Joshua Burkhow: Welcome to Alter Everything, a podcast about AI analytics and the future of work. I'm Joshua Burkout. You know, getting AI to produce a complicated and complex answer can feel like a real challenge. And then you're inevitably going to be asked to bring that answer into a business meeting where you also likely know someone is going to ask you where those numbers came from. And someone else Is going to want to change an assumption that you used deep in your prompt. And then on top of all that, most importantly, the person responsible for the decision needs to know whether they can stand behind it. You know, they need sources and facts. Now, suddenly, you're having a much bigger and crucial conversation about how work actually gets done. Today, We're looking at what it's gonna take for AI to earn a place in the everyday work of a business. You know, the kind of work where AI and humans are gonna have to complete it alongside of each other. Now, I want you to think about the knowledge that lives inside your team, the person who knows why a calculation works a certain way, the colleague who can spot a problem before anyone else sees it, the experience that takes years to build. And can be difficult to pass along. What happens when those people can put that knowledge to work through AI? And what needs to be in place for the rest of the business to trust what they build? Those are the questions, those are questions about technology. They're also questions about leadership and the people we're asking to make this change work. And that's where today's exciting conversation takes us today. Let's jump into it.
Joshua: Hello everybody. Sometimes I get guests that I've been longing to have on the podcast for a long time. sometimes I schedules work out and it just happens right away. But this is one that I've been wanting on for a long time. Th this person has had an immense impact on me, and I I wanna get into a really meaty conversation about AI today, with none other than our own Alteric CEO, Andy McMillan. Andy, welcome to the podcast.
Andy: Thanks for having me on and that's a very kind intro of you. I'm I'm not a tough guest to get when you're running the Alteryx podcast, and I'm willing to come on and chat with you and I've been looking forward to this as well.
Joshua: I love it. I love it so much. Before we get into the meat of of talking about all this stuff going on in the world with with AI, we have some things to talk about for sure. I I'd like to have people understand how you got there. I don't think you have a prototypical CEO path that you know every every other CEO has had. And I I find it really interesting. That I I'd love for you to share if you you don't mind.
Andy: Well, I I don't know, I mean, maybe there's I don't know if there's a prototypical path or not. That's a fair point. I do think a lot of folks come up through the finance side of the world. So you do see a lot of CFOs going into CEO roles. you see a lot of people in kind of COO or operational roles as well. know, occasionally some folks out of sales backgrounds. and I think now what we are seeing more and more often, especially with this massive sort of technology shift happening around AI, especially in software companies, as people sort of transform what they're doing. We are seeing people with my kind of background, which is a product background, starting to take on more CEO roles, which I think is interesting because we think of founders often as having that kind of product background. but now I think we're seeing more professional CEOs coming out of that area. my background is is not only out of the kind of product management side of the world, actually forever ago, I actually started my career as a As a software developer and I I joke that I was so good at writing code
Joshua: Amazing.
Andy: that people introduced me to product management and thought maybe just come up with the ideas and let somebody else build them. but yeah, I very much come
Joshua: Right, right.
Andy: from the technical side of the universe.
Joshua: Yeah. C can you sort of highlight why you think that helps you? Or what gaps does it fill in having this sort of knowledge and experience come to the table for for at least for all turks?
Andy: I mean, I think there's certainly some value, perhaps obvious value in really deeply understanding the thing that your company builds or does. So if you're I don't know, running an auto company, having a background building cars is probably helpful. we're a software company, we build software for our customers, and so I think having a background in building software is useful. I do think companies go through points in time where sometimes the building of the thing isn't the core problem of the company. You're scaling distribution, managing the financial structure of the company. And and sometimes when companies go through those periods of time, I can understand why someone with a a background in that part of the business ends up taking the leadership role of the company overall. I think that's frankly the position a lot of software companies were in, even three and four years ago. There was a lot of focus on, hey, founders have come up with really great ideas, especially in like the B2B software side of the world. How do you scale that? Like, how do you get that out to the world? How do you ramp up sales team and a marketing effort and customer success and all those things? And so that's really valuable. I don't mean to diminish people that come from those backgrounds running companies. I think now there is this huge shift going on where every software company is kind of looking at themselves in the mirror and saying, okay, like why is our software valuable in a kind of post-AI world? And that's not a distribution question. I think that's a core product. question. And so I think that's why the rotation of sort of product centric folks at the helm is happening right now.
Joshua: Yeah, it's interesting because one of the things that I've observed here, is your perspective and your energy and your drive towards innovation. You And I I I think maybe that's just part of the the understanding of how product innovation works, but also, you know, maybe maybe a really valuable asset in the world that we're like in the thick of, the chaos of innovation. Does that make sense?
Andy: It makes perfect sense. I think one of the things about CEOs right now, whether they come from a product background or not, is you better be pretty curious about how this massive technology shift is going to impact your business. I think part of being curious, at least for me, is getting my hands on stuff. And so, you know, I've got my work laptop loaded up with a million tools that I'm using and integrating with Altrix and building stuff. I've got my home set up. wired up and I'm spending some of my evenings building stuff and thinking about how all this comes together. And I genuinely enjoy that. I it is something I've always liked building stuff. I've always been interested in software. It's again how I started my career as a software developer was I thought it was fun to build stuff with software. So I think having that core interest helps because you're trying to think about, hey, how is this new technology going to impact my customers and how they use my products and services? And so understanding that technology is is useful. And at least for me, the way to understand something is not reading the manual. It's installing it on my machine and trying to do stuff with it and and sort of iterating
Joshua: Yeah. Do it. Yep.
Andy: and and you know, playing with it a bit.
Joshua: So w one of the things I wanna shift gears towards is is this first concept and idea. I I really have been wanting to pick your brain on this topic. you and I have talked a lot about sort of personal AI ventures and all the stuff we're learning and and learning how to do hands-on keyboard. But then w we come to work, we put our you know, our work hat on and we we sort of say, Okay, how do we make this work for our customers. Like how do we make Alteryx and AI tangible and and practical for enterprise AI? And I'm wondering if you have just general themes and then I'm gonna poke and broad on a on a couple questions I have.
Andy: Well, I think maybe one of the big framings is talking about who our customers are is an important part of that conversation. we have a customer base that I would say is technical. You know, they understand data very well, right? So if you're a tax accountant, you're pretty technically adept at how data works. if you're in sales ops or supply chain management, you know data and logic behind data really well. What our customer base often is not is software developers, right? And and that
Joshua: Right.
Andy: is a different sort of landscape of technology. And so one of the things that I think we have done for a long time really well is help make people that are sort of data technical able to use software to do incredible things with their data without having to double click on everything and get into writing Python or SQL and starting to feel like they're a software developer. instead of, to use my example, sort of a tax accountant. And one of the things that I think we spend too much time in AR and not talking about is this idea that somehow software developers or forward deployed engineers are gonna write agents that are tax accountants when the folks writing that stuff don't really know a ton about tax accounting. And yes, you can ask the AI to go do some research on tax accounting for you, but I think that's different than being somebody who deeply understands not only tax accounting, but how the taxi accounting works inside your company and and how it works across your different divisions and geos and decisions you've made. And so I believe one of the opportunities with AI, much like we did with data tooling like Python and SQL, how do we make that accessible to somebody who has the kind of technical competence in their area and their domain of expertise that's not software development, but is data technical for them to consume. And so I'm spending a lot of time and effort thinking through how do we enable people like that to be wildly successful using AI to solve problems and to think about how it impacts their role and their job and their scale. And I'll say that group of folks from the moment I took this job have told me what they love about Altrix is it helps them automate things and scale because there is, in their words, a never ending stack of work, business changes and requests on their plate. So how
Joshua: Absolutely. Yep.
Andy: do we help those folks use AI to be more effective, more efficient and sort of scale that across their company, I think is a really interesting opportunity space for us to sort of lean in and listen and help them.
Joshua: you and I have talked to a lot of customers and this is one of the things that we don't do that by giving them two thousand lines of Python or a bunch of SQL to to have to read through every single time to to be as productive as they can. And I always come back to this. I I think it's such a fundamental concept that opens up all the other doors around AI is like, hey, do is there an opportunity to make it simple and easy? It does it have to actually be rocket science. W we can make all the, you know, AI agents in the world and do all these incredible things. Cool. Amazing. But at the end of the day, like not not everybody to your point is is a a fifteen year data engineer. Well, you know, they're a you know, an accountant. They they are very good at what they do. Can we meet them where they are? And I I I find it really fascinating. And and so this is we led right to where my questions were gonna go is what what is different about, you know, you and I are at home, we're we're in cloud or codex or maybe an open source model and we're we're building stuff and that's great. It's all sort of personal endeavors. When we come into the enterprise AI world, what changes? Like what what is it about enterprise AI that you're really trying to get through to other CEOs and other leaders across companies?
Andy: I think this is one of the core challenges we're all dealing with is I would say AI seems particularly useful in in two areas. One is personal productivity. So, you know, again, clean up this email, you know, help me think through this thing, help me search for this, all really useful. and then the second is when we build things for ourselves, all the sort of assumptions that AI makes in the background feels like magic. So for example, I built a little project at home playing with data to sort of go through personal finances, right? So we all deal with this. You get sent all these statements. I think banks do a remarkably good job of making it hard for you to get to the actual data. So you get, you know,
Joshua: Just a bit.
Andy: I still get literally, you know, paper copies sent to me or PDFs. And I thought this would be an interesting project. What if I load all this into an agent? And so I feed the PDFs into an agent. I'm using a little local open source models for not sending all my financial data out to the internet and it rips through all the PDFs and it puts it all into a nice table for me and I can I can now pull all this apart. And I ask this little agent, which was really fun to build, and I'm writing some posts about how I built all this stuff that I'll share at some point. But
Joshua: Good, good.
Andy: I'm asking it questions about what I spend money on. And you know it comes back and it's like, you know, your biggest spending area are these categories. And it made up these categories. And it's super useful. And a few weeks later I asked it another question, it sort of goes back and it recalculates some categories and it tells me what I'm doing. At home, I'm not really hung up on the specifics of did it calculate those categories the same way each time? I'm not, you know, managing my family on financial performance metrics based on how they hit these categories, thank God for my children, right?
Joshua: Right, right. I
Andy: My mother used to tell my dad he can't run our family like a business unit. So maybe there's a lesson in there that I'm I'm trying to think through. But but I think there's
Joshua: You're trying to learn, yeah.
Andy: we literally had to make cash flow statements for college, like month by month of what my father was providing and what we were doing. So it's a story for another day. But
Joshua: It's amazing. I love this guy, huh? Yeah.
Andy: but I think look, you know, those those assumptions are what make AI feel like magic. What I didn't have to do was all this nitty gritty work of going through line item by line item and trying to categorize all my spending. So that's awesome.
Joshua: Mm. Yeah.
Andy: If I take that same concept and I bring it into
Joshua: Exactly.
Andy: work and I go, hey, every time I run the accounting process in the background, this AI makes a bunch of, again, reasonable assumptions, just different
Joshua: Reasonable.
Andy: than last time I ran it. I think we called changing the accounting
Joshua: Oops.
Andy: rules every month. I think those are technically financial crimes, right? So you're you're sort of so
Joshua: Pretty close.
Andy: so that what sort of makes it awesome when I'm not living in that finite business space makes it hard to roll out. And I think it it also manifests itself in how we interact with things at work. If I go into my favorite, you know, LLM environment and I build an agent and it pulls together, I did this the other day. I had it look at all of our web traffic. I just plug
Joshua: Mm-hmm.
Andy: it right into, I've plugged into Altrix, I've plugged into Snowflake, and I'm analyzing our web traffic and looking for some patterns. And it comes back with this whole analysis. It's really interesting. But I know the minute I show that to our web marketing team, we're not going to go to the conclusions. We're going to go like, how did it calculate that? Where did that come from? Why is it making that assumption? Like it will not survive first impact on a conversation
Joshua: That's right.
Andy: with other people because none of it is visible to them. They're not going to understand what went on behind the scenes to pull this data together. If they went in and asked their own Claude or Gemini or Codex, the same question, they might get a similar but slightly different answer. So it's not repeatable. So
Joshua: Yeah, not repeat or yep.
Andy: how do I compensate or track or measure my web marketing team if every time I run the report, the numbers change a little bit how we calculated it? but that those set of rules and that consistency is how businesses operate. I can't come to you and go, hey Josh, we're gonna put together your KPIs for the quarter. I'm gonna tie your compensation to those KPIs. Neither you or I are exactly sure how those are gonna get calculated, but at the end of the quarter,
Joshua: Right.
Andy: we'll ask the robots and they'll tell us how they think you did. And then I'll pay you based on that. You're gonna go, hey Andy, I don't feel super comfortable with this arrangement, right? And
Joshua: Yeah, yeah. Good luck. Yeah, exactly.
Andy: and so what was magic in a single player game of me working with some data that sort of gave me directional usefulness is the major. hurdle for adoption internally. And we see it time and time again.
Joshua: Yeah. you're sort of hinting at this this idea of the single player trap. And I want you to sort of h help folks understand that more because I I feel like there's still a lot to unpack in that in that concept.
Andy: Well, I would say this is where maybe the agentic coding is different than how the normal folks are using AI. So if I'm on an engineering team and we're using, let's say, Claude code to build something, I've got an infrastructure behind me that Claude's been taught to use that can sort of make this multiplayer. I can have Claude Do what's called a pull request where it goes out to a Git repository, which is where all my code is sitting, and it can pull in some code that somebody else wrote. And I can have Claude make some changes to it. And then I can say, Okay, Claude, now I want you to submit this as a merge request, which is sort of to tell the person that owns the code I made some changes. And you know, their agent can look at the code and maybe merge that code in. And so I'm sort of working on this shared asset in a way that makes sense. And so I think in that universe, that sort of works. As soon as it gets outside of that universe, I get to I'm using the web interface for Chat GPT and I'm asking it questions. and then it goes off and maybe it writes a little bit of code in a little sandbox to figure something out, like putting, I don't know, two spreadsheets together for some logic. And I go, that was really interesting. And then I go over my CFO, John. I'm like, Hey John, look at this thing. John's gonna go, Well, how can you send me the thing that built that? And I'm gonna go, no, I don't like it's it's somewhere behind this web interface. I don't I don't You you could try to tell your chat GPT to do the same thing, but again, it'll might do it a little differently. So your your mileage may vary, as they say. even if I'm in something like a CLI. So let's say I'm a little bit more technical, and I'm I'm using, I don't know, anti-gravity or Gemini or something like on my computer and I'm in the command line interface and I'm building something, and it's writing a bunch of Python for me in the background. Again, same problem. When I go to my CFO and I say, hey, like I built this really cool thing and it's recategorizing some stuff that we're doing. What am I doing? Am I sending him some saved Python files? Better yet, am I saying to my CFO, hey, what's your
Joshua: Screenshots. Let's go with
Andy: what's your git handle? I'll I'll put this in a git repository for you. I mean like, what are we what are we doing here? What are we doing, right? So so that's what I mean by it's
Joshua: Can we just do this? I just wanna see John's face. be priceless, but
Andy: it's kind of single player, right? Like I I'm I'm
Joshua: Right.
Andy: doing all this cool stuff. My my team's doing cool stuff. probably everybody listening has this maybe I'm in the in their story, the the kind of pointy hair Dilbert boss where it's like I'm out using AI and
Joshua: Yeah.
Andy: I come up with some cool thing on a weekend and I send it out to the whole team. And everybody's like,
Joshua: Let's go for it. Yeah, yeah.
Andy: great, like you know, the boss is doing random stuff in AI that none of us can repeat and we don't understand it and
Joshua: He's going Yeah.
Andy: there's probably flaws in the logic and none of us can dig into the logic, but he's sending us reports where he thinks the whatever the email marketing campaign didn't go well. Like not
Joshua: That's right.
Andy: helpful. And that's the sort of again single player. Like none of it's operating off real repeatable and understandable assets where the marketing team could go, I'm looking at what he did and I, you know, he's using the data assets that we created that make sense, that actually represent our email marketing campaigns, the way they're actually labeled, like whatever that is, we haven't all got there yet with AI.
Joshua: Yeah, I it it's so fascinating to me because you go into a customer and they have this natural pent-up excitement around AI. So do we. It's it's amazing, there's a lot of cool stuff in it. But to your point, it's all this sort of individual personal productivity or projects that they've been able to make progress on. But there's almost this sort of assumption, I guess, is is that this just automatically scales. Like it just automatically becomes this thing that everybody can tap into and all share the same sort of intelligence. But reality's not that. Like reality is that you're you're teaching your models and your your agentic systems on all the things that are important to you. But y I would never take that and use it to do do my own thing. You know, I I this is one where I I think a a lot of customers Got excited, got into it, did the POCs, how many reports were written by industry saying, you know, so so few pilots are getting out of their sort of incubation phase into production? I I think this is part of the that challenge that you're highlighting. You think? You agree with that?
Andy: Well, and it's I I agree with that. I also think it's the natural evolution we see of new technologies. You know, this is the whole, you know, history doesn't
Joshua: Yeah. Across the board. Yeah.
Andy: repeat itself but it but it certainly rhymes. if I if I use an analogy to another time, if I think of the early days of the internet, back into the early days of my career. So I was my first job was actually as a web developer, working
Joshua: Yeah, you just dated yourself, right?
Andy: at EDS. I I helped build the original websites for General Motors. So, like gm.com and Pontiac.com. Yeah, yeah.
Joshua: Wow. See, that's a great story. Ugh, so good.
Andy: it actually a related story, you'll be amazed how long code
Joshua: Please.
Andy: lives. The little window sticker that you used to see when you would price a vehicle, the code
Joshua: Yeah.
Andy: I wrote that built that was live for like 15 years. They just retired it. For for years I would go online and be my I can't believe that's still running. so so anyhow, it it in the early
Joshua: Yeah, we gotta call them, put it back. That's awesome.
Andy: days, there was this big issue that. Building a website was actually very technical, right? You you hand wrote a lot of code
Joshua: Hm. Right.
Andy: and you hand wrote a lot of JavaScript and HTML and all these things that today I wouldn't say are extremely complicated coding environments, but nonetheless, they're coding environments. And so you had marketing teams and commerce teams that wanted to build web applications, but they weren't technical. They were marketers and e-commerce and product manager experts for product lines. And so they would hire consultants and agencies and people to come in and build their site. Their site wasn't editable really because they didn't have any access and they were, you know, it was really expensive and it never quite did what they wanted it to do. And we had all this friction. and everybody was like, basically the internet is a playground for techies, you it's for people that that write software. And, you know, we as an industry eventually fixed that. We came up with these web content management platforms. And what the web content management platforms basically did was acknowledge that a lot of this is really technical. So the security, the web applications, the single sign-on, the e-commerce engines, like all those things are going to be managed by IT teams, technical teams. They're gonna it's gonna take real software. But we also can create spaces in that real software where people can write content and post images and things that aren't technical. We didn't go out and teach everybody in marketing to hand code HTML. We taught them how to use web content management platforms. And we put in place workflows and approvals, right? Everybody in my marketing team can't just click a button and publish stuff out to the internet on our website. There's an approval process and and we have staging environments and all these things that are in place that make it enterprise class, business led, supported by IT, you know, technically secure. So I think we're gonna go through that same thing with AI, which is we're gonna Go through this phase where, you know, phase one is people that want to write Python and play the single player game and have a Git repository
Joshua: Turkeys. Yep.
Andy: are gonna build stuff. And pretty soon, and we see it happening already, people that work in normal jobs in you know finance and in supply chain and and sales operations are gonna say, I need all those things, but I really don't need a Git account and I don't plan to learn to write Python for a living. Not only that, if it can write the Python for me, I still need to know what it does. Right. I I can't I can't bet my job that this thing I can't even read is the right answer. especially I'm in finance. I can't stand in front of my auditor and go, I don't know, like I've got two thousand lines of Python here. Like, yeah, it seems like it kind of works. Like, I'm gonna want some
Joshua: Think it's right. It's probably
Andy: level of understanding and I'm gonna need that shareability,
Joshua: That's right.
Andy: right? It's it's not just, each tax accounting rule was written by one person and only they have access to it, and then they, you know, so like We're going to come up with a way like like a repository to sort of share all these things and build all these things and wire it into the business. So I think all that will happen. And I think when that happens, that's when I think enterprise adoption can really take off. That's when we can say, like I have a I have a you know a workspace full of of skills and logic that the team has built, that the experts know is correct, that I can point my AI agents to go use. So if my AI Agent needs to know about revenue recognition, it it goes and looks up our revenue recognition calculation and actually uses
Joshua: Always knows, yep. Yep.
Andy: it. What it doesn't do is make a bunch of assumptions of like, like here's a reasonable way to calculate revenue recognition and then make a bunch of decisions. It it needs to be constrained to how we actually do it. And so I think we will get there. I think we're helping our customers get there right now. And I think that's the big unlock for real enterprise AI is to get beyond this, you know, single player techie. mode, which we're in now, which is again not uncommon, and we get to a multiplayer business user friendly way to apply the technology, and then I think, you know, things take off.
Joshua: Yeah. there's two two points. Your your example when when I hate even bringing this up 'cause it dates me as well, but I got crazy when JavaScript first came out. Because it was I was knee deep in HTML. I thought it was the coolest thing. and then when JavaScript came out and stuff could move, I was like, this is incredible. Like life life will never be the same. Sorta right, but I I did realize that you you hit on this point that as soon as WordPress came out, right? Talking about CMS, WordPress came out and it gave you UI and it gave you all these buttons and clicks and stuff. And behind the scenes, it's handling the HTML, the JavaScript. And if you you look forward, like we've seen this Tableau did this, right? Having to write p Python to to to write all this code to create visuals. Guess what? Tableau came in. It's like, yeah, just drag and drop these pills and you get this, you know, beautiful state of the art imagery and and and you know charts and graphs and all this stuff. And it so to your point, like, history repeats itself and and it's one of those things where it it it's due for it. I I find this hilarious. I was at a a customer on site not too long ago and One guy says to the other, like, I was on Claude last night. I was doing the amazing things. I I I got all this stuff. It did everything that I needed to do for my meeting this morning. And you you could just tell he was ecstatic about it. He's like, just like clockwork. He was like, Yeah, but it's the only problem is it it produced it all in Python. I don't know Python. I was like, I don't know what to do with that. Do I do I save that? And I was just I I I w it was struggling because I wanted to come up to him and and chat about it, but I also like, yeah, he he'll he'll figure it out. But if I know if one person's having that experience, thousands, millions, hundreds of thousands are are having this experience as well, is where they're they're right on that cusp. And I think the the underlying current behind all that is when a problem starts to surface across the board, it's gotta, you know, it's gotta get solved. They w sort of call it the poppy seed. syndrome where it's like you watch poppy seeds. Once there's enough up there, you it time to cut off and solve that problem.
Andy: I think with a lot of this tooling, one of the challenges is that for the technical folks that build a lot of the tooling, it takes a little while for the need to emerge because they're inherently technical. So you don't see the need right away. I get asked sometimes
Joshua: Yeah. A hundred percent.
Andy: by data engineers and Python developers, you know, what does Altrix that I can't do in Python or SQL? And I'll tell them not I mean,
Joshua: Every day.
Andy: you can you can write whatever you want in Python. The point is there's this massive group of people that don't have the Python skills that you have, that understand data really well, that need tooling that helps them solve problems. So the goal of of Altrix or kind of business user solutions is not to create things you can do in that environment that you can't do a level or two lower in the stack if you want to write lower in the stack. The point is a lot of people don't want to make their living Learning how to write lower in the stack. And even when these agentic tools help you do that, I still have to understand what it's doing. I use this analogy all the time with customers, which is with enough context on topic, and if it's a simple enough conversation, I can sort of understand what's going on when people speak Spanish. Took some Spanish in high school, it's been a long time. Okay, but but but again, a lot.
Joshua: Yo hablo español, ¿sí?
Andy: has to be in context. there
Joshua: Great.
Andy: is absolutely no way I would sign a contract of something important in Spanish. I don't I don't know enough Spanish, right? That that would be crazy.
Joshua: Perfect point. Y Amen. Right, right. Yeah.
Andy: and maybe I could hire a Spanish interpreter to sit next to me and tell me I was going to a very efficient way to go through a contract, right? Which is
Joshua: Yeah, what's this word?
Andy: Which is what we do when we say, well, bring in a forward deployed engineer or have a software developer sit with the tax accounting team.
Joshua: So to
Andy: And every time they want to make a change, they can tell the software engineer who can then code the change and maybe there's some lost in translations to take the translation analogy a little bit too far. or a better
Joshua: You can do the translation. Yeah, yeah.
Andy: answer is like maybe the contract's in English, right? And and and I that works for me. So same
Joshua: Yeah. Yeah.
Andy: situation here, right? The the goal is not to say there are things you can do in low code, no code environments. That you can't do in code-heavy environments. The point is to say, who are you empowering to do those things? And what level of trust and understanding do they have in what's being built? And I don't think the goal of AI is to make sophisticated, important things like your tax accounting completely black box. You're not going to bring in your auditor and go, we have no idea, but we think the robots are running a pretty good tax accounting regime that we can't defend. what I want is help me build a more efficient and effective and AI empowered tax accounting regime in an environment that I understand as a tax account, because I have to put my name next to it and say this is correct. And I think there's a lot being built right now that's gonna help people do that soon. And again, it's not gonna be requiring them to understand Python or getting a Git repository or understanding how to do pull requests and merge requests. Like that's not where the world of tax accounting goes. It's to take those same concepts and put them into the environment where those folks are working, and to have Claude or Gemini or whatever help them be a better tax accountant, help them maybe write an agent or build an agent that does tax accounting with them. Like again, I'm I'm not saying they're not going to agentically build these things, but I think this whole idea that you need a, you know, a quote unquote forward deployed engineer who's not a tax accountant to show up and try to build your tax accounting agent to me is completely backwards. Right. You already have a forward deployed employee. You have a tax accountant right there. They know your tax accounting rules. How do we give them the the
Joshua: Yep. Empower them, yeah.
Andy: tools and the skills through AI that they can build stuff where they go, yeah, this is our tax accounting rule. This makes sense. I will stand behind this and this thing can run in an autonomous way. I can manage what it's doing. I can have it work across my other tax accounting rules. I can know that when other agents do things that impact our tax accounting, that this logic is going to be part of that agent's decision. Like that's a very empowering role to be in. And it really drives up
Joshua: That's right.
Andy: your value as a tax accountant. But all of that is sort of premised on the idea that you're probably not trying to learn how to be a software developer. And I really like your analogy of the HTML to JavaScript jump because I remember being in that world
Joshua: Yeah.
Andy: and the idea was. We're just gonna teach everybody in marketing to write HTML. And HTML is not that complicated. So people were
Joshua: Right. No.
Andy: learning. And right about the time people in marketing were like, yeah, I think I've got this figured out. We introduced JavaScript. And they were like, you gotta be kidding me. Like, this is like real coding. Like this is, you know, this is hard. And so that's when everybody's
Joshua: Wait, functions. yeah.
Andy: like, this is not, I didn't want to be a software developer. Like I could have gone to school
Joshua: Yeah, yeah.
Andy: to be a software developer. I went to school to be a marketer. Right. And
Joshua: That's right.
Andy: so you're right. Then we figured out as technology folks, hey, people
Joshua: Yeah, yeah.
Andy: don't want to write JavaScript by hand. They want to use WordPress. They want to use things like that,
Joshua: That's right. Yeah.
Andy: right? Doesn't mean JavaScript went away. Doesn't mean nobody writes anything in JavaScript. Nobody made an argument that you can do stuff in WordPress that you simply can't do in JavaScript. It's like, no, but the point is, who's using WordPress? And so that
Joshua: It's just absolutely yeah.
Andy: that inflection point is on the horizon here for AI. And
Joshua: Yeah, exactly. Yep.
Andy: that's when it goes from being single player techie tool to being multiplayer business tool. where the people that understand the business are building things with AI, scaling things with AI, standing behind the things that they built in AI. I think that's the sort of exciting frontier that we're just coming up to now.
Joshua: Yeah. All right. I wanna shift a little bit and just talk to software development because it's so front and forward. To all the stuff that's going in AI and the fact that you have a very close background to this this field. what like my message always to to software developers is like, are you working on the most valuable stuff? Because every software developer I've run into, have a few high value projects, one or two. But the vast majority is just the the the translation, the speak trying to speak Spanish in an English world. Like if software developers would get a lot of the tools and functionality that they don't need to work on out to the masses, let them use those tools, Alteryx being one. And then they focus on high value things that literally change the foundation of what is possible. Is am I do I have this right or do you think of it differently?
Andy: Of it as there is an incredible amount of mundane work that happens in software engineering today, whether you're talking about
Joshua: That's right.
Andy: the kind of work that you're describing, which is, hey, I built an entire application for one of our business units and you know helps them solve a bunch of problems, but now they are constantly coming to me with small changes that they want to make to that application, and I'm spending my life maintaining little changes. So
Joshua: Yeah, exactly
Andy: How do I
Joshua: right.
Andy: say, hey, great, like let me give you some tooling so you can make those small changes? And again, might need approval workflows and things like that. But like I maintaining a bunch of stuff that you built previously is not what most people go into software engineering saying, you what I hope I can do is maintain a library of a thousand Python scripts for the accounting team is not, is not where most people say, I'm really glad I
Joshua: Exactly So true. So true.
Andy: did that master's degree in CopSci. I think the second part is
Joshua: Right.
Andy: even even in an engineering team, the reality is. I want to build something. I've got this great idea. And I think my skills as a software developer, even if I'm agentically building, help me think through the logic of how that should work. I still think that's incredibly valuable. And I can give some
Joshua: Mm. Right. Super valuable. Super.
Andy: examples, even in my own life right now, like some stuff I'm doing where the the the knowledge of how stuff kind of works at Ultrix helps me build the things that I'm building because I understand what we're trying to do. When I have an idea like that as a software developer or product manager, what often happens is you go, hey, I have this great idea, I'm gonna build this whole thing. However, I have these mundane dependencies, which are you know, other parts of the code base I need to be different. I need an API on top of this service that's sitting somewhere. Writing that API
Joshua: Sure. I need a service, yeah.
Andy: is not an interesting computer science problem. It's just jamming out some code to like wire that into
Joshua: Yep. Yep.
Andy: whatever the API layer is, right? And unfortunately, that's where a lot of innovation kind of goes to die, where you add up three or four of those things, and there's a team that owns that service, and they're not ready to build the API right now, and they're not sure they want me to build the API on top of their service. And you kind of go like, okay, I'll go work on something else, even though building that thing would have been really
Joshua: Yeah.
Andy: useful to some customers. And my hope and I believe what's gonna happen in this sort of post AI universe is I could say, well, I want to build this thing and I'm and I'm using Claude code to help me build it. And Claude's identified a bunch of the dependencies ahead of time because it's seen a bunch of the code base that maybe I wouldn't have had the ability to go read through myself. And it's identified that yeah, there's some places where I need some mundane work done, some some APIs or an updated library somewhere, whatever that is. And it could actually spec out for me. And I could go to those teams and say, Hey guys that own the APIs, like I I know you're not planning to build this API for this thing, but I need an API on this for this project. And I've spec'd it out with Claude Code. And here's the thing that it wants to implement. Can you take a look? And if you guys are good with this, I'm gonna have Claude build that. I'll send you the merge request so you can see it and then we'll we'll push it on. And I I don't think we are honest with ourselves enough as a software industry of again, just how much of that stuff people deal with every day.
Joshua: Yeah,
Andy: In any kind
Joshua: exactly.
Andy: of scaled engineering team, right? Yeah, if you're a startup and there's five people in a room and I show like, hey, Bob, I'm gonna update your API. Bob goes, okay, fine. Like that's different than you work at a
Joshua: Yeah, exactly.
Andy: large scaled organization, whether you're building software as a company or just implementing software, right? You work at a large global bank or something. Like you as a developer don't just get to go put an API on top of somebody else's service. But again, like I think there's an emerging language of like, hey, I think this thing needs an API for me to talk to it. And our agentic coding platform that we share has identified what it would look like to build that API. It's not super sophisticated. Again, it's not new computer science. I'm not changing how your underlying service works. Here's kind of what it would go do. Is everybody good if this thing just implements this? And that's a maybe a conversation over Slack documented in Jira, and all of a sudden that thing's got an API, and my project moves on. Like that, that's an exciting.
Joshua: That's progress, yeah.
Andy: universe. And I think that same again, everything I think about with software engineering teams doing, I then turn and think about if we had the right abstraction layers and tooling for our users, how do they do the same thing? Hey, I'm building an agent. The agent is doing a revenue recognition calculation, but for it to do a revenue recognition calculation, I actually need this bit of data that sits over here in this universe. Is there a workflow behind that? Is that calculated somewhere? Can I use something that someone's already built? If not, can I figure out how that gets put together? Can I communicate with those teams what that would be? It's no different. I'm just not doing it in Python and I'm not going to use words like merge request and pull request and get repo. I'm going to talk about, hey, on our Altrix workspace, we've got a set of workflows that define this business logic. I need to access some of it, change some of it. AI is going to help me do it. That's the world we're moving towards.
Joshua: So let's dive into that. Let's let's let's talk about the world of this business logic. You you and I and and a lot of folks at Altrick sort of really tied to elevating the importance of solid deterministic business logic and and this idea that we've got all roads are lead to AI and Folks are already getting burned. They're already getting you know, having struggles of various kinds, whether it's token costs or it's risk in in all the forms, security risk, but financial risk, reputational risk. let's start broad and and just say like how do you think about this problem? Can you expand on on that maybe and how you say how do you think
Andy: Yeah.
Joshua: about business logic in the concept of in the in the context of AI?
Andy: The core concept is, and you mentioned sort of deterministic, probabilistic, like sort of fancy words we use in the AI space around, you know, AI does this cool new stuff where it it figures out things in a way that doesn't have to be if then, like we did before. And
Joshua: Yeah. Yep.
Andy: I think that's awesome. Like that's a new set of capabilities we hadn't had before. It is probabilistic. It's using models to figure out what might be the right answer and using logic and reasoning. I tell teams I work with, you know, we never had probably in our software before, right? We didn't we didn't run a dashboard and tell you sales were down in Central Europe in your mid market. And it's probably because of this. Like
Joshua: Right, right.
Andy: software just didn't do that, right? You had to go in and figure it out yourself. And so that's awesome. And the idea that the software could tell me it's probably because, you know, currency fluctuations in the market and a supply chain disruption in Central Europe. You go, that's really useful. I would have had to Start sending emails because the dashboard is red and and why is that? And here AI has helped me do that. There are other places where the word probably is not helpful. This will probably pass audit. we we did we do
Joshua: It's not a good idea. Yeah.
Andy: the month end close the same way this month as we did last month? Yeah, probably. Like, no one feels good about that, right?
Joshua: Probably. Right, right. Yeah, yeah.
Andy: you know, am I gonna get paid my full commission when I close this deal? Probably. Like, that's not the kind of motivating factor for people.
Joshua: Okay.
Andy: So So I I want to be careful. I don't think it's that agents are bad or that deterministic logic
Joshua: Right, right, right.
Andy: is so good. It's that these are these are two capabilities we now have, and we have to figure out how to weave them together and have them make sense. And so when we talk
Joshua: So perfect, yeah.
Andy: about business logic, I think about things that we want to have always be the same. So if I'm calculating my month end close, I want that done using the same rules every time. And importantly,
Joshua: Yeah, exactly.
Andy: I don't want it using whatever it deems to be the best. Current rules available to any company in the market, right? This is not a place for artisanal accounting where you know Claude says, hey,
Joshua: Yeah.
Andy: there's a trendy new way to do month end closed, and I did it for you this month. And I want to close the books the same way I closed them last month. And if you have a better idea, Claude, I want you to tell me, hey, there's a better way to do this. I'd like to propose a policy change. And I can decide if I want to
Joshua: Yeah.
Andy: implement that going forward. Great, super helpful.
Joshua: Yeah we
Andy: Now Inside something like a month-end close, I might want to introduce
Joshua: Yeah, yeah.
Andy: a probably, right? So maybe I've got a 10-step workflow that today does part of my financial close. Maybe it's a reconciliation. What would be really neat is what happens when something doesn't reconcile? Maybe that same deterministic workflow has a new step. And the new step goes, hey, I'm going call Claude, and Claude is going to tell me what it probably might be that didn't reconcile. And it's gonna make a record of that. It's not, it's not doing it for me. It's just going like, hey, Andy or accounting team, you know, this record didn't reconcile. There's this other record over here that didn't reconcile. And I think that's probably the same record. You should have someone look at that. That's a great step to add to my month end close. Then I
Joshua: Yeah. Yeah. huge.
Andy: want it to continue on its deterministic path of doing my month end close. And so for first pass of the month end close, it goes, hey Andy, great, 99%, everything's all set. You got actually three things you gotta look at that. didn't reconcile the first path. And I think I know what they probably are. Awesome. Way
Joshua: Yeah. Yeah.
Andy: better than, hey, I made some assumptions on those, did it myself, didn't
Joshua: Good luck. Yeah.
Andy: tell you about it. And when your auditors show up, you can try to explain to them why I did the thing that I didn't tell you about because I felt like doing it as your AI agent. So business logic is the ability for someone on that financial team or that accounting team that does the month end closed to say, I want to implement some logic. That I know is our business rules that runs the same way. We talk about Vora at Ultics, that is visible,
Joshua: Every time. Yep.
Andy: understandable, repeatable, and auditable. Can I go to my boss or my auditors or my team and say, Yep, like here's what we did? And they're gonna understand it. And again, this is where that sort of low-code, no-code environment matters. When I show somebody an Ultix workflow, I can show an Ultics workflow to somebody who doesn't know Ultrics. And you can walk through
Joshua: Absolutely.
Andy: and go, I see what this does. Like it, you know, okay.
Joshua: Yeah. That's right.
Andy: Is it understandable? Like can I read it and go, makes sense? Is it repeatable? Hey, I ran the same thing this month and I ran it next month, and I'm gonna run it the month after that. And if you run it and I run it, we get the same answer. Great. And again, maybe with the exception of those probably steps. Maybe, maybe your AI suggested a different solve than mine, but it's going to a person who's gonna look at it. That might be fine, right? and then is it auditable? Can I go back at the end of the month and say, hey, we ran the month end closed? Do we all stand by what it did? And can I show it to somebody else and
Joshua: Did two plus two equal four? Yeah.
Andy: review it? Right. And so I think that mindset is important for folks that run processes. And again, this could be supply chain or this could be, you know, how I do my commerce reconciliation, like whatever it is, a Sarve's Oxley process. that business logic needs to exist. We also need to be able to tell AI where it is and how to use it. Right? We talked a little bit about kind of skills and tools. And so there's an emerging architecture for AI for those of you that are kind of digging in and building agents and things. You start to think about skills and context. and that's the other thing that that makes this feel kind of you know single player right now. Is that if you and I both boot up our own instances of Claude and you've been talking to Claude about some things and I've been talking to Claude about other things, and we ask it the same question, it has different context and it might
Joshua: That's right.
Andy: use different tools. And so what we have to be able to do with in an enterprise is be able to say, hey, if you're I don't know, pricing out a proposal for a customer, when we price out proposals, I need my agent to understand and comply with the following rules. Maybe I've got a skill that does import-export compliance. Maybe I've got a skill that does revenue recognition compliance. And it's not just that it can calculate the revenue recognition, it understands like, hey, here's how we do revenue recognition. Because when you're iterating
Joshua: Yeah.
Andy: on deal pricing, you kind of have to know all the constraints and the rules and I have to have conviction that you're sort of following our policies. And so I have to be able to sort of inflict upon AI a set of rules and guidelines that I know it will follow if I'm going to trust it. And again, if that's the case, then that, in this example, revenue recognition rule, I think has to be owned by the team that does revenue recognition. Back to our earlier point of nobody on the software engineering team wants to own a thousand Python libraries, nor do I want to have an update to our revenue recognition where I'm waiting on the IT team to implement some kind of Python code change that we both think kind of does our revenue recognition. Like I've got to be able to go do that. Hit save, update it, maybe it goes through an approval process. Now it's live. Now our agent uses our new revenue recognition rules. Like that's the world we've got to get to. And I think, again, I think we're on the verge of of doing it.
Joshua: Yeah. I I know that if we go to any leader in any company, they know that they these these things that w you're you're mentioning. You know, the the numbers have to be correct. The the processes need to be there and clear. the the team needs to be working off the same page, essentially. But still there's such a fever pitch excitement still around going after AI. And I'm sorta I I don't know if I'm as nice as you are, Andy, and I sort of say that there's this delusion towards thinking that AI is gonna solve all your problems. And I I literally I try to make it funny when I was like, you know, all Alter's history has al always been about this aha moment, right? And there's There's this first aha moment is just that moment when you realize that it the thing you're doing is, you know, Alteryx or AI is way more powerful than you ever imagined. And then, and and that's everybody has to get to that that point. And I you you got there, I got there, everybody. But the second one, I I don't know if people get to. And the second one is around this this aha moment that is a little bit more somber. It's the moment when you realize that AI can't or shouldn't. do everything, right? That there are limitations, there are capabilities that are not inherent to what we want to have happen. And it's like, okay. Like I I get it now. And I almost can talk to someone for 30 seconds and figure out where their head's at, because you know, if I if I ask you a question, you you sort of have all this the construct of like, okay, I'm going to use it here. not gonna use it here or I'm if I do use it here, I need it to have the skills and the context and the the pieces to it. Where there's some people you're like, yeah, I just throw everything in there and it it just figures it out. You know? And it it's sort of I I I it's this switch that I I just am so passionate about trying to get people to see that second aha moment that that there are limitations that we have to account for. There are there are lawsuits coming out. There's legal risk that that that is happening. There's companies that are actually losing money on this topic, right?
Andy: Well, look, I Josh, I think you're right. I think I'm a massive AI optimist, and I think what's happening and what you're describing is sort of a like a Dunning-Kruger effect, right? So, you know, Dunning-Kruger effect
Joshua: yeah.
Andy: is when you don't know much about a problem, it seems simple. Then you learn a bunch about
Joshua: That's right.
Andy: it and you realize how complicated it is, but then you've learned a bunch about it, so you get smarter about it and it sort of gets simpler
Joshua: That's right.
Andy: again, right? It's sort of a very, very simple interpretation of the Dunning-Kruger effect. And I think this is what we're all experiencing, which is if I know a lot about a topic and I go in with AI and I you know I connect a bunch of my data and I'm I I can sort of communicate with it in a way where I get pretty good results. If I don't know anything about the thing that I'm doing, I can also ask it a question. It makes a whole bunch of assumptions. I go, my gosh, amazing. Like it it calculated our entire
Joshua: It's yeah. Yeah.
Andy: revenue recognition model with one prompt. And
Joshua: Mm-hmm.
Andy: then you go, Dunning Kruger effect, when I click on that again, surviving first impact. If I took that to my revenue
Joshua: Choo That's right.
Andy: recognition team, they would go like. It's not wrong. It's just not how we calculate revenue, right? It's this is artisanal.
Joshua: What it w Yeah.
Andy: Like it just came up with a random way to do it and that's not helpful inside
Joshua: Hundred percent.
Andy: a company. And so I think we're going through that at scale, both
Joshua: Think you're right.
Andy: individually and as companies. What I again think will happen is we will realize how complicated our businesses are and how do
Joshua: Mm-hmm.
Andy: we represent that in AI? We will need tooling to
Joshua: Yeah.
Andy: do it. We'll go through that. my gosh, this is harder than we thought it was. And then as we figure out, actually, if we empower all the people that know those things, we give them the ability to get that logic codified in a way and we put it into some skills that can be shared across the organization, we'll sort of come out the other side. Like, actually now when I talk to AI, I don't have to understand our Import export rule details, they don't have to understand Reverec. I can just talk to our pricing bot and it comes up with really good pricing. And when I show it to teams, I'll go, yeah, that looks really good. And it gets simple again, right? But we've got to go through that. Like that's that's just where we are
Joshua: Yeah, that works. Yeah, yeah. Yeah. Yeah.
Andy: right now. And and this is, I joke, this is why CEOs think AI is amazing. We are the masters of the Dunning Kruger effect, where we're up here at the top. We don't understand all the complexity. Look how easy this was. I just I don't know why it's taken
Joshua: Yeah yeah. It's like it's all amazing. Yeah.
Andy: you guys so long to calculate this thing. I just asked Claude and it did it for me, but it's like, yeah, but Claude
Joshua: Yeah.
Andy: didn't. follow any of the company's policies, didn't understand the complexity of our
Joshua: Ha ha ha.
Andy: contracts and our deals, and it made a million assumptions that made this easy and simple.
Joshua: Yeah.
Andy: When I learn more about all the complicated aspects of the business, the answer isn't don't use AI anymore. The answer is,
Joshua: Right.
Andy: hey, how do I empower my teams in a multiplayer way to encode that logic, make it Vora,
Joshua: So good. Yeah.
Andy: make it shareable, wire it back into the AI and come out the other side, and that's the journey we're
Joshua: Yeah.
Andy: all on.
Joshua: I I don't know if it's just CEOs though, Andy. I I I I love our GTM folks across the board too. They're they're probably right there with you, don't you think? Like the GTM you right? Yeah, yeah.
Andy: I think it's everybody it's it's it is everybody with with I just I laugh because CEOs are like, my gosh, AI is gonna replace you know all this capability in my company. It's like, well, I think you got a ways to go before it really understands
Joshua: Yeah.
Andy: your business. it doesn't mean it won't shift how we staff things. You know, I'm a big believer in
Joshua: Yeah, right.
Andy: the theory of constraints, and I think AI is gonna massively change the constraints in every part of our business. We're gonna have sort of this abundance of new intelligence. We're gonna have to implement all kinds of new business logic to support that intelligence. And so I think the makeup of our teams will change a lot. It doesn't necessarily mean it's job destructing. It just means, hey, if one of my teams is five times more productive, but they're constrained by a team that's only twice as productive, I'm going to reallocate the staffing levels of my teams and hopefully grow a lot faster and have a more efficient business because of AI. So I think all that sort of happens. But that's the transition
Joshua: It's all positive, yeah.
Andy: companies have to go through. I think individually we're all going through this dunning Kruger. It's not related to CEOs.
Joshua: Percent.
Andy: Every one of us has this. Hey, I didn't know a lot about this. I asked AI. The results are amazing. Hey, I know a lot about this. I asked AI. And it's like I really had to refine what it was doing. It wasn't that it was wrong or not worthwhile. I just knew how to ask the right follow-up questions. actually, here, let me bring in little bit of extra data. Let me make that make a little bit more sense. Let me r refine your thinking here. now that's really useful. Now it's really useful, but now I want to send it to you. You don't have all that context I provided. You sort of go like this thing doesn't make any sense to me. So again, a lot of work to do to make all that sort of wired together and and make sense.
Joshua: Yeah. No, it's I I like these sort of being able to apply the advanced thinking that's taken fifty, hundred years to to get to because it does help us sort of make the ground not feel like quicksand sometimes 'cause it it's, you know, there's we're all in the same boat. We're all learning what's possible. if we keep going, we're gonna end up talking for another two hours. So I'm I'm hoping you would be up for what what I call rapid fire. I I have this little tidbit towards the end where I like to throw stuff at you and get your sort of off the cuff gut reaction and and quick quick thought. You up for it?
Andy: Sounds slightly terrifying, but let's do it.
Joshua: Let's go, let's go. I want no personal questions. I won't ask about finances. This is this is all socks compliance. so
Andy: It's a safe space. Okay, great. Let's go.
Joshua: yes, I I have been called a therapist in pretty in other podcasts, but first one, personal AI freedom or enterprise AI guardrails.
Andy: I want to say both. I'd go freedom first. I think we're still in the experimentation phase.
Joshua: Okay. Ooh, good one. All right. Directional insight or auditable answer?
Andy: I think we need to get to auditable answer. I think we're currently living in directional insight.
Joshua: Perfect. Single player speed or multiplayer trust. Sort of
Andy: I think we gotta get to multiplayer trust. I think this is the whole we all go further together analogy. if we really wanna
Joshua: Yeah, yeah, yeah.
Andy: get to fast AI in a company, we've gotta get to multiplayer a lot a lot
Joshua: Yeah.
Andy: sooner.
Joshua: But but there is value to that single player speed, you know, peop figure it yep.
Andy: Yeah, yeah. No, absolutely. We but we have that now. So like I I think if you if we go to like why is B to B AI not just taken off? Why are all of us not every single day talking about all the processes that we have agents for? Like we gotta
Joshua: Yeah.
Andy: get to multiplayer.
Joshua: Love it. All right. One assumption that AI should be allowed to make.
Andy: I think it I think it's lots, frankly. I just need to know which assumptions it's making. So one of the things I would love if
Joshua: Mm-hmm.
Andy: AI did a better job was saying, hey, like I assumed these things when I did this. but
Joshua: Yeah, yeah.
Andy: I think that's part of the magic is, you know, hey, I'm trying to solve a problem. I get that it's complicated, make some reasonable assumptions and let's move forward. I I think many, but I need documented.
Joshua: It it's interesting when I when I just did a quick rabbit hole on this, when I've gone to you know, multiple steps down a AI workflow and have got to an end result I didn't want or didn't expect, I I would sort of back channel, ask the hey, how did you get here? What were the assumptions you make? And Good lord if it didn't say well, six months ago you you and I had this conversation about XYZ and I assume that that's still in place. And I'm like, the stuff that you like is keeps me on my toes. Like, actually you're right. Like my assumption was wrong.
Andy: Well, I'll give I'll give you I'll give you a very concrete example. I built a little agent, again, just playing around with this stuff. I built a little fitness agent. and I wrote this one up as well, I'll share at some point. and what I like about it is I can ask it to plan a workout for me. It knows like what equipment I have at my gym and things like that. but I wanted to do more of like food and diet tracking and I hate those tracker apps where I have to like enter every little thing.
Joshua: I know, three hundred and twenty four calories. Yeah.
Andy: Well, or like, you know, I went to a restaurant that's not in the app and I had a BLT and now I'm like,
Joshua: yeah.
Andy: do am I assembling like one piece of lettuce with like two slices of tomato? You know, it's like, was it a Roma tomato? I'm like, I don't know. Like it was it's it was just a tomato. Right. And
Joshua: Yeah. So so painful. Yeah yeah.
Andy: so the agent I wrote at home runs on my local machine, and so it's all private and great. But I basically say to like I had a BLT for lunch. And it goes, Okay. And it goes and makes some reasonable assumptions of like what the hell's on a BLT? Great.
Joshua: Yeah.
Andy: And do I really
Joshua: Yeah.
Andy: care if it's off by 50 calories? Not in the slightest. Like, great. Like it's a BLT. So that's the kind of assumption that is really useful.
Joshua: Let's go. Yeah.
Andy: And you realize like good software should be able to make all kinds of assumptions like that. That's useful, right? I'm
Joshua: They're reasonable. Yep. And
Andy: also not running a restaurant and ordering my monthly supply of lettuce based on this assumption, which if it was off by 10%,
Joshua: Exactly. That's right.
Andy: might be a really big deal. So again, knowing
Joshua: Yeah yeah yeah.
Andy: where you're using assumptions and where you're not.
Joshua: Yeah.
Andy: Is sort of the power of AI.
Joshua: Yeah. Well and it's a it's a power of a a leader too to understand. I I like it. one enterprise AI metric leaders should stop celebrating.
Andy: I think it's token usage, but I think we're getting here as an industry. People are realizing, number one, it doesn't necessarily drive value. Number two,
Joshua: Totally
Andy: we are often paying people to each independently run the same calculations and get different answers and paying the AI companies for the privilege. So a lot to be done there.
Joshua: Yeah yeah. Amen to that. that's a that's a that hurts me just thinking about it. all right, last one. These haven't been too bad, right? Like
Andy: All good.
Joshua: one hands on AI project that most changed the way you lead.
Andy: I think it was the first time I wired an agent into both Ultrix and Snowflake, and I could start doing my own deep dive analysis into the data and basically using it like a scratch pad to figure stuff out and then publishing it into Ultrix. And now I'm able to go in, look at my business in detail, but actually share out what I'm doing. And that's been really fun.
Joshua: So good. So good. amazing. All right. So we gotta wrap up. Mr. Andy, can you leave people with one thought. Like what's on front of mind? What's what's something that you you want everybody to to leave the podcast with today?
Andy: I mean, one, I think everybody should be playing around with AI and and getting into not just the web interfaces, but like try these C L I's, these command line interfaces and starting to really learn how to
Joshua: Yeah.
Andy: build things like on your machine and then immediately start to transition like, okay, if I can build stuff like this, how am I gonna do that across my team? and so I think that's a a big exciting new area for people to to get hands on and think about how to how to scale building.
Joshua: Absolutely. And then last thing is we I think most people could figure out that you're on LinkedIn. Is there any other places that people can reach out to you or contact you if they had a question for ya?
Andy: I'd say LinkedIn and X are the two big ones for me. my handle
Joshua: Mm-hmm.
Andy: on both of them is just A P McMillan. P is in Paul. So A P. McMillan, you can find me on either place. happy to
Joshua: Cool. Perfect.
Andy: engage with folks.
Joshua: Love it. Andy, thank you. Sincerely I appreciate it. I love all our interactions and a whole lot to digest today for sure in an hour. So thanks again. Appreciate ya being on.
Andy: It was really fun. Thank you.
Joshua: Alright, take care.
Joshua Burkhow: You know what stayed with me from this conversation is that your knowledge of the business still matters. AI needs the rules that your team understands and the judgment that you've built through experience. Andy gave us a useful way to think about it. AI can suggest why something happened, but the rules behind a business decision need to be clear enough for people to check and use over and over again. So here's something you could maybe try this week. Take one of these AI experiments that helped you, show it to a teammate. Can you see the assumptions? Can you understand the steps? And more importantly, can they use it without having to start all over? That's a practical step towards making AI useful across the team. I want to thank Andy McMillan for joining me. Thanks to you all for listening to Ultra Everything. You can listen to it on any podcast that shows alter everything. I'm Joshua Burkow. See you next time.
Joshua: Hello everybody. Sometimes I get guests that I've been longing to have on the podcast for a long time. sometimes I schedules work out and it just happens right away. But this is one that I've been wanting on for a long time. Th this person has had an immense impact on me, and I I wanna get into a really meaty conversation about AI today, with none other than our own Alteric CEO, Andy McMillan. Andy, welcome to the podcast.
Andy: Thanks for having me on and that's a very kind intro of you. I'm I'm not a tough guest to get when you're running the Alteryx podcast, and I'm willing to come on and chat with you and I've been looking forward to this as well.
Joshua: I love it. I love it so much. Before we get into the meat of of talking about all this stuff going on in the world with with AI, we have some things to talk about for sure. I I'd like to have people understand how you got there. I don't think you have a prototypical CEO path that you know every every other CEO has had. And I I find it really interesting. That I I'd love for you to share if you you don't mind.
Andy: Well, I I don't know, I mean, maybe there's I don't know if there's a prototypical path or not. That's a fair point. I do think a lot of folks come up through the finance side of the world. So you do see a lot of CFOs going into CEO roles. you see a lot of people in kind of COO or operational roles as well. know, occasionally some folks out of sales backgrounds. and I think now what we are seeing more and more often, especially with this massive sort of technology shift happening around AI, especially in software companies, as people sort of transform what they're doing. We are seeing people with my kind of background, which is a product background, starting to take on more CEO roles, which I think is interesting because we think of founders often as having that kind of product background. but now I think we're seeing more professional CEOs coming out of that area. my background is is not only out of the kind of product management side of the world, actually forever ago, I actually started my career as a As a software developer and I I joke that I was so good at writing code
Joshua: Amazing.
Andy: that people introduced me to product management and thought maybe just come up with the ideas and let somebody else build them. but yeah, I very much come
Joshua: Right, right.
Andy: from the technical side of the universe.
Joshua: Yeah. C can you sort of highlight why you think that helps you? Or what gaps does it fill in having this sort of knowledge and experience come to the table for for at least for all turks?
Andy: I mean, I think there's certainly some value, perhaps obvious value in really deeply understanding the thing that your company builds or does. So if you're I don't know, running an auto company, having a background building cars is probably helpful. we're a software company, we build software for our customers, and so I think having a background in building software is useful. I do think companies go through points in time where sometimes the building of the thing isn't the core problem of the company. You're scaling distribution, managing the financial structure of the company. And and sometimes when companies go through those periods of time, I can understand why someone with a a background in that part of the business ends up taking the leadership role of the company overall. I think that's frankly the position a lot of software companies were in, even three and four years ago. There was a lot of focus on, hey, founders have come up with really great ideas, especially in like the B2B software side of the world. How do you scale that? Like, how do you get that out to the world? How do you ramp up sales team and a marketing effort and customer success and all those things? And so that's really valuable. I don't mean to diminish people that come from those backgrounds running companies. I think now there is this huge shift going on where every software company is kind of looking at themselves in the mirror and saying, okay, like why is our software valuable in a kind of post-AI world? And that's not a distribution question. I think that's a core product. question. And so I think that's why the rotation of sort of product centric folks at the helm is happening right now.
Joshua: Yeah, it's interesting because one of the things that I've observed here, is your perspective and your energy and your drive towards innovation. You And I I I think maybe that's just part of the the understanding of how product innovation works, but also, you know, maybe maybe a really valuable asset in the world that we're like in the thick of, the chaos of innovation. Does that make sense?
Andy: It makes perfect sense. I think one of the things about CEOs right now, whether they come from a product background or not, is you better be pretty curious about how this massive technology shift is going to impact your business. I think part of being curious, at least for me, is getting my hands on stuff. And so, you know, I've got my work laptop loaded up with a million tools that I'm using and integrating with Altrix and building stuff. I've got my home set up. wired up and I'm spending some of my evenings building stuff and thinking about how all this comes together. And I genuinely enjoy that. I it is something I've always liked building stuff. I've always been interested in software. It's again how I started my career as a software developer was I thought it was fun to build stuff with software. So I think having that core interest helps because you're trying to think about, hey, how is this new technology going to impact my customers and how they use my products and services? And so understanding that technology is is useful. And at least for me, the way to understand something is not reading the manual. It's installing it on my machine and trying to do stuff with it and and sort of iterating
Joshua: Yeah. Do it. Yep.
Andy: and and you know, playing with it a bit.
Joshua: So w one of the things I wanna shift gears towards is is this first concept and idea. I I really have been wanting to pick your brain on this topic. you and I have talked a lot about sort of personal AI ventures and all the stuff we're learning and and learning how to do hands-on keyboard. But then w we come to work, we put our you know, our work hat on and we we sort of say, Okay, how do we make this work for our customers. Like how do we make Alteryx and AI tangible and and practical for enterprise AI? And I'm wondering if you have just general themes and then I'm gonna poke and broad on a on a couple questions I have.
Andy: Well, I think maybe one of the big framings is talking about who our customers are is an important part of that conversation. we have a customer base that I would say is technical. You know, they understand data very well, right? So if you're a tax accountant, you're pretty technically adept at how data works. if you're in sales ops or supply chain management, you know data and logic behind data really well. What our customer base often is not is software developers, right? And and that
Joshua: Right.
Andy: is a different sort of landscape of technology. And so one of the things that I think we have done for a long time really well is help make people that are sort of data technical able to use software to do incredible things with their data without having to double click on everything and get into writing Python or SQL and starting to feel like they're a software developer. instead of, to use my example, sort of a tax accountant. And one of the things that I think we spend too much time in AR and not talking about is this idea that somehow software developers or forward deployed engineers are gonna write agents that are tax accountants when the folks writing that stuff don't really know a ton about tax accounting. And yes, you can ask the AI to go do some research on tax accounting for you, but I think that's different than being somebody who deeply understands not only tax accounting, but how the taxi accounting works inside your company and and how it works across your different divisions and geos and decisions you've made. And so I believe one of the opportunities with AI, much like we did with data tooling like Python and SQL, how do we make that accessible to somebody who has the kind of technical competence in their area and their domain of expertise that's not software development, but is data technical for them to consume. And so I'm spending a lot of time and effort thinking through how do we enable people like that to be wildly successful using AI to solve problems and to think about how it impacts their role and their job and their scale. And I'll say that group of folks from the moment I took this job have told me what they love about Altrix is it helps them automate things and scale because there is, in their words, a never ending stack of work, business changes and requests on their plate. So how
Joshua: Absolutely. Yep.
Andy: do we help those folks use AI to be more effective, more efficient and sort of scale that across their company, I think is a really interesting opportunity space for us to sort of lean in and listen and help them.
Joshua: you and I have talked to a lot of customers and this is one of the things that we don't do that by giving them two thousand lines of Python or a bunch of SQL to to have to read through every single time to to be as productive as they can. And I always come back to this. I I think it's such a fundamental concept that opens up all the other doors around AI is like, hey, do is there an opportunity to make it simple and easy? It does it have to actually be rocket science. W we can make all the, you know, AI agents in the world and do all these incredible things. Cool. Amazing. But at the end of the day, like not not everybody to your point is is a a fifteen year data engineer. Well, you know, they're a you know, an accountant. They they are very good at what they do. Can we meet them where they are? And I I I find it really fascinating. And and so this is we led right to where my questions were gonna go is what what is different about, you know, you and I are at home, we're we're in cloud or codex or maybe an open source model and we're we're building stuff and that's great. It's all sort of personal endeavors. When we come into the enterprise AI world, what changes? Like what what is it about enterprise AI that you're really trying to get through to other CEOs and other leaders across companies?
Andy: I think this is one of the core challenges we're all dealing with is I would say AI seems particularly useful in in two areas. One is personal productivity. So, you know, again, clean up this email, you know, help me think through this thing, help me search for this, all really useful. and then the second is when we build things for ourselves, all the sort of assumptions that AI makes in the background feels like magic. So for example, I built a little project at home playing with data to sort of go through personal finances, right? So we all deal with this. You get sent all these statements. I think banks do a remarkably good job of making it hard for you to get to the actual data. So you get, you know,
Joshua: Just a bit.
Andy: I still get literally, you know, paper copies sent to me or PDFs. And I thought this would be an interesting project. What if I load all this into an agent? And so I feed the PDFs into an agent. I'm using a little local open source models for not sending all my financial data out to the internet and it rips through all the PDFs and it puts it all into a nice table for me and I can I can now pull all this apart. And I ask this little agent, which was really fun to build, and I'm writing some posts about how I built all this stuff that I'll share at some point. But
Joshua: Good, good.
Andy: I'm asking it questions about what I spend money on. And you know it comes back and it's like, you know, your biggest spending area are these categories. And it made up these categories. And it's super useful. And a few weeks later I asked it another question, it sort of goes back and it recalculates some categories and it tells me what I'm doing. At home, I'm not really hung up on the specifics of did it calculate those categories the same way each time? I'm not, you know, managing my family on financial performance metrics based on how they hit these categories, thank God for my children, right?
Joshua: Right, right. I
Andy: My mother used to tell my dad he can't run our family like a business unit. So maybe there's a lesson in there that I'm I'm trying to think through. But but I think there's
Joshua: You're trying to learn, yeah.
Andy: we literally had to make cash flow statements for college, like month by month of what my father was providing and what we were doing. So it's a story for another day. But
Joshua: It's amazing. I love this guy, huh? Yeah.
Andy: but I think look, you know, those those assumptions are what make AI feel like magic. What I didn't have to do was all this nitty gritty work of going through line item by line item and trying to categorize all my spending. So that's awesome.
Joshua: Mm. Yeah.
Andy: If I take that same concept and I bring it into
Joshua: Exactly.
Andy: work and I go, hey, every time I run the accounting process in the background, this AI makes a bunch of, again, reasonable assumptions, just different
Joshua: Reasonable.
Andy: than last time I ran it. I think we called changing the accounting
Joshua: Oops.
Andy: rules every month. I think those are technically financial crimes, right? So you're you're sort of so
Joshua: Pretty close.
Andy: so that what sort of makes it awesome when I'm not living in that finite business space makes it hard to roll out. And I think it it also manifests itself in how we interact with things at work. If I go into my favorite, you know, LLM environment and I build an agent and it pulls together, I did this the other day. I had it look at all of our web traffic. I just plug
Joshua: Mm-hmm.
Andy: it right into, I've plugged into Altrix, I've plugged into Snowflake, and I'm analyzing our web traffic and looking for some patterns. And it comes back with this whole analysis. It's really interesting. But I know the minute I show that to our web marketing team, we're not going to go to the conclusions. We're going to go like, how did it calculate that? Where did that come from? Why is it making that assumption? Like it will not survive first impact on a conversation
Joshua: That's right.
Andy: with other people because none of it is visible to them. They're not going to understand what went on behind the scenes to pull this data together. If they went in and asked their own Claude or Gemini or Codex, the same question, they might get a similar but slightly different answer. So it's not repeatable. So
Joshua: Yeah, not repeat or yep.
Andy: how do I compensate or track or measure my web marketing team if every time I run the report, the numbers change a little bit how we calculated it? but that those set of rules and that consistency is how businesses operate. I can't come to you and go, hey Josh, we're gonna put together your KPIs for the quarter. I'm gonna tie your compensation to those KPIs. Neither you or I are exactly sure how those are gonna get calculated, but at the end of the quarter,
Joshua: Right.
Andy: we'll ask the robots and they'll tell us how they think you did. And then I'll pay you based on that. You're gonna go, hey Andy, I don't feel super comfortable with this arrangement, right? And
Joshua: Yeah, yeah. Good luck. Yeah, exactly.
Andy: and so what was magic in a single player game of me working with some data that sort of gave me directional usefulness is the major. hurdle for adoption internally. And we see it time and time again.
Joshua: Yeah. you're sort of hinting at this this idea of the single player trap. And I want you to sort of h help folks understand that more because I I feel like there's still a lot to unpack in that in that concept.
Andy: Well, I would say this is where maybe the agentic coding is different than how the normal folks are using AI. So if I'm on an engineering team and we're using, let's say, Claude code to build something, I've got an infrastructure behind me that Claude's been taught to use that can sort of make this multiplayer. I can have Claude Do what's called a pull request where it goes out to a Git repository, which is where all my code is sitting, and it can pull in some code that somebody else wrote. And I can have Claude make some changes to it. And then I can say, Okay, Claude, now I want you to submit this as a merge request, which is sort of to tell the person that owns the code I made some changes. And you know, their agent can look at the code and maybe merge that code in. And so I'm sort of working on this shared asset in a way that makes sense. And so I think in that universe, that sort of works. As soon as it gets outside of that universe, I get to I'm using the web interface for Chat GPT and I'm asking it questions. and then it goes off and maybe it writes a little bit of code in a little sandbox to figure something out, like putting, I don't know, two spreadsheets together for some logic. And I go, that was really interesting. And then I go over my CFO, John. I'm like, Hey John, look at this thing. John's gonna go, Well, how can you send me the thing that built that? And I'm gonna go, no, I don't like it's it's somewhere behind this web interface. I don't I don't You you could try to tell your chat GPT to do the same thing, but again, it'll might do it a little differently. So your your mileage may vary, as they say. even if I'm in something like a CLI. So let's say I'm a little bit more technical, and I'm I'm using, I don't know, anti-gravity or Gemini or something like on my computer and I'm in the command line interface and I'm building something, and it's writing a bunch of Python for me in the background. Again, same problem. When I go to my CFO and I say, hey, like I built this really cool thing and it's recategorizing some stuff that we're doing. What am I doing? Am I sending him some saved Python files? Better yet, am I saying to my CFO, hey, what's your
Joshua: Screenshots. Let's go with
Andy: what's your git handle? I'll I'll put this in a git repository for you. I mean like, what are we what are we doing here? What are we doing, right? So so that's what I mean by it's
Joshua: Can we just do this? I just wanna see John's face. be priceless, but
Andy: it's kind of single player, right? Like I I'm I'm
Joshua: Right.
Andy: doing all this cool stuff. My my team's doing cool stuff. probably everybody listening has this maybe I'm in the in their story, the the kind of pointy hair Dilbert boss where it's like I'm out using AI and
Joshua: Yeah.
Andy: I come up with some cool thing on a weekend and I send it out to the whole team. And everybody's like,
Joshua: Let's go for it. Yeah, yeah.
Andy: great, like you know, the boss is doing random stuff in AI that none of us can repeat and we don't understand it and
Joshua: He's going Yeah.
Andy: there's probably flaws in the logic and none of us can dig into the logic, but he's sending us reports where he thinks the whatever the email marketing campaign didn't go well. Like not
Joshua: That's right.
Andy: helpful. And that's the sort of again single player. Like none of it's operating off real repeatable and understandable assets where the marketing team could go, I'm looking at what he did and I, you know, he's using the data assets that we created that make sense, that actually represent our email marketing campaigns, the way they're actually labeled, like whatever that is, we haven't all got there yet with AI.
Joshua: Yeah, I it it's so fascinating to me because you go into a customer and they have this natural pent-up excitement around AI. So do we. It's it's amazing, there's a lot of cool stuff in it. But to your point, it's all this sort of individual personal productivity or projects that they've been able to make progress on. But there's almost this sort of assumption, I guess, is is that this just automatically scales. Like it just automatically becomes this thing that everybody can tap into and all share the same sort of intelligence. But reality's not that. Like reality is that you're you're teaching your models and your your agentic systems on all the things that are important to you. But y I would never take that and use it to do do my own thing. You know, I I this is one where I I think a a lot of customers Got excited, got into it, did the POCs, how many reports were written by industry saying, you know, so so few pilots are getting out of their sort of incubation phase into production? I I think this is part of the that challenge that you're highlighting. You think? You agree with that?
Andy: Well, and it's I I agree with that. I also think it's the natural evolution we see of new technologies. You know, this is the whole, you know, history doesn't
Joshua: Yeah. Across the board. Yeah.
Andy: repeat itself but it but it certainly rhymes. if I if I use an analogy to another time, if I think of the early days of the internet, back into the early days of my career. So I was my first job was actually as a web developer, working
Joshua: Yeah, you just dated yourself, right?
Andy: at EDS. I I helped build the original websites for General Motors. So, like gm.com and Pontiac.com. Yeah, yeah.
Joshua: Wow. See, that's a great story. Ugh, so good.
Andy: it actually a related story, you'll be amazed how long code
Joshua: Please.
Andy: lives. The little window sticker that you used to see when you would price a vehicle, the code
Joshua: Yeah.
Andy: I wrote that built that was live for like 15 years. They just retired it. For for years I would go online and be my I can't believe that's still running. so so anyhow, it it in the early
Joshua: Yeah, we gotta call them, put it back. That's awesome.
Andy: days, there was this big issue that. Building a website was actually very technical, right? You you hand wrote a lot of code
Joshua: Hm. Right.
Andy: and you hand wrote a lot of JavaScript and HTML and all these things that today I wouldn't say are extremely complicated coding environments, but nonetheless, they're coding environments. And so you had marketing teams and commerce teams that wanted to build web applications, but they weren't technical. They were marketers and e-commerce and product manager experts for product lines. And so they would hire consultants and agencies and people to come in and build their site. Their site wasn't editable really because they didn't have any access and they were, you know, it was really expensive and it never quite did what they wanted it to do. And we had all this friction. and everybody was like, basically the internet is a playground for techies, you it's for people that that write software. And, you know, we as an industry eventually fixed that. We came up with these web content management platforms. And what the web content management platforms basically did was acknowledge that a lot of this is really technical. So the security, the web applications, the single sign-on, the e-commerce engines, like all those things are going to be managed by IT teams, technical teams. They're gonna it's gonna take real software. But we also can create spaces in that real software where people can write content and post images and things that aren't technical. We didn't go out and teach everybody in marketing to hand code HTML. We taught them how to use web content management platforms. And we put in place workflows and approvals, right? Everybody in my marketing team can't just click a button and publish stuff out to the internet on our website. There's an approval process and and we have staging environments and all these things that are in place that make it enterprise class, business led, supported by IT, you know, technically secure. So I think we're gonna go through that same thing with AI, which is we're gonna Go through this phase where, you know, phase one is people that want to write Python and play the single player game and have a Git repository
Joshua: Turkeys. Yep.
Andy: are gonna build stuff. And pretty soon, and we see it happening already, people that work in normal jobs in you know finance and in supply chain and and sales operations are gonna say, I need all those things, but I really don't need a Git account and I don't plan to learn to write Python for a living. Not only that, if it can write the Python for me, I still need to know what it does. Right. I I can't I can't bet my job that this thing I can't even read is the right answer. especially I'm in finance. I can't stand in front of my auditor and go, I don't know, like I've got two thousand lines of Python here. Like, yeah, it seems like it kind of works. Like, I'm gonna want some
Joshua: Think it's right. It's probably
Andy: level of understanding and I'm gonna need that shareability,
Joshua: That's right.
Andy: right? It's it's not just, each tax accounting rule was written by one person and only they have access to it, and then they, you know, so like We're going to come up with a way like like a repository to sort of share all these things and build all these things and wire it into the business. So I think all that will happen. And I think when that happens, that's when I think enterprise adoption can really take off. That's when we can say, like I have a I have a you know a workspace full of of skills and logic that the team has built, that the experts know is correct, that I can point my AI agents to go use. So if my AI Agent needs to know about revenue recognition, it it goes and looks up our revenue recognition calculation and actually uses
Joshua: Always knows, yep. Yep.
Andy: it. What it doesn't do is make a bunch of assumptions of like, like here's a reasonable way to calculate revenue recognition and then make a bunch of decisions. It it needs to be constrained to how we actually do it. And so I think we will get there. I think we're helping our customers get there right now. And I think that's the big unlock for real enterprise AI is to get beyond this, you know, single player techie. mode, which we're in now, which is again not uncommon, and we get to a multiplayer business user friendly way to apply the technology, and then I think, you know, things take off.
Joshua: Yeah. there's two two points. Your your example when when I hate even bringing this up 'cause it dates me as well, but I got crazy when JavaScript first came out. Because it was I was knee deep in HTML. I thought it was the coolest thing. and then when JavaScript came out and stuff could move, I was like, this is incredible. Like life life will never be the same. Sorta right, but I I did realize that you you hit on this point that as soon as WordPress came out, right? Talking about CMS, WordPress came out and it gave you UI and it gave you all these buttons and clicks and stuff. And behind the scenes, it's handling the HTML, the JavaScript. And if you you look forward, like we've seen this Tableau did this, right? Having to write p Python to to to write all this code to create visuals. Guess what? Tableau came in. It's like, yeah, just drag and drop these pills and you get this, you know, beautiful state of the art imagery and and and you know charts and graphs and all this stuff. And it so to your point, like, history repeats itself and and it's one of those things where it it it's due for it. I I find this hilarious. I was at a a customer on site not too long ago and One guy says to the other, like, I was on Claude last night. I was doing the amazing things. I I I got all this stuff. It did everything that I needed to do for my meeting this morning. And you you could just tell he was ecstatic about it. He's like, just like clockwork. He was like, Yeah, but it's the only problem is it it produced it all in Python. I don't know Python. I was like, I don't know what to do with that. Do I do I save that? And I was just I I I w it was struggling because I wanted to come up to him and and chat about it, but I also like, yeah, he he'll he'll figure it out. But if I know if one person's having that experience, thousands, millions, hundreds of thousands are are having this experience as well, is where they're they're right on that cusp. And I think the the underlying current behind all that is when a problem starts to surface across the board, it's gotta, you know, it's gotta get solved. They w sort of call it the poppy seed. syndrome where it's like you watch poppy seeds. Once there's enough up there, you it time to cut off and solve that problem.
Andy: I think with a lot of this tooling, one of the challenges is that for the technical folks that build a lot of the tooling, it takes a little while for the need to emerge because they're inherently technical. So you don't see the need right away. I get asked sometimes
Joshua: Yeah. A hundred percent.
Andy: by data engineers and Python developers, you know, what does Altrix that I can't do in Python or SQL? And I'll tell them not I mean,
Joshua: Every day.
Andy: you can you can write whatever you want in Python. The point is there's this massive group of people that don't have the Python skills that you have, that understand data really well, that need tooling that helps them solve problems. So the goal of of Altrix or kind of business user solutions is not to create things you can do in that environment that you can't do a level or two lower in the stack if you want to write lower in the stack. The point is a lot of people don't want to make their living Learning how to write lower in the stack. And even when these agentic tools help you do that, I still have to understand what it's doing. I use this analogy all the time with customers, which is with enough context on topic, and if it's a simple enough conversation, I can sort of understand what's going on when people speak Spanish. Took some Spanish in high school, it's been a long time. Okay, but but but again, a lot.
Joshua: Yo hablo español, ¿sí?
Andy: has to be in context. there
Joshua: Great.
Andy: is absolutely no way I would sign a contract of something important in Spanish. I don't I don't know enough Spanish, right? That that would be crazy.
Joshua: Perfect point. Y Amen. Right, right. Yeah.
Andy: and maybe I could hire a Spanish interpreter to sit next to me and tell me I was going to a very efficient way to go through a contract, right? Which is
Joshua: Yeah, what's this word?
Andy: Which is what we do when we say, well, bring in a forward deployed engineer or have a software developer sit with the tax accounting team.
Joshua: So to
Andy: And every time they want to make a change, they can tell the software engineer who can then code the change and maybe there's some lost in translations to take the translation analogy a little bit too far. or a better
Joshua: You can do the translation. Yeah, yeah.
Andy: answer is like maybe the contract's in English, right? And and and I that works for me. So same
Joshua: Yeah. Yeah.
Andy: situation here, right? The the goal is not to say there are things you can do in low code, no code environments. That you can't do in code-heavy environments. The point is to say, who are you empowering to do those things? And what level of trust and understanding do they have in what's being built? And I don't think the goal of AI is to make sophisticated, important things like your tax accounting completely black box. You're not going to bring in your auditor and go, we have no idea, but we think the robots are running a pretty good tax accounting regime that we can't defend. what I want is help me build a more efficient and effective and AI empowered tax accounting regime in an environment that I understand as a tax account, because I have to put my name next to it and say this is correct. And I think there's a lot being built right now that's gonna help people do that soon. And again, it's not gonna be requiring them to understand Python or getting a Git repository or understanding how to do pull requests and merge requests. Like that's not where the world of tax accounting goes. It's to take those same concepts and put them into the environment where those folks are working, and to have Claude or Gemini or whatever help them be a better tax accountant, help them maybe write an agent or build an agent that does tax accounting with them. Like again, I'm I'm not saying they're not going to agentically build these things, but I think this whole idea that you need a, you know, a quote unquote forward deployed engineer who's not a tax accountant to show up and try to build your tax accounting agent to me is completely backwards. Right. You already have a forward deployed employee. You have a tax accountant right there. They know your tax accounting rules. How do we give them the the
Joshua: Yep. Empower them, yeah.
Andy: tools and the skills through AI that they can build stuff where they go, yeah, this is our tax accounting rule. This makes sense. I will stand behind this and this thing can run in an autonomous way. I can manage what it's doing. I can have it work across my other tax accounting rules. I can know that when other agents do things that impact our tax accounting, that this logic is going to be part of that agent's decision. Like that's a very empowering role to be in. And it really drives up
Joshua: That's right.
Andy: your value as a tax accountant. But all of that is sort of premised on the idea that you're probably not trying to learn how to be a software developer. And I really like your analogy of the HTML to JavaScript jump because I remember being in that world
Joshua: Yeah.
Andy: and the idea was. We're just gonna teach everybody in marketing to write HTML. And HTML is not that complicated. So people were
Joshua: Right. No.
Andy: learning. And right about the time people in marketing were like, yeah, I think I've got this figured out. We introduced JavaScript. And they were like, you gotta be kidding me. Like, this is like real coding. Like this is, you know, this is hard. And so that's when everybody's
Joshua: Wait, functions. yeah.
Andy: like, this is not, I didn't want to be a software developer. Like I could have gone to school
Joshua: Yeah, yeah.
Andy: to be a software developer. I went to school to be a marketer. Right. And
Joshua: That's right.
Andy: so you're right. Then we figured out as technology folks, hey, people
Joshua: Yeah, yeah.
Andy: don't want to write JavaScript by hand. They want to use WordPress. They want to use things like that,
Joshua: That's right. Yeah.
Andy: right? Doesn't mean JavaScript went away. Doesn't mean nobody writes anything in JavaScript. Nobody made an argument that you can do stuff in WordPress that you simply can't do in JavaScript. It's like, no, but the point is, who's using WordPress? And so that
Joshua: It's just absolutely yeah.
Andy: that inflection point is on the horizon here for AI. And
Joshua: Yeah, exactly. Yep.
Andy: that's when it goes from being single player techie tool to being multiplayer business tool. where the people that understand the business are building things with AI, scaling things with AI, standing behind the things that they built in AI. I think that's the sort of exciting frontier that we're just coming up to now.
Joshua: Yeah. All right. I wanna shift a little bit and just talk to software development because it's so front and forward. To all the stuff that's going in AI and the fact that you have a very close background to this this field. what like my message always to to software developers is like, are you working on the most valuable stuff? Because every software developer I've run into, have a few high value projects, one or two. But the vast majority is just the the the translation, the speak trying to speak Spanish in an English world. Like if software developers would get a lot of the tools and functionality that they don't need to work on out to the masses, let them use those tools, Alteryx being one. And then they focus on high value things that literally change the foundation of what is possible. Is am I do I have this right or do you think of it differently?
Andy: Of it as there is an incredible amount of mundane work that happens in software engineering today, whether you're talking about
Joshua: That's right.
Andy: the kind of work that you're describing, which is, hey, I built an entire application for one of our business units and you know helps them solve a bunch of problems, but now they are constantly coming to me with small changes that they want to make to that application, and I'm spending my life maintaining little changes. So
Joshua: Yeah, exactly
Andy: How do I
Joshua: right.
Andy: say, hey, great, like let me give you some tooling so you can make those small changes? And again, might need approval workflows and things like that. But like I maintaining a bunch of stuff that you built previously is not what most people go into software engineering saying, you what I hope I can do is maintain a library of a thousand Python scripts for the accounting team is not, is not where most people say, I'm really glad I
Joshua: Exactly So true. So true.
Andy: did that master's degree in CopSci. I think the second part is
Joshua: Right.
Andy: even even in an engineering team, the reality is. I want to build something. I've got this great idea. And I think my skills as a software developer, even if I'm agentically building, help me think through the logic of how that should work. I still think that's incredibly valuable. And I can give some
Joshua: Mm. Right. Super valuable. Super.
Andy: examples, even in my own life right now, like some stuff I'm doing where the the the knowledge of how stuff kind of works at Ultrix helps me build the things that I'm building because I understand what we're trying to do. When I have an idea like that as a software developer or product manager, what often happens is you go, hey, I have this great idea, I'm gonna build this whole thing. However, I have these mundane dependencies, which are you know, other parts of the code base I need to be different. I need an API on top of this service that's sitting somewhere. Writing that API
Joshua: Sure. I need a service, yeah.
Andy: is not an interesting computer science problem. It's just jamming out some code to like wire that into
Joshua: Yep. Yep.
Andy: whatever the API layer is, right? And unfortunately, that's where a lot of innovation kind of goes to die, where you add up three or four of those things, and there's a team that owns that service, and they're not ready to build the API right now, and they're not sure they want me to build the API on top of their service. And you kind of go like, okay, I'll go work on something else, even though building that thing would have been really
Joshua: Yeah.
Andy: useful to some customers. And my hope and I believe what's gonna happen in this sort of post AI universe is I could say, well, I want to build this thing and I'm and I'm using Claude code to help me build it. And Claude's identified a bunch of the dependencies ahead of time because it's seen a bunch of the code base that maybe I wouldn't have had the ability to go read through myself. And it's identified that yeah, there's some places where I need some mundane work done, some some APIs or an updated library somewhere, whatever that is. And it could actually spec out for me. And I could go to those teams and say, Hey guys that own the APIs, like I I know you're not planning to build this API for this thing, but I need an API on this for this project. And I've spec'd it out with Claude Code. And here's the thing that it wants to implement. Can you take a look? And if you guys are good with this, I'm gonna have Claude build that. I'll send you the merge request so you can see it and then we'll we'll push it on. And I I don't think we are honest with ourselves enough as a software industry of again, just how much of that stuff people deal with every day.
Joshua: Yeah,
Andy: In any kind
Joshua: exactly.
Andy: of scaled engineering team, right? Yeah, if you're a startup and there's five people in a room and I show like, hey, Bob, I'm gonna update your API. Bob goes, okay, fine. Like that's different than you work at a
Joshua: Yeah, exactly.
Andy: large scaled organization, whether you're building software as a company or just implementing software, right? You work at a large global bank or something. Like you as a developer don't just get to go put an API on top of somebody else's service. But again, like I think there's an emerging language of like, hey, I think this thing needs an API for me to talk to it. And our agentic coding platform that we share has identified what it would look like to build that API. It's not super sophisticated. Again, it's not new computer science. I'm not changing how your underlying service works. Here's kind of what it would go do. Is everybody good if this thing just implements this? And that's a maybe a conversation over Slack documented in Jira, and all of a sudden that thing's got an API, and my project moves on. Like that, that's an exciting.
Joshua: That's progress, yeah.
Andy: universe. And I think that same again, everything I think about with software engineering teams doing, I then turn and think about if we had the right abstraction layers and tooling for our users, how do they do the same thing? Hey, I'm building an agent. The agent is doing a revenue recognition calculation, but for it to do a revenue recognition calculation, I actually need this bit of data that sits over here in this universe. Is there a workflow behind that? Is that calculated somewhere? Can I use something that someone's already built? If not, can I figure out how that gets put together? Can I communicate with those teams what that would be? It's no different. I'm just not doing it in Python and I'm not going to use words like merge request and pull request and get repo. I'm going to talk about, hey, on our Altrix workspace, we've got a set of workflows that define this business logic. I need to access some of it, change some of it. AI is going to help me do it. That's the world we're moving towards.
Joshua: So let's dive into that. Let's let's let's talk about the world of this business logic. You you and I and and a lot of folks at Altrick sort of really tied to elevating the importance of solid deterministic business logic and and this idea that we've got all roads are lead to AI and Folks are already getting burned. They're already getting you know, having struggles of various kinds, whether it's token costs or it's risk in in all the forms, security risk, but financial risk, reputational risk. let's start broad and and just say like how do you think about this problem? Can you expand on on that maybe and how you say how do you think
Andy: Yeah.
Joshua: about business logic in the concept of in the in the context of AI?
Andy: The core concept is, and you mentioned sort of deterministic, probabilistic, like sort of fancy words we use in the AI space around, you know, AI does this cool new stuff where it it figures out things in a way that doesn't have to be if then, like we did before. And
Joshua: Yeah. Yep.
Andy: I think that's awesome. Like that's a new set of capabilities we hadn't had before. It is probabilistic. It's using models to figure out what might be the right answer and using logic and reasoning. I tell teams I work with, you know, we never had probably in our software before, right? We didn't we didn't run a dashboard and tell you sales were down in Central Europe in your mid market. And it's probably because of this. Like
Joshua: Right, right.
Andy: software just didn't do that, right? You had to go in and figure it out yourself. And so that's awesome. And the idea that the software could tell me it's probably because, you know, currency fluctuations in the market and a supply chain disruption in Central Europe. You go, that's really useful. I would have had to Start sending emails because the dashboard is red and and why is that? And here AI has helped me do that. There are other places where the word probably is not helpful. This will probably pass audit. we we did we do
Joshua: It's not a good idea. Yeah.
Andy: the month end close the same way this month as we did last month? Yeah, probably. Like, no one feels good about that, right?
Joshua: Probably. Right, right. Yeah, yeah.
Andy: you know, am I gonna get paid my full commission when I close this deal? Probably. Like, that's not the kind of motivating factor for people.
Joshua: Okay.
Andy: So So I I want to be careful. I don't think it's that agents are bad or that deterministic logic
Joshua: Right, right, right.
Andy: is so good. It's that these are these are two capabilities we now have, and we have to figure out how to weave them together and have them make sense. And so when we talk
Joshua: So perfect, yeah.
Andy: about business logic, I think about things that we want to have always be the same. So if I'm calculating my month end close, I want that done using the same rules every time. And importantly,
Joshua: Yeah, exactly.
Andy: I don't want it using whatever it deems to be the best. Current rules available to any company in the market, right? This is not a place for artisanal accounting where you know Claude says, hey,
Joshua: Yeah.
Andy: there's a trendy new way to do month end closed, and I did it for you this month. And I want to close the books the same way I closed them last month. And if you have a better idea, Claude, I want you to tell me, hey, there's a better way to do this. I'd like to propose a policy change. And I can decide if I want to
Joshua: Yeah.
Andy: implement that going forward. Great, super helpful.
Joshua: Yeah we
Andy: Now Inside something like a month-end close, I might want to introduce
Joshua: Yeah, yeah.
Andy: a probably, right? So maybe I've got a 10-step workflow that today does part of my financial close. Maybe it's a reconciliation. What would be really neat is what happens when something doesn't reconcile? Maybe that same deterministic workflow has a new step. And the new step goes, hey, I'm going call Claude, and Claude is going to tell me what it probably might be that didn't reconcile. And it's gonna make a record of that. It's not, it's not doing it for me. It's just going like, hey, Andy or accounting team, you know, this record didn't reconcile. There's this other record over here that didn't reconcile. And I think that's probably the same record. You should have someone look at that. That's a great step to add to my month end close. Then I
Joshua: Yeah. Yeah. huge.
Andy: want it to continue on its deterministic path of doing my month end close. And so for first pass of the month end close, it goes, hey Andy, great, 99%, everything's all set. You got actually three things you gotta look at that. didn't reconcile the first path. And I think I know what they probably are. Awesome. Way
Joshua: Yeah. Yeah.
Andy: better than, hey, I made some assumptions on those, did it myself, didn't
Joshua: Good luck. Yeah.
Andy: tell you about it. And when your auditors show up, you can try to explain to them why I did the thing that I didn't tell you about because I felt like doing it as your AI agent. So business logic is the ability for someone on that financial team or that accounting team that does the month end closed to say, I want to implement some logic. That I know is our business rules that runs the same way. We talk about Vora at Ultics, that is visible,
Joshua: Every time. Yep.
Andy: understandable, repeatable, and auditable. Can I go to my boss or my auditors or my team and say, Yep, like here's what we did? And they're gonna understand it. And again, this is where that sort of low-code, no-code environment matters. When I show somebody an Ultix workflow, I can show an Ultics workflow to somebody who doesn't know Ultrics. And you can walk through
Joshua: Absolutely.
Andy: and go, I see what this does. Like it, you know, okay.
Joshua: Yeah. That's right.
Andy: Is it understandable? Like can I read it and go, makes sense? Is it repeatable? Hey, I ran the same thing this month and I ran it next month, and I'm gonna run it the month after that. And if you run it and I run it, we get the same answer. Great. And again, maybe with the exception of those probably steps. Maybe, maybe your AI suggested a different solve than mine, but it's going to a person who's gonna look at it. That might be fine, right? and then is it auditable? Can I go back at the end of the month and say, hey, we ran the month end closed? Do we all stand by what it did? And can I show it to somebody else and
Joshua: Did two plus two equal four? Yeah.
Andy: review it? Right. And so I think that mindset is important for folks that run processes. And again, this could be supply chain or this could be, you know, how I do my commerce reconciliation, like whatever it is, a Sarve's Oxley process. that business logic needs to exist. We also need to be able to tell AI where it is and how to use it. Right? We talked a little bit about kind of skills and tools. And so there's an emerging architecture for AI for those of you that are kind of digging in and building agents and things. You start to think about skills and context. and that's the other thing that that makes this feel kind of you know single player right now. Is that if you and I both boot up our own instances of Claude and you've been talking to Claude about some things and I've been talking to Claude about other things, and we ask it the same question, it has different context and it might
Joshua: That's right.
Andy: use different tools. And so what we have to be able to do with in an enterprise is be able to say, hey, if you're I don't know, pricing out a proposal for a customer, when we price out proposals, I need my agent to understand and comply with the following rules. Maybe I've got a skill that does import-export compliance. Maybe I've got a skill that does revenue recognition compliance. And it's not just that it can calculate the revenue recognition, it understands like, hey, here's how we do revenue recognition. Because when you're iterating
Joshua: Yeah.
Andy: on deal pricing, you kind of have to know all the constraints and the rules and I have to have conviction that you're sort of following our policies. And so I have to be able to sort of inflict upon AI a set of rules and guidelines that I know it will follow if I'm going to trust it. And again, if that's the case, then that, in this example, revenue recognition rule, I think has to be owned by the team that does revenue recognition. Back to our earlier point of nobody on the software engineering team wants to own a thousand Python libraries, nor do I want to have an update to our revenue recognition where I'm waiting on the IT team to implement some kind of Python code change that we both think kind of does our revenue recognition. Like I've got to be able to go do that. Hit save, update it, maybe it goes through an approval process. Now it's live. Now our agent uses our new revenue recognition rules. Like that's the world we've got to get to. And I think, again, I think we're on the verge of of doing it.
Joshua: Yeah. I I know that if we go to any leader in any company, they know that they these these things that w you're you're mentioning. You know, the the numbers have to be correct. The the processes need to be there and clear. the the team needs to be working off the same page, essentially. But still there's such a fever pitch excitement still around going after AI. And I'm sorta I I don't know if I'm as nice as you are, Andy, and I sort of say that there's this delusion towards thinking that AI is gonna solve all your problems. And I I literally I try to make it funny when I was like, you know, all Alter's history has al always been about this aha moment, right? And there's There's this first aha moment is just that moment when you realize that it the thing you're doing is, you know, Alteryx or AI is way more powerful than you ever imagined. And then, and and that's everybody has to get to that that point. And I you you got there, I got there, everybody. But the second one, I I don't know if people get to. And the second one is around this this aha moment that is a little bit more somber. It's the moment when you realize that AI can't or shouldn't. do everything, right? That there are limitations, there are capabilities that are not inherent to what we want to have happen. And it's like, okay. Like I I get it now. And I almost can talk to someone for 30 seconds and figure out where their head's at, because you know, if I if I ask you a question, you you sort of have all this the construct of like, okay, I'm going to use it here. not gonna use it here or I'm if I do use it here, I need it to have the skills and the context and the the pieces to it. Where there's some people you're like, yeah, I just throw everything in there and it it just figures it out. You know? And it it's sort of I I I it's this switch that I I just am so passionate about trying to get people to see that second aha moment that that there are limitations that we have to account for. There are there are lawsuits coming out. There's legal risk that that that is happening. There's companies that are actually losing money on this topic, right?
Andy: Well, look, I Josh, I think you're right. I think I'm a massive AI optimist, and I think what's happening and what you're describing is sort of a like a Dunning-Kruger effect, right? So, you know, Dunning-Kruger effect
Joshua: yeah.
Andy: is when you don't know much about a problem, it seems simple. Then you learn a bunch about
Joshua: That's right.
Andy: it and you realize how complicated it is, but then you've learned a bunch about it, so you get smarter about it and it sort of gets simpler
Joshua: That's right.
Andy: again, right? It's sort of a very, very simple interpretation of the Dunning-Kruger effect. And I think this is what we're all experiencing, which is if I know a lot about a topic and I go in with AI and I you know I connect a bunch of my data and I'm I I can sort of communicate with it in a way where I get pretty good results. If I don't know anything about the thing that I'm doing, I can also ask it a question. It makes a whole bunch of assumptions. I go, my gosh, amazing. Like it it calculated our entire
Joshua: It's yeah. Yeah.
Andy: revenue recognition model with one prompt. And
Joshua: Mm-hmm.
Andy: then you go, Dunning Kruger effect, when I click on that again, surviving first impact. If I took that to my revenue
Joshua: Choo That's right.
Andy: recognition team, they would go like. It's not wrong. It's just not how we calculate revenue, right? It's this is artisanal.
Joshua: What it w Yeah.
Andy: Like it just came up with a random way to do it and that's not helpful inside
Joshua: Hundred percent.
Andy: a company. And so I think we're going through that at scale, both
Joshua: Think you're right.
Andy: individually and as companies. What I again think will happen is we will realize how complicated our businesses are and how do
Joshua: Mm-hmm.
Andy: we represent that in AI? We will need tooling to
Joshua: Yeah.
Andy: do it. We'll go through that. my gosh, this is harder than we thought it was. And then as we figure out, actually, if we empower all the people that know those things, we give them the ability to get that logic codified in a way and we put it into some skills that can be shared across the organization, we'll sort of come out the other side. Like, actually now when I talk to AI, I don't have to understand our Import export rule details, they don't have to understand Reverec. I can just talk to our pricing bot and it comes up with really good pricing. And when I show it to teams, I'll go, yeah, that looks really good. And it gets simple again, right? But we've got to go through that. Like that's that's just where we are
Joshua: Yeah, that works. Yeah, yeah. Yeah. Yeah.
Andy: right now. And and this is, I joke, this is why CEOs think AI is amazing. We are the masters of the Dunning Kruger effect, where we're up here at the top. We don't understand all the complexity. Look how easy this was. I just I don't know why it's taken
Joshua: Yeah yeah. It's like it's all amazing. Yeah.
Andy: you guys so long to calculate this thing. I just asked Claude and it did it for me, but it's like, yeah, but Claude
Joshua: Yeah.
Andy: didn't. follow any of the company's policies, didn't understand the complexity of our
Joshua: Ha ha ha.
Andy: contracts and our deals, and it made a million assumptions that made this easy and simple.
Joshua: Yeah.
Andy: When I learn more about all the complicated aspects of the business, the answer isn't don't use AI anymore. The answer is,
Joshua: Right.
Andy: hey, how do I empower my teams in a multiplayer way to encode that logic, make it Vora,
Joshua: So good. Yeah.
Andy: make it shareable, wire it back into the AI and come out the other side, and that's the journey we're
Joshua: Yeah.
Andy: all on.
Joshua: I I don't know if it's just CEOs though, Andy. I I I I love our GTM folks across the board too. They're they're probably right there with you, don't you think? Like the GTM you right? Yeah, yeah.
Andy: I think it's everybody it's it's it is everybody with with I just I laugh because CEOs are like, my gosh, AI is gonna replace you know all this capability in my company. It's like, well, I think you got a ways to go before it really understands
Joshua: Yeah.
Andy: your business. it doesn't mean it won't shift how we staff things. You know, I'm a big believer in
Joshua: Yeah, right.
Andy: the theory of constraints, and I think AI is gonna massively change the constraints in every part of our business. We're gonna have sort of this abundance of new intelligence. We're gonna have to implement all kinds of new business logic to support that intelligence. And so I think the makeup of our teams will change a lot. It doesn't necessarily mean it's job destructing. It just means, hey, if one of my teams is five times more productive, but they're constrained by a team that's only twice as productive, I'm going to reallocate the staffing levels of my teams and hopefully grow a lot faster and have a more efficient business because of AI. So I think all that sort of happens. But that's the transition
Joshua: It's all positive, yeah.
Andy: companies have to go through. I think individually we're all going through this dunning Kruger. It's not related to CEOs.
Joshua: Percent.
Andy: Every one of us has this. Hey, I didn't know a lot about this. I asked AI. The results are amazing. Hey, I know a lot about this. I asked AI. And it's like I really had to refine what it was doing. It wasn't that it was wrong or not worthwhile. I just knew how to ask the right follow-up questions. actually, here, let me bring in little bit of extra data. Let me make that make a little bit more sense. Let me r refine your thinking here. now that's really useful. Now it's really useful, but now I want to send it to you. You don't have all that context I provided. You sort of go like this thing doesn't make any sense to me. So again, a lot of work to do to make all that sort of wired together and and make sense.
Joshua: Yeah. No, it's I I like these sort of being able to apply the advanced thinking that's taken fifty, hundred years to to get to because it does help us sort of make the ground not feel like quicksand sometimes 'cause it it's, you know, there's we're all in the same boat. We're all learning what's possible. if we keep going, we're gonna end up talking for another two hours. So I'm I'm hoping you would be up for what what I call rapid fire. I I have this little tidbit towards the end where I like to throw stuff at you and get your sort of off the cuff gut reaction and and quick quick thought. You up for it?
Andy: Sounds slightly terrifying, but let's do it.
Joshua: Let's go, let's go. I want no personal questions. I won't ask about finances. This is this is all socks compliance. so
Andy: It's a safe space. Okay, great. Let's go.
Joshua: yes, I I have been called a therapist in pretty in other podcasts, but first one, personal AI freedom or enterprise AI guardrails.
Andy: I want to say both. I'd go freedom first. I think we're still in the experimentation phase.
Joshua: Okay. Ooh, good one. All right. Directional insight or auditable answer?
Andy: I think we need to get to auditable answer. I think we're currently living in directional insight.
Joshua: Perfect. Single player speed or multiplayer trust. Sort of
Andy: I think we gotta get to multiplayer trust. I think this is the whole we all go further together analogy. if we really wanna
Joshua: Yeah, yeah, yeah.
Andy: get to fast AI in a company, we've gotta get to multiplayer a lot a lot
Joshua: Yeah.
Andy: sooner.
Joshua: But but there is value to that single player speed, you know, peop figure it yep.
Andy: Yeah, yeah. No, absolutely. We but we have that now. So like I I think if you if we go to like why is B to B AI not just taken off? Why are all of us not every single day talking about all the processes that we have agents for? Like we gotta
Joshua: Yeah.
Andy: get to multiplayer.
Joshua: Love it. All right. One assumption that AI should be allowed to make.
Andy: I think it I think it's lots, frankly. I just need to know which assumptions it's making. So one of the things I would love if
Joshua: Mm-hmm.
Andy: AI did a better job was saying, hey, like I assumed these things when I did this. but
Joshua: Yeah, yeah.
Andy: I think that's part of the magic is, you know, hey, I'm trying to solve a problem. I get that it's complicated, make some reasonable assumptions and let's move forward. I I think many, but I need documented.
Joshua: It it's interesting when I when I just did a quick rabbit hole on this, when I've gone to you know, multiple steps down a AI workflow and have got to an end result I didn't want or didn't expect, I I would sort of back channel, ask the hey, how did you get here? What were the assumptions you make? And Good lord if it didn't say well, six months ago you you and I had this conversation about XYZ and I assume that that's still in place. And I'm like, the stuff that you like is keeps me on my toes. Like, actually you're right. Like my assumption was wrong.
Andy: Well, I'll give I'll give you I'll give you a very concrete example. I built a little agent, again, just playing around with this stuff. I built a little fitness agent. and I wrote this one up as well, I'll share at some point. and what I like about it is I can ask it to plan a workout for me. It knows like what equipment I have at my gym and things like that. but I wanted to do more of like food and diet tracking and I hate those tracker apps where I have to like enter every little thing.
Joshua: I know, three hundred and twenty four calories. Yeah.
Andy: Well, or like, you know, I went to a restaurant that's not in the app and I had a BLT and now I'm like,
Joshua: yeah.
Andy: do am I assembling like one piece of lettuce with like two slices of tomato? You know, it's like, was it a Roma tomato? I'm like, I don't know. Like it was it's it was just a tomato. Right. And
Joshua: Yeah. So so painful. Yeah yeah.
Andy: so the agent I wrote at home runs on my local machine, and so it's all private and great. But I basically say to like I had a BLT for lunch. And it goes, Okay. And it goes and makes some reasonable assumptions of like what the hell's on a BLT? Great.
Joshua: Yeah.
Andy: And do I really
Joshua: Yeah.
Andy: care if it's off by 50 calories? Not in the slightest. Like, great. Like it's a BLT. So that's the kind of assumption that is really useful.
Joshua: Let's go. Yeah.
Andy: And you realize like good software should be able to make all kinds of assumptions like that. That's useful, right? I'm
Joshua: They're reasonable. Yep. And
Andy: also not running a restaurant and ordering my monthly supply of lettuce based on this assumption, which if it was off by 10%,
Joshua: Exactly. That's right.
Andy: might be a really big deal. So again, knowing
Joshua: Yeah yeah yeah.
Andy: where you're using assumptions and where you're not.
Joshua: Yeah.
Andy: Is sort of the power of AI.
Joshua: Yeah. Well and it's a it's a power of a a leader too to understand. I I like it. one enterprise AI metric leaders should stop celebrating.
Andy: I think it's token usage, but I think we're getting here as an industry. People are realizing, number one, it doesn't necessarily drive value. Number two,
Joshua: Totally
Andy: we are often paying people to each independently run the same calculations and get different answers and paying the AI companies for the privilege. So a lot to be done there.
Joshua: Yeah yeah. Amen to that. that's a that's a that hurts me just thinking about it. all right, last one. These haven't been too bad, right? Like
Andy: All good.
Joshua: one hands on AI project that most changed the way you lead.
Andy: I think it was the first time I wired an agent into both Ultrix and Snowflake, and I could start doing my own deep dive analysis into the data and basically using it like a scratch pad to figure stuff out and then publishing it into Ultrix. And now I'm able to go in, look at my business in detail, but actually share out what I'm doing. And that's been really fun.
Joshua: So good. So good. amazing. All right. So we gotta wrap up. Mr. Andy, can you leave people with one thought. Like what's on front of mind? What's what's something that you you want everybody to to leave the podcast with today?
Andy: I mean, one, I think everybody should be playing around with AI and and getting into not just the web interfaces, but like try these C L I's, these command line interfaces and starting to really learn how to
Joshua: Yeah.
Andy: build things like on your machine and then immediately start to transition like, okay, if I can build stuff like this, how am I gonna do that across my team? and so I think that's a a big exciting new area for people to to get hands on and think about how to how to scale building.
Joshua: Absolutely. And then last thing is we I think most people could figure out that you're on LinkedIn. Is there any other places that people can reach out to you or contact you if they had a question for ya?
Andy: I'd say LinkedIn and X are the two big ones for me. my handle
Joshua: Mm-hmm.
Andy: on both of them is just A P McMillan. P is in Paul. So A P. McMillan, you can find me on either place. happy to
Joshua: Cool. Perfect.
Andy: engage with folks.
Joshua: Love it. Andy, thank you. Sincerely I appreciate it. I love all our interactions and a whole lot to digest today for sure in an hour. So thanks again. Appreciate ya being on.
Andy: It was really fun. Thank you.
Joshua: Alright, take care.
Joshua Burkhow: You know what stayed with me from this conversation is that your knowledge of the business still matters. AI needs the rules that your team understands and the judgment that you've built through experience. Andy gave us a useful way to think about it. AI can suggest why something happened, but the rules behind a business decision need to be clear enough for people to check and use over and over again. So here's something you could maybe try this week. Take one of these AI experiments that helped you, show it to a teammate. Can you see the assumptions? Can you understand the steps? And more importantly, can they use it without having to start all over? That's a practical step towards making AI useful across the team. I want to thank Andy McMillan for joining me. Thanks to you all for listening to Ultra Everything. You can listen to it on any podcast that shows alter everything. I'm Joshua Burkow. See you next time.
