Why Sage CTO Aaron Harris Isn't Worried About AI Killing SaaS.

Attention: This is a machine-generated transcript. As such, there may be spelling, grammar, and accuracy errors throughout. Thank you for your understanding!

Matthew Lescault: So we've got a really special episode today. I had a really awesome chance to sit down with Doug, Emily, myself, and Aaron Harris, the CTO of Sage, to talk things about AI innovation, maybe a little bit about his origin story. Such a great conversation wasn't just about this episode. We have two and we're really excited to present them. The first one, we're focused on AI and how Aaron looks at AI from a [00:00:30] business perspective, but also a personal one. And then our episode two will really talk about the origin story from Aaron's eyes, how it came about, how it got to where it is. We're really excited. I hope you love these episodes. See you on the other side.

Doug Lewis: Whiteboard. Notes. App. Spreadsheet. Where [00:01:00] are you taking your notes? Or is it just all up here?

Aaron Harris: No, no. So I, I, I mean, I hate to admit this, but I've been using Evernote for 15 years.

Doug Lewis: Really?

Aaron Harris: Um, I recently coded a tool to export it all into a database that one of my AI agents can access. This is all in an attempt to get me off of Evernote. I hope it's not watching because I've been a customer for 15 years. But yeah, Evernote is, is where I take notes. [00:01:30] Um, I've tried this. The, the, what is it? The readable or the remarkable? The remarkable. I tried remarkable. It's a really cool device. It lasted about a month and I was back to Evernote. Uh, it's just text. Text notes.

Matthew Lescault: So how how is that journey right now? Getting off Evernote. Are you are you back?

Aaron Harris: I'm back on it. I haven't found anything that works across every device that I use really seamlessly. However, my agents Have consumed all of my notes, including [00:02:00] my personal journals, going back 15 years. So theoretically, I could just switch to talking to AI agents as a form of note taking.

Emily Madere: I do that. Um, do you feel like your AI agents, like actually help you? Absolutely. Are they to the point where they can help you?

Aaron Harris: Yeah, yeah yeah, yeah.

Emily Madere: So they help me with my.

Aaron Harris: It's kind of cool when note taking can turn into when it can be more conversational. Right. So you can inject. So if you're in a meeting [00:02:30] and I say, um, some, some abbreviation that I should have explained, but I didn't, and you're taking notes and you're actually having the exchange with AI. Ai can explain to you exactly what it was that I was talking about. That didn't make any sense during the conversation.

Matthew Lescault: I got a text recently from a Sage person. It was like acronyms. I'm like, I'm about to ask him what he actually meant. And this would, that would that would help. Now we're not here for AI, but I do have one question on that. How many agents [00:03:00] do you deploy to support yourself? Like personally? Yeah.

Aaron Harris: Uh, I have. Okay. So we got to be clear though, that I like a lot of it is just experiments and learning. Right. So I probably have 6 or 7 agents that I can talk to at any time. Um, running on. I've got a Mac mini at home. That's, that's running some agents. I've got an old MacBook that's running an agent. Um, and then I've got obviously Claude ChatGPT, [00:03:30] whatever else is interesting at the moment. So I feel like it's a real army of them.

Emily Madere: I feel like you name your agents. Yeah, yeah.

Aaron Harris: I have Deckard. Uh, Deckard is my personal assistant.

Matthew Lescault: Um, why?

Aaron Harris: The name Deckard comes from Blade Runner. And if you're a fan of Blade Runner, like the real question of Blade Runner is. Is Harrison's for Harrison Ford's character human or a replicant? Oh, you don't know. So [00:04:00] when I when I when I brought Deckard to life, I wanted him to embody this tension and actually be wondering whether he's real or he's AI. Now they're trained to, to, to, to not ever assume that they're human. But I really pushed that. I want, I want, I want these human interactions. So I've got Deckard, who's my, my personal assistant. Uh, I have Arthur, who I talked about at the conference. He's my personal accountant. He's terrible. [00:04:30] He's not very good at his job. Um, I've got Stu. Stu is basically capable of short 2 or 3 word interactions. Like, I'll say what's up? And Stu will say, hey, man. And and the reason for this is I'm like, I'm trying to get to the, like the bare bones minimum. Like like, can I host a really small model that is tiny [00:05:00] and really not powerful compared to, you know, the big models. So I've got Stu that allows me to play around with, with, with little models. Um, I've got, I've got an artist model, um, named Chris, also from Blade Runner. I don't know why I named it Chris. I just thought she was a cool character. Uh, she connects to, to various, um, image generation. So she made all the little avatars for the rest of the agents.

Matthew Lescault: Okay. Can I, can I assume that Blade Runner [00:05:30] is your favorite movie?

Aaron Harris: Blade runner is my favorite movie.

Matthew Lescault: How many times have you seen it? I've lost count. Okay, so if we wanted to have, like, a movie night at Sage with all 20,000 employees, it would be Blade Runner. Probably not. It's.

Aaron Harris: It's too personal.

Matthew Lescault: Okay.

Aaron Harris: And nobody. It's it's not like a crowd pleaser.

Emily Madere: I've never seen it, just transparent.

Doug Lewis: I think it's got a cult kind of, you know.

Aaron Harris: It's dark.

Doug Lewis: Yeah.

Matthew Lescault: You know, somebody doesn't understand. It feels like [00:06:00] it's slow, you know? When I first watched it, I was like, what is this? And then as you got older, you really started to like, get into like what the concept was and why and so forth. I've seen it a long time. You're gonna make me now go watch Blade Runner again.

Emily Madere: You're gonna watch on the plane.

Matthew Lescault: No flight from here to Maryland is not is not short enough for that. Yeah. It's like or long.

Emily Madere: 52 minutes for me.

Doug Lewis: You know what's really impressive about this? It took us, like, less than ten minutes to completely go off the rails.

Emily Madere: Sorry.

Doug Lewis: Usually [00:06:30] we wait a little bit longer before we go off the rails, but.

Matthew Lescault: Well, your job was to bring us back.

Aaron Harris: Let me just. I have one more agent I want to describe. I mean, I have a couple, right? But but I have another agent called Holmes. Like Sherlock, Sherlock Holmes and I have empowered Sherlock Holmes with 15 years of my personal history, as I mentioned. And so if I need to. And all of my notes. Right. So if I need to say, hey, what is my son's social security number? I can say that to [00:07:00] Holmes and he will go off in Sherlock, right? He will, he will look through my notes to, to find where I somehow recorded or what is frequent flier number is for Delta Airlines, right. Um, so Holmes is my, my research.

Emily Madere: I think we all.

Aaron Harris: Not research for important things. Research for my personal history.

Matthew Lescault: So you, you actually brought all of your journaling and notes and your personal history into this agent to support you from not anything professional, [00:07:30] just everyday life. Yeah. Okay. And what we said we were going to do the I think I have to ask, everybody is worried about putting personal information into AI.

Aaron Harris: You guys told me we weren't going to talk about heavy stuff, but but go ahead.

Matthew Lescault: We did and we did, but I it's a personal question because me and my wife have this battle. It's like she's like, not put pictures of our kids into ChatGPT or Claude. She's just like adamant about this. But you're obviously very comfortable putting certain information. [00:08:00] I obviously don't know that. What would you say to the general public about your your concern over personal security?

Aaron Harris: Uh, so I mean, the devil's in the details. Um, the models themselves, right? They're just, they're just algorithms, right? There's, there's, it's the systems around them, right? It's, it's who is hosting the model and, and, you know, who has built the systems for interacting with the model. Uh, so for, for, for [00:08:30] Sage Intacct, we've obviously developed a lot of our own models, um, which, you know, we, we train and we operate for like high level reasoning tasks. We will use an anthropic model or an open AI model, but in those cases, we access those models either through AWS bedrock or through Azure because we can actually, um, we can have agreements in place with those model operators [00:09:00] to not access any of the data that is used to generate the, the inferences or the come in the prompts. Right. So you have to kind of look at the agreements. Uh, most of the, the chat bots, if you will, you know, Claude ChatGPT, there is a setting, right? You can set a privacy setting on these. So you just need to be kind of careful about who's operating the model and what, what the privacy agreements they have in place. I think it's a very legitimate concern. Uh, [00:09:30] but just, you know, be educated about, you know, who's operating the model for the models that I use with my agents most. I mean, I actually operate a lot of the models myself, but they're the, they're like stew, right? They're not very smart, right? They can't hold a candle to anthropic models. So for the big, important things, I do have to go out to anthropic. And you either have to sacrifice your personal privacy for the power that you get right. Or be [00:10:00] really careful about the agreements.

Matthew Lescault: Makes sense. Makes sense. Thank you.

Emily Madere: My first experience hearing you speak, it was in Vegas. Um, and you were live demoing the AI features and intact. I believe it was like a partner day and you were like, the team is really going to get me for this one. Uh, and you.

Aaron Harris: I remember that presentation really well.

Emily Madere: Yeah. And you live demoed, uh, some AI features. Um, that, that was my first time hearing you speak. And I've, um, heard you talk about [00:10:30] patchouli probably every time after that.

Aaron Harris: Totally. Yeah. Yeah. I still regret not pushing that as the name for Sage copilot.

Matthew Lescault: I, I probably shouldn't say this, but I was like, why Copilot. Like from a naming perspective. I feel like it could. Sage. Ai would have been a better name than Sage copilot, but that's that. Now you're probably like, I should not be talking to this guy.

Doug Lewis: They didn't ask you?

Matthew Lescault: No.

Aaron Harris: Well, the rationale, the rationale was at the time, there was there [00:11:00] was pretty broad belief that it would become a generic term, right? That that people would talk about their copilots and that would be a category of AI solution. And so, you know, we took a little bit of a gamble that, um, you know, this would become sort of Kleenex or Xerox, right? Um, it did cause a little bit of confusion because, you know, I always want people to know that no, we built this, this is not Microsoft's, we built this with our own two hands. Mhm. Um, but, but fundamentally, the, the idea was [00:11:30] this is going to become just a standard terminology.

Matthew Lescault: And you actually explain that you probably, I mean, as much as you speak and people you meet. But a few years ago you had presented in Canada at account tax. And I came up to you and I said, is copilot just Microsoft copilot? You're like, no.

Aaron Harris: You're one of those.

Matthew Lescault: Well, I was asking, I didn't, I wasn't sure.

Aaron Harris: Yeah.

Matthew Lescault: And you, you corrected me very, very specifically on the fact that you have multiple models around that, that it's just a term. It was more of a, a, a brand [00:12:00] than it is a, a product that's.

Aaron Harris: Yeah, yeah. And, and, you know, we've, we've got a lot of terminology that we use around this, we call the broader AI initiative just Sage AI, right. Um, and copilot is the experience, right? That's, that's how you interact with our AI solutions.

Doug Lewis: How big is the AI team today?

Aaron Harris: Well, so there's no, it's, there's no longer just an AI team. So the, so the core AI team, there's sort of [00:12:30] two teams that do nothing but AI between the two of them. There's a few hundred. Um, but most developers in the business Now in some way, our developing AI solutions. But if you're just talking like pure data scientists or AI engineers, there's there's a couple hundred.

Matthew Lescault: There's a pretty big team. Now, how many do those teams report up to you? Is that is that part of your role at this point?

Aaron Harris: No, they they report up to the Chief Product Officer. [00:13:00]

Matthew Lescault: Okay.

Aaron Harris: And I, I get to advise and influence and spend time with them without actually having any accountability for what they do. And I did build the teams. Um, so I, I had those teams until about two years ago. Um, we moved them into product and I get to devote more of my time to strategy and vision and, uh, being on podcasts.

Emily Madere: Did you say recently they found some of your original code?

Aaron Harris: Oh, there's a lot of my original [00:13:30] code.

Emily Madere: Yeah.

Doug Lewis: That's a, that's a really good question.

Aaron Harris: I was coding, I was writing code for intact up until, uh, I took the global CTO job.

Matthew Lescault: So what would you say would be the percentage of your code still in intact today?

Aaron Harris: Oh, geez.

Doug Lewis: The original.

Aaron Harris: Uh. So until 2017, I was the most prolific coder in terms of lines of code in the product. Um, but [00:14:00] we've had, you know, lots of brilliant developers who are very prolific. I, I'm probably down to 1 or 2% really of the code.

Matthew Lescault: I thought it was more than that. I guess it was.

Aaron Harris: Well, it's always evolving. Yeah. We're always modernizing. I think this is this is the, um, the thing that people don't appreciate about software as a service because we're like, we literally put a new version out on a weekly basis, sometimes multiple times per week. And so, [00:14:30] you know, whatever, whatever technical debt we've incurred is just on the current version. And so we're constantly evolving the code and, you know, modernizing the code, evolving the architecture, bringing in new technology where it makes sense. And so it's not fair to say intact is a 27 year old product, right? Right. A, you know, 50, 50% of the code was probably written in the last five years. Right? So, you know, it is a [00:15:00] modern product. Um, we have thousands and thousands of customers, so we have to be more deliberate and cautious about making changes to the product. But, um, you know, a lot of the capabilities and the product are written on the, you know, the latest standards for how you would build software if you started from scratch today.

Matthew Lescault: Yeah.

Doug Lewis: You brought us up to the SaaS apocalypse, right? And everyone loves talking about that. People have been just predicting doomsday since the dawn of time. Where's your head at with that? Are we in some [00:15:30] bubble that's going to pop, you know?

Aaron Harris: Yeah. So so, um, we, we build software for humans so that humans can do high quality work productively. Why would we not want AI agents to do high quality work productively? Right? You know, the, the, the big breakthrough in technology that enables us to have agents is that, you know, we've replaced deterministic coding with large language [00:16:00] models that sort of problem solve the way a human does, which enables them to do human like tasks. So what does that mean? Right. It means in that creative process, they make mistakes, right? In that creative process, uh, left to their own devices, they're sort of unpredictable in the way they go about things. Why would we not want to empower them with software that makes them more productive, that makes them more predictable, that that controls the quality, etc.. [00:16:30] And essentially what we argue is, and in fact, because this is now a digital workforce, you need more predictability, you need more control, you need more governance, right? So so the first thing that I would say is AI needs a platform like Sage Intacct as much or more than the human users who are using it today, right? So incredibly valid. The second thing I would say is, and I've been saying this for years, general [00:17:00] purpose AI is really good at a at a broad set of things. But when it comes to doing accounting tasks, you need AI that's built for those accounting tasks, right? That understands the nuances of accounting, that understands the language of accounting, knows how our products work, and has been fine tuned to be really good at the tasks that we need to do repetitively in the world of accounting.

Aaron Harris: And so, you know, the way, the way our agents work is, you know, we do use the [00:17:30] large, you know, the big provider, large language models to do the reasoning, right? To have the conversational interaction. But when it comes to the work, they actually do the tools. If it's a tool that requires AI, more often than not, we train the models for that, right? We build those tools. So the second reason why the SaaS apocalypse just isn't, you know, isn't a thing for us is that we are more sort of qualified, um, and, [00:18:00] and set up to build the AI that actually works in accounting because we've got, you know, 45 years of history because we've got millions of customers, right? Because our customers trust us to train models on their data, right? Which the big providers can't do. So it's a, the platform is more valuable or more important than ever. And B, we're going to build agents that are fit for purpose for accounting. And we're we have a lot [00:18:30] of conviction that we can build more capable, reliable, trustworthy agents for accounting tasks than a broad provider of agents can.

Emily Madere: I just took all that in, like, because I'm in sales, right? So I get asked these questions all the time about like Sage's AI like, what are what's going on? What are they doing? Um, I just respectfully have no follow up questions from that. I kind of need to sit down and think about what you said.

Matthew Lescault: Well, I think one of the things that I say and I like, I like [00:19:00] you to challenge the statements like if Doug has the same I, I, I as Aaron has, as Emily has and as I have, what's the differentiator? And it's the individual. And so when we keep talking about how AI is going to impact the workforce, I keep going. Like you still have to prove why you are more equipped to solve or be the the provider and so forth. And I think that's getting [00:19:30] lost in some of the translation that's being talked about when it comes to it.

Aaron Harris: Yeah. Um, it's, you know, we all have to operate more and more as managers of, of talent. And so, you know, you wouldn't hire an engineering manager who doesn't understand the craft of engineering, right? So if you're going to be effective at writing code in a world where you have coding agents, you need to be [00:20:00] a skilled, experienced software developer, right? Um, it's just impossible to, to work with these agents without being able to review the work, without being able to guide the work, right? Without being able to make smart decisions when there's a decision to be made. So, you know, this, this idea that the career of a, of a, you know, of a, of a computer scientist is going to be replaced by a AI accounting agents is just flat wrong. And what I [00:20:30] what I said at our last conference is there is an unlimited demand for software, right? There are so many things that either haven't been solved well or haven't been solved at all with software. Right? You think about, you know, the projects that need to get done, you know, inside a company like Sage, there's so much demand for software. Why, why, why would we go to, you know, 10%, the number of developers that, that we had [00:21:00] in the past because they're, they're ten X more productive. Why don't we write ten times more software?

Matthew Lescault: Some of your competitors, I don't think, have the same vision.

Aaron Harris: Well, everybody.

Doug Lewis: Feel free to dunk on anybody.

Speaker 7: Like every everybody everybody.

Aaron Harris: Everybody is under pressure to be more productive as a result of of generative AI. Right. Of of, of access to these agents. And, you know, there's absolutely an imperative to [00:21:30] demonstrate that, that it leads to more productivity. And what's, what's happening is you have to start to evolve your workforce to, to really leverage the power of that. And so on the one hand, you're going to see, you know, big tech companies do big layoffs, in my view, in a lot of cases, they overhired, right. They were sort of growing recklessly. And, you know, when the music stopped, they, [00:22:00] you know, they, they, they really had a workforce that wasn't fit for purpose. And so under the guise of AI productivity, they're, you know, they're letting big parts of their their company go. Sage has always been disciplined in the way we grow our business, our workforce. In fact, though the company from a revenue perspective is, I don't know, know, 50 or 60% bigger than it was when Sage acquired intact. We've got something like 2000 fewer employees, [00:22:30] because over time, we've been evolving our workforce to take advantage of technology to, you know, to to be more efficient, to bring in people with, with, with different skills. So you're always going to see Sage be more deliberate about these things and be a bit cautious if you will. But what it also means is, you know, the SaaS apocalypse hurts, but if you look it into it, their stock is down 60%, six 0% over the last [00:23:00] year. Uh, ours is down some, you know, in the high 20s, which is, you know, I'm not happy about. But but, you know, our overperformance versus other SaaS companies is a reflection of how steady we've been and the way we've grown the business.

Matthew Lescault: I've always looked at it, and it's been, I find it interesting is that Sage has, from a stock perspective, has never had the deep pits like the deep [00:23:30] falls or the the high rises. It's been far more of a steady a steady pricing. You know, from a stock perspective, I'm not going to ask you too much around that because I think that's that's unfair. But it does talk a little bit about the difference in in, in strategy that.

Doug Lewis: Yeah, don't put him on the spot for financial. Oh no, no, put him on the spot with that. Yeah. Let's let's get deep into financials.

Aaron Harris: I guess what I would say is it's a very diverse company, right. And diversity leads to stability. Um, [00:24:00] so we've got a lot of products and a lot of markets so we can absorb things that happen. Um, and we've worked hard to build an efficient business that returns profit to shareholders. That also leads right to, to, to, to the stability. So yeah, I would really love for the stock to, uh, to, to sort of perform the way it did in the past. But we've got to get through this, this educational process that the software is more valuable now, not less valuable as a result of AI.

Emily Madere: And you're working on it. [00:24:30] I mean, I see the ads that say just pushing out there. So, I mean, y'all are well on your way. It just takes time.

Doug Lewis: It is a weird path, no question about it. For those listening, those watching, last piece of final advice you'd give to those who are out there building from scratch, early stage, you know, maturing through the life cycle of the business. Great, great advice from Aaron Harris. What do you got? One piece.

Aaron Harris: Well, I'll tell you the thing that I'm the most proud of. Um, so so the thing that is, that has made me [00:25:00] feel most proud over the years is the quality of the people who have decided to work with me, whether it's, it's amazing salespeople or brilliant engineers or, you know, people in finance and, and legal that the, the fact that brilliant, capable people have chosen to work with me on this product is the thing that that that that makes me feel [00:25:30] the most proud, I guess. Um, that's that's the wrong word. It's kind of, oh, it's, it's kind of, it's, it's humbling. Um, so I'm, you know, translating that into advice, um, you, you know, you can attract, right and recruit the best people whether you realize it or not. One of the things I learned is that a lot of really brilliant engineers want to work for people that have a vision, right? [00:26:00] So, so even though these people can out code me, you know, they can code me under the table. Um, they want to work for somebody who, you know, sets a direction, who has a vision and can create some excitement. And if you can do that, there's no limit to to the quality of the people that you can you can recruit. So that's, that's, you know, obviously I've got lots of lessons, lots of.

Doug Lewis: Good people and hang on to them, but that seems to be, yeah.

Emily Madere: Be friends with.

Doug Lewis: Them.

Aaron Harris: I think Warren Buffett said you're, you're the average [00:26:30] of the five people you're closest to, right? You know, in terms of, you know, your, your I don't actually know what he was referring to.

Speaker 7: You're the average.

Aaron Harris: But you can sort of get the essence of what he's saying. Like if you surround yourself with, with excellent people, then the outcome is going to be excellent.

Doug Lewis: And we have taken up too much of your time. I would say, uh, even though now I know that you quite literally have nothing on your calendar, so nothing to do, we will have to have you back. But we thank you so much for joining us today. It's actually, it's been a pleasure. And I can't say that about all [00:27:00] of these. So that's.

Speaker 7: Uh, that's fun.

Doug Lewis: I appreciate it.

Aaron Harris: Thank you very much. It's great to be here.

Matthew Lescault: Thank you.