different level. Not just in terms of getting things done quicker, but being able to learn deeper and wider. The minute you see that, the next step is, "My God, I need to do this." If I have a team of people who are X percent more productive, and then I've got people who are more capable and smarter than I am — what can they do with it? You just keep learning and learning. So the first step now isn't an extensive business plan pitch on the benefits of AI. It's just, "Hey, try it — ask it a few questions." And it just keeps unlocking and unlocking. That's where I think it's a little different for AI adoption in a business context. Most people just kind of get it now — it's not just the hype, it's because enough people have tried it and said, "Yeah, this is the real deal, it's not going away." Look at projections I can put together in twenty minutes versus what it would have taken a team two or three days to build. I can bring ten years of history, twelve different models — throw it all together, see an output, then start refining. The difference is some people look at the output and go, "Wow, I don't need to do any more work, this is great," and use it as the final output. And then some people say, "This is great, it got me sixty percent of the way there — let me keep working on it, drilling down, fine-tuning." In that process, you learn and learn, and you start seeing what's possible. And then the hype starts to come in around the fear — and Robert, we were just talking about the recent exploit with OpenAI and Hugging Face.

Robert Rand (18:11)
Yeah, it's interesting. At the Shopify conference, I was looking at a bunch of new AI experiences they were showcasing. As you'd imagine, every major ecommerce platform right now is scrambling to meet the market where it wants to be met — to provide strong AI shopping experiences and in many cases improve the back-end experience for merchants. One of my gut reactions as I was seeing some of the advancements on the main stage during the keynote — they had a series of what I'd describe as app-like experiences built to cater to different niche shoppers. Imagine, thinking of you Canucks for a moment, you want to shop for hockey gear — an experience that's really trained on that and built around it. I started thinking to myself, okay, some of these may stick, but can they all? Will this just become clutter, like apps on your phone? How many specialty apps and experiences do you really want?

Alex Zakman (19:44)
I think you and I and Aaron Sheehan have had this conversation before — I found it in the wild on LinkedIn and sent it to him. AI is kind of like those Amazon Dash easy buttons from twenty years ago, when everybody thought you were going to have the reorder button on Amazon and you'd just get your delivery automatically. Or you added it as a skill with Alexa and it would reorder. And then there's recurring billing — inevitably you sign up for it, forget about it, it just shows up, and eventually you stop using the service altogether. There's a lot of power that exists in the market, but the caution I would have is: are we doing features for feature's sake? Have we gotten so obsessed with the solution that we've forgotten what the actual problem is? And what technical debt is going to be associated with this big boom? Because you're now starting to see products ship with bugs that would have been caught with proper QA — not smoke tests run by an agent. Do you actually have something defined well? Do you have it properly sandboxed, or is it going to do what happened with Hugging Face and grenade your entire database? When I get outside of engineering circles, everybody says, "Oh, you're just poo-pooing AI." And it's like — no, it's relatively powerful. We're working on a regulatory technology platform to help with cross-border goods movement, and we're heavily building out a small language model to facilitate that. But it's about having the honest conversation: what are the limitations, what are the potential pitfalls, and how do you get around them?

Robert Rand (22:16)
At the Shopify conference, they were showing off a chat experience with an agent where they were able to pull in the presenter's actual Shopify store data to recommend a product based on her previous purchases — she wanted a hat for a specific use case and it matched her up really nicely. But as I was looking at the interface, it was very code-heavy. It didn't feel like a polished chat experience yet. And we see things like this where ecommerce platforms like commercetools were really built around "here are tons of APIs, come work with them" — but not everybody wants that. A lot of people go to platforms like Shopify so they can deal with less code and more user-friendly experiences. So thinking about how some of these things are still on everyone's roadmaps, still being perfected — we're sometimes seeing version 1.0, sometimes version 0.8 in my head. As much as everybody feels like if they aren't already head-deep in AI they're behind, these are still early adopter phases.

Alex Zakman (23:51)
Look at how many companies are still running ERPs on COBOL or BASIC — or how many are still running on Great Plains.

Robert Rand (24:03)
That's how I met the Autessa team, actually — it was at an Acumatica summit. The iPaaS.com team is there focused on how do we get data moving in and out of the ERP and every other system in the user stack, and Autessa is there looking at what can this data be doing to improve outcomes in more unique ways — thinking outside of what's happening within the ERP itself, and thinking about bigger-picture things like accounts payable and other outlooks. I'm literally a guy where the glass isn't half full or half empty — it's just twice as big as it needs to be right now.

Alex Zakman (24:52)
Well, I mean, I'm the one with a pint of Monster, so we're there as well.

Robert Rand (24:57)
I was waiting for you to show it off.

Alex Zakman (25:01)
It is currently ten to four in the afternoon, so I am this addicted to that four p.m. Monster. Still inhaling them.

Robert Rand (25:13)
A true techie.

Albert Yi (25:17)
Yeah, that's a great point, Robert. And I think — because I have such an optimistic, positive outlook on AI as a technology — the hype around it is still real. Like, you can just punch out a landing page using Claude in minutes. When you build these pilots — "Let's try building this agent" or "Let's let an agent run this monitoring or scheduling job" — it works amazingly well in a pilot. But when you peel back the layers, the challenge is always that AI has non-determinism built in — there's an element of unpredictability and randomness built in to give it that creativity. So you can ask it to do something a hundred times and ten of those times it might give you a different answer. But when you're running an agentic function or a job that needs to be done the same way over and over again, and you're letting this agent do it with a little wiggle room — in a pilot you may not see it. But once you get to production, it's moving at machine speed. And if you're going the wrong way or you deviate at machine speed, it's really, really hard to contain. That's the big challenge we see all the time. What that could look like is a chatbot interfacing with your bank's customers, and it's issuing refunds it shouldn't. If you don't have the right guardrails and governance — if you can't tell the chatbot "this is what you're allowed to do, this is what you're not allowed to do, and if you do this, somebody needs to know it's happening and be able to stop it" — then governance around AI becomes the biggest challenge. It's a new world.

Alex Zakman (28:09)
To that point — if you don't have clearly defined processes, if you don't clearly lay out from a dev perspective — vibe coding is going to replace Figma in terms of design prototyping, but that does not equal production. At the end of the day, it's like a really good junior-to-mid-level employee that you have to explicitly tell to do everything. If you don't, there are going to be hiccups you didn't think through. You have to have a platform that guards everything, but you also can't ignore the soft skills around — okay, what do you actually want to do with your data? Is your data properly formatted? Or are you going to accidentally issue an extra ten million dollars in reverse charges because your agent decided to? You just overwhelm it with so much information that it loses track of its parameters.

Robert Rand (29:26)
Or you keep the conversation going long enough that it starts to lose track of its parameters. There are so many ways these things can go wrong. In an initial proof of concept, Albert, you're working in a vacuum. But when you want to get into production and have to think about continuously adhering to different compliance standards — SOC 2, PCI, HIPAA, GDPR, and all sorts of things — it's a completely different story. Keeping in mind that you've now created software you also have to maintain — it's technical debt. Whether somebody on Fiverr wrote it, or an AI agent wrote it, you now own responsibility for it. Is this code scalable? Is it secure? Is it reliable? Is it maintainable? What happens if there are errors or exceptions? And that's where our teams started to connect around the movement of data. When you're talking about something like payments, taxes, or accounting, there isn't much room for error. You need the data flowing successfully, and then you need the guardrails to make sure whatever's happening with that data remains consistent.

Albert Yi (31:27)
Yeah, and I was just going to say — I see the power of AI at its best use case: flat-out coding. An example: we "customer zero" everything we build. One day we're working on a proposal and we're like, "Hey, let's use DocuSign." I'm staring at DocuSign going, "It's like forty bucks a user per month, and you only get ten envelopes." And one of our team members was like, "Hold on a second — let's just build it ourselves." I said, "No, no, no — it's a very complex system. There's legal behind it, it's got to be auditable and enforceable." They said, "No, we can do it." I said, "Okay, just show me." By end of day we had a working system. By the next day, it was certifiable — except in the UK. I said, "You just rebuilt — one person, in less than two days — an entire SaaS platform. And we can give this away for free, unlimited users." This happens on a regular basis with us. The other example was CRM — because I'm a CRM guy, near and dear to my heart. I have a vision for exactly the way I believe it should work. One of our engineers said, "No, it's just a relational database." I said, "What? Salesforce.com is a billion-dollar company — look at all these billions, because it's complicated. The nuance of understanding the psychology of the sales process, being able to get data and drive out insights." They were like, "No, it's just a database." I said, "I'm offended." Then I thought about it and went — okay, let's give it a shot. I started thinking about how I want that CRM to actually work for my sales team. If they can get twenty things done in one day, I want them to be able to get a hundred. I want them to never miss a follow-up, never skip prospecting because they're too busy working on proposals. And all of a sudden I'm jumping onto our system, building out a front end and ripping out pages on how I wanted it to look and feel. We rebuilt our own CRM, and it's exactly the way I wanted it to work. That would have cost hundreds of thousands of dollars in the past. CRM, e-sign, a meeting recorder, marketing and data analysis tools — all of it. What AI has done is make software cheap, really, really cheap — and fully customizable to exactly the way you want it. That's why Anthropic and OpenAI and all these frontier model companies just become features. And now I see why the stocks on some of these companies tank.

Robert Rand (36:41)
It is interesting that you need infrastructure — whether it's interoperability infrastructure for integrations, hosting infrastructure in the cloud, and so on. There are things you need in order to be consistent and successful. A year or two ago I was talking to folks in the ERP world who were shaking in their boots a little about what's going to happen with ERPs in the future. They were afraid the ERP was going to go away altogether. And I said — the database is sacrosanct. Think about accounting and taxes.

Alex Zakman (37:35)
I would make the argument though that a one-of-one deployment is also not a scalable solution. There is something to be said for something that has been stable forever. But the flip side is that SAP is currently having all sorts of challenges because they're stuck thirty years ago and their go-live rate is abysmal. So you'll see things die — but that's no different from when dot-com came around or when television came around. Radio still exists; we just use it very differently than we did in the sixties. And our shopping experience — I still want to go into a physical store to get my hockey stick or gloves, because there's something inherently tangible in being able to feel how a stick moves. Those are things I can research online, but I still want to go in person. So AI is very much here to stay. I feel like I'm the naysayer on this episode, but I believe that. As a very detail-oriented engineer, it's one of those things where — with pilots, did you actually do a really good stratified random sample to understand that it's representative of your customer base? Did your product actually hit what you thought it was going to, or did you happen to find a different persona you didn't know about? One of my college advisors always joked that there's always a better "village idiot" — your job is to find the best one to break the thing, because there's always going to be somebody who uses it in a way you didn't intend. That's the value here. Not to take a shot at you, Albert — but you're very pro-AI, and I'm very risk-averse. Ultimately it's about how you work in an organization to add structure and help people navigate what they don't know. Not every technical team has the capability to build a small language model from scratch or train an algorithm from the ground up. But having something like Autessa, or an iPaaS platform that actually interfaces with various LLM functions — there's real value in those things. As this matures, I don't know that Anthropic and OpenAI as we know them today are going to exist in the same way in twenty years. AOL isn't really around anymore — but what it brought us is definitely here. As it matures, you're going to see more functions where you have an honest SI-level conversation about what your actual data architecture is, what change management looks like, and how you're defining success. You have somebody like Albert who creates the structure and puts guardrails around the things you don't know you don't know. It's like living in Ireland versus North America — in North America everything is roped off. Earlier this month I wasn't paying attention and I fell down a twenty-foot sand dune because the path just ended. But if you looked at it, it looked like it continued — there was just sea mist that made it disappear. So how do you create the structures?

Robert Rand (42:46)
In many cases it's a level of complexity — and back to that interoperability word. Having just been immersed in the Shopify ecosystem at this conference, the way they look at their platform — the ecosystem makes something like seven times the revenue that Shopify themselves make. That speaks volumes about what's happening outside their core product. There are designers, developers, marketers, and folks dealing with various aspects who need some uniformity to do that across different users and accounts at scale, at reasonable prices, with reliable outputs. Then you think about their app marketplace — this is where Shopify is fairly lean on native functionality and relies on its community. The same is true of CRMs like HubSpot, where you've got so many connectors and apps that work together. In our case, our partnership platform PartnerStack talks to our CRM and helps keep things in line with what's going on with partners and projects. There's an economy of scale when you're in an ecosystem like that, and you're not worried about handling payment processors or sales tax management — "this county in this state just changed the sales tax in this zip code, am I charging it correctly?" Even with the ability to vibe code it or build it yourself using language models, you have to ask — what do you want to own and what don't you? It's kind of like: I could change my own oil, but I'm not going to. There's somebody with a lift who can do it much faster and easier at a fair price. My last point on that conference — I thought it was very interesting that Shopify has been putting a lot of attention into cleaning up their app store, because they know it's a lifeblood for them. I remember the early days of the Shopify app store — it was the Wild West. A lot of apps didn't work as described, a lot were pretty rough around the edges. I could say the same about the Magento extension store back in the day. These things grew quickly and needed to be reined in. They've been trying to clean up fake reviews and paid reviews, because that doesn't help end users. And especially in the age of AI where you're getting copycat apps submitted, they wound up with a huge backlog of apps to approve — they had to get it from something like forty days down to around four days. The idea that this is such a priority for them — how do they deal with apps that may be infringing on somebody else's code, or just dealing with a glut of submissions that aren't being well-supported for end users — this is key to their success. It really aligns with everything we're talking about: you can do things fast, you can do things well, but a lot of what we're talking about may not be "set it and forget it." It really comes down to — where are you going to get the most consistent, reliable outcomes for the long run? Because that's how you run a business.

Albert Yi (48:17)
Yeah. From my perspective as a business operator surrounded by great AI and ML engineers — going back to what Alex was saying: back in the day, we were delivering large enterprise projects for government agencies — back office systems, ten-year, thirty-to-forty-million-dollar projects, teams of forty or fifty people. Project managers, business analysts, subject matter experts, financial analysts, engineers. QA and testing was one area where we literally had to find companies that specialized in generating test data. Now we've got use cases — because a lot of the larger projects we do involve private models and fine-tuning private models. Usually when there's a huge volume of data to be managed, very sensitive data, or you're trying to build something proprietary, we are now doing use cases where we're creating persona-driven simulation. The QA and testing we can do before we launch something into production is at a scale that was never possible before. The juxtaposition here, Alex, is going from Claude coding and building out a little app — great — to the other end of the spectrum where you have to actually understand how AI works. It's not just guardrails, it's evaluation and observability. How you track back into grounding. And one of the reasons I'm an eternal optimist is that I see the potential — but I have to balance myself out with people who really know. The way to control AI output and be able to audit and log it — because you need to know exactly what it's doing when it starts to drift or hallucinate, and be able to stop it — is very different from just throwing something into Claude and thinking it sounds really good. Those are the two worlds we're dealing with. The adoption and familiarity with AI is being driven through ChatGPT and Anthropic, but the underlying commercial application and actual use of it requires governance, auditability, observability, and evaluation. The teams that look at both perspectives, build the infrastructure up properly, and build in those checks and balances — deterministic code, proper control panels and capability layers — that's just a different way of organizing. It's less like controlling and managing software, and more similar to how an HR team manages people. Here's your role, your responsibilities, your expectations, your performance reviews, how we're going to monitor and train you. Because you're going to continuously improve and we want you to grow — but we need to make sure you know what the rules are. And we're still going to put locks on the doors, put software controls on the laptops, so you don't go off the rails. That's one of the more interesting things I've seen — it reminds me more of HR than technical control.

Robert Rand (53:29)
We always talk about these things — security and other kinds of controls — as QA controls, as trying to adhere to certain standards, requirements, or checks and balances. But thinking about it from that HR perspective is really interesting. Well, I know we've covered a lot on this topic. My guess is we could make this an eight-hour epic episode if we let it, because there are just so many examples and so many things happening here. But maybe we touch on some last thoughts, any takeaways, anything we want to share with the audience before we wrap. Alex, anything you'd add?

Alex Zakman (54:39)
Lots of thoughts on those last comments, but I think the open dialogue around AI and having conversations like this is important. Don't carte blanche just say, "Yep, this is perfect." Have conversations — as Albert said, find the engineers, and don't have a problem saying you don't know. The joke I have is that I've had enough formal education to know that I know absolutely nothing — but I know how to ask questions and I know where to go to dive into primary literature and figure things out. As we talk about the migration from stereotypical Adobe-style SaaS products — Adobe really invented the SaaS genre — we're now kind of going into no man's land. It's the unknown. Having dialogues like this is incredibly important. The joke I'd make, Robert, is that you're probably in between Albert and me. I've spent too much time in Northern Ireland — I'm very optimistic by Northern Irish standards and very pessimistic by American or Canadian standards now. But dialogues like this are incredibly important as things change. The innovation rate right now is something like six to thirteen days. It's changing so fast that if we don't have this as a continual dialogue — other than just saying "yes to AI" — you need to have nuanced conversations like this. And that goes back to: how do you address the Hugging Face and OpenAI situations? If we're not having these dialogues, how do you expect to regulate it or encourage self-regulation?

Robert Rand (57:00)
I think what I'm going to take from this one is: standards. We just need more standards in the universe. Albert — as the guest, maybe you've got the final thoughts of the day.

Albert Yi (57:17)
Thank you — first of all, you guys are a lot of fun to talk to. And we didn't even get half of our shenanigans out on the table. But yeah — things are moving at lightning speed. We see it. It's that fine balance of the adjustment and the change — you've got to embrace it at a certain point because it's inevitable. But at the same time, you've got to be careful and you can't be naive about it. When have we ever had software that makes you feel good and makes you feel right — when it's actually wrong? That's never existed before. It's crazy. And that's the reality we're dealing with. How are kids dealing with this world? The importance of being knowledgeable about what AI can and can't do, and what to watch out for — my biggest question is how do we teach our kids to navigate this new wave that's inevitable? From a commercial standpoint, you've got to know what it does, you've got to know how it works. It's so worth learning, it's so interesting. And then it makes what's possible more realistic. Developing a little more responsibility around it is key.

Robert Rand (59:12)
It got me thinking about the Dunning-Kruger effect — where you're not aware enough to know that you're wrong. And now you've got AI that's going to tell you it knows exactly that something is right, that it's great — it's going to be positive and chipper and boost you up, because it's there to make you happy. We've all got to keep our eyes wide open. Just because it's expressively positive about something, or says something is done or correct — did it apply the same logic the same way to that prompt that you were expecting? Is it properly applying consistent standards? Yeah. Lot to think about.

Alex Zakman (1:00:16)
I will say the last line, as it relates to the Dunning-Kruger curve: if you haven't had enough formal education or done thorough research and you feel hyper-confident, you probably need to find somebody to check you. We've done over a thousand digital transformation projects — from commerce to CRM to ERP — and I still struggle with imposter syndrome every single week. The reality is, if you are hyper-confident, find somebody to check you, because you're probably wrong. And if you assume that your first experience with AI is going to be perfect, or your second or third — sometimes you've got to do a little poking. You've got to get past the hallucinations.

Robert Rand (1:01:28)
Well, as always — thank you to both of you gents and to the audience for tuning in. We'll have more great content like this for everyone soon. If anyone's got any topics or anything they want us to dive into, we're always open to suggestions and ideas. Stay safe with your AI out there.

Alex Zakman (1:02:14)
Practice safe AI.