Transkrip penuh untuk Spencer Penn, CEO at LightSource - What AI Means for the Future of Procurement

that I think about it. 00:04:29.680 --> 00:04:33.920 With AI and procurement, gosh, there's like so many things to think about. 00:04:34.160 --> 00:04:43.920 I think it is, it is inevitable that a very, very large portion of white-collar work will be either supplemented or replaced by AI. 00:04:44.160 --> 00:04:55.199 And the question of whether or not something gets replaced by AI or supplemented by AI, I think falls into there's two questions I have to ask. 00:04:55.519 --> 00:05:01.120 One is how tech forward and how tech ambitious are the people in the field? 00:05:01.360 --> 00:05:07.920 And then the second is how much room does that job have for expansion and change. 00:05:08.079 --> 00:05:12.399 And these two vectors I call tech ambition and role plasticity. 00:05:12.560 --> 00:05:27.439 And so when I think about tech ambition, if somebody says, hey, look, I'm really excited about AI and I want to use it in my job, I want to be able to do more faster, then what's gonna likely happen is that AI will become a tailwind for them. 00:05:27.519 --> 00:05:39.759 If they say, hey, look, I'm really not interested in AI and I really want to push back on it and I don't want to engage with it, I'm leaning out, not leaning in, then eventually what'll happen is that the role will be better done without their presence there. 00:05:39.920 --> 00:05:40.959 And that's the reality. 00:05:41.920 --> 00:05:45.519 The second uh vector that I think about is role plasticity. 00:05:45.680 --> 00:05:49.279 And so role plasticity is how much can the job itself change? 00:05:49.519 --> 00:05:54.319 An example of a role that has very low plasticity would be like a toll booth operator. 00:05:54.480 --> 00:06:11.680 And so if I think about a toll booth operator, maybe let's say that theoretically you meet a toll booth operator that's super into AI and they love AI, it doesn't really matter because that job function is so well defined that even if there's an adoption of AI, they can't modify their role. 00:06:11.920 --> 00:06:20.000 By comparison, you know, if we think about software engineering, it's a it's a role that has reinvented itself hundreds of times, right? 00:06:20.160 --> 00:06:29.439 And so when you think about what someone was doing as a software engineer 50 years ago, it was punch cards, it was setting up physical servers, that kind of thing. 00:06:29.680 --> 00:06:34.160 And then you might look at software engineering and say, oh my God, we're gonna replace it with technology. 00:06:34.399 --> 00:06:36.959 We have cloud, so you don't have to set up your own servers. 00:06:37.199 --> 00:06:40.959 We have also languages that deal with memory management directly. 00:06:41.040 --> 00:06:43.680 So you don't need to spend all your time doing that kind of work. 00:06:43.759 --> 00:06:50.800 And so if you look at the software engineer of 50 years ago, that role has been 95% automated. 00:06:50.959 --> 00:06:53.519 But that doesn't mean that software engineers are

now out of work. 00:06:53.600 --> 00:06:59.839 It means that they say, now we can assume the server as given by AWS or Google Cloud. 00:07:00.000 --> 00:07:03.199 Now I have other things that I can focus on and work on. 00:07:03.360 --> 00:07:10.160 And so they found, I think, in software engineering in general, a really strong growth in terms of the compensation that they've received. 00:07:10.240 --> 00:07:15.040 Technology has been a tailwind, not a replacement versus toll booth operator, cashier. 00:07:15.199 --> 00:07:21.759 These are roles where there's often not a lot of tech ambition, but also the role plasticity is very firm. 00:07:22.000 --> 00:07:28.160 And so my predictions for, you know, I think people often, I think there's a few different elements of this question. 00:07:28.319 --> 00:07:32.720 People will ask about what are the areas that AI is going to help automate. 00:07:32.879 --> 00:07:38.959 And then the simple thing is that anything that can be done behind a computer screen in its entirety is automatable. 00:07:39.120 --> 00:07:40.240 I think that's very clear. 00:07:40.480 --> 00:07:43.120 Doing things in the real world is much harder. 00:07:43.360 --> 00:07:48.639 My guess is that 50% of white-collar jobs will be automated within the next three years. 00:07:48.879 --> 00:07:58.639 That's I know it's a very bold claim, but I I think that's actually conservative when I talk to folks who are at OpenAI and Anthropic about the trajectory that the technology is moving in. 00:07:58.800 --> 00:08:09.839 And then the the real metric of like whether or not your job will be automated comes down to, you know, if you can do it from behind a computer screen, if you have a lot of tech ambition and if your role has the ability to change, right? 00:08:09.920 --> 00:08:23.759 And so if your job is primarily just, you know, imagine like a very, very transactional kind of buyer where you're placing orders, you're checking invoice receipts, you're reconciling things between systems, that feels very automatable. 00:08:23.920 --> 00:08:42.480 You know, if you think about someone who's developing a category strategy and then going out to market, building credibility with suppliers to ensure long-term supply continuity, a lot of the things, frankly, in direct materials, I think that that kind of stuff is very, very difficult to automate, especially if you find yourself often going into the field. 00:08:42.639 --> 00:08:52.159 Because, you know, again, LLMs can do a lot on screen, but there isn't uh today uh a really easy way to bring that intelligence to the real world. 00:08:52.320 --> 00:09:03.600 And so if your role involves interacting with people, if your role involves getting out into the public, if it involves going to visit suppliers, I think that you're gonna find it very difficult to automate. 00:09:03.840 --> 00:09:16.960 I don't know if that answers your question, Aaron, but you know, I I I don't have the predictions that are like AI

is gonna replace everything and you know, in the in the but I I do think that uh there are certain there are certain tides that will just shift over time. 00:09:17.360 --> 00:09:19.759 No, I I think it's you know really interesting. 00:09:19.919 --> 00:09:30.000 I think we can definitely see your prediction coming true with more and more advancements in AI and how it's automating, like you said, the white collar work. 00:09:30.240 --> 00:09:35.440 So, you know, is that a prediction that you had even before 2025 and it's just something you're kind of holding on to? 00:09:35.519 --> 00:09:40.320 You said within three years, so we've got until the end of this year, end of next year to come. 00:09:40.799 --> 00:09:42.000 No, that's a very good question. 00:09:42.159 --> 00:09:52.720 Well, every year I'll push it back by one year, but but but right now, no, that that that uh that insight that 50% of white-collar jobs will be automated within the next three years. 00:09:53.120 --> 00:10:09.200 Is actually something that I came to over the summer, so about six months ago, and that was in a conversation again with I went to a I went to a really informative dinner with uh with leaders from OpenAI and Anthropic, and we talk about these topics, right? 00:10:09.360 --> 00:10:21.440 I mean, I think I will say that I think that you know, at Light Source, we have an unbelievable unfair advantage simply in being in San Francisco at this exact time. 00:10:21.679 --> 00:10:36.879 Companies that were started 10 years ago could not have taken advantage of the AI revolution because these things were far less prominent, and companies that are not in the Bay Area don't have access to that earliest sort of cauldron of knowledge. 00:10:37.039 --> 00:10:38.320 And so, you know, it's funny. 00:10:38.399 --> 00:10:44.559 I was looking at, I've been looking at new office space for our headquarters, and I think we're getting close to finding a space. 00:10:44.720 --> 00:10:46.480 But as I look at the listings, it's funny. 00:10:46.559 --> 00:10:54.399 There's this new thing where people are calling one of the neighborhoods the San Francisco AI district, and it's this area called Mission Rock. 00:10:54.559 --> 00:11:04.320 And so real estate brokers, maybe they're the canaries in the coal mine, you know, they're looking for ways to market, you know, certain areas of the city as being, you know, where where AI is happening. 00:11:04.399 --> 00:11:05.279 And it's not untrue. 00:11:05.440 --> 00:11:07.759 I mean, it is a really revolutionary technology. 00:11:08.000 --> 00:11:11.600 I I find myself always amazed by what it's able to do. 00:11:11.679 --> 00:11:20.720 Like I'll be on, I'll be just using even just in my own life, I'll be using Chat GPT and I'll I'll start asking some questions and ask it to do something. 00:11:20.879 --> 00:11:22.240 And I'm like amazed. 00:11:22.399 --> 00:11:24.879 I'm I'm still amazed at what it's able to do. 00:11:24.960 --> 00:11:28.480 I'm always impressed, even though I know it should be able to do this. 00:11:28.559 --> 00:11:31.200 I'm still impressed by what it's capable of producing. 00:11:31.360 --> 00:11:32.320 It's it's just great. 00:11:32.399 --> 00:11:38.720 But the thing is, like, I do feel that it is gonna be a like many technologies, not gonna be a fair distribution of outcome. 00:11:38.879 --> 00:11:44.320 You know, the people that benefit most from AI are probably gonna be the people who, again, are the ones knowing how to use it, right? 00:11:44.480 --> 00:11:58.000 If you, you know, if you think about like the things that procurement has experienced over the last year, I think something I didn't predict is how much people have been asking procurement to do more while while asking them for reduced headcount. 00:11:58.080 --> 00:12:07.759 I think that a lot of a lot of CPOs, it's funny, a lot of CPOs we meet with, they're not, they're not telling me, hey, Spencer, we're looking for savings necessarily. 00:12:07.919 --> 00:12:09.840 What they say is we're looking for efficiency. 00:12:09.919 --> 00:12:11.600 I'm like, what do you mean by efficiency? 00:12:11.759 --> 00:12:12.639 Like cost efficiency? 00:12:12.799 --> 00:12:23.919 Like, no, headcount efficiency, because our headcount is being held firm and our business is growing, or our headcount is being reduced and our expectations are staying the same. 00:12:24.080 --> 00:12:28.320 And so a lot of times what people are looking for are ways to extend their workforce. 00:12:28.559 --> 00:12:31.519 And I think that AI is actually a phenomenal way to do that. 00:12:31.759 --> 00:12:40.000 But if you're gonna say AI is a force multiplier and not a people replacer, then it has to be that the people know how to work the knobs, right? 00:12:40.320 --> 00:12:51.039 Yeah, so you're you're you're just not gonna get anywhere with the tools, less teams and people are given the proper training and can understand how to harness it best for the work that they do day in, day out. 00:12:51.120 --> 00:13:03.360 And you know, I think it's really interesting on those about on San Francisco, you know, I for the first time kind of late last year, and it does feel how unique as a city to you know anyone in the States. 00:13:03.600 --> 00:13:08.320 You know, I I don't think you can ever really get over it the first time you see a like a Waymo car. 00:13:08.559 --> 00:13:08.879 Oh, yeah. 00:13:09.519 --> 00:13:12.879 Did you drive it on around on the roof and you're like, what is that? 00:13:13.120 --> 00:13:22.559 And you ask like a local and they explain to you, you think, my my god, like you that was just someone I would never see or expect to see in in you know in my life and and to see it like that. 00:13:22.720 --> 00:13:33.759 So, you know, for someone like Light Source, who, you know, are at the very cut edge of of AI, you know, San Francisco, obviously it benefits you, but you know, you can't help that, so you just gotta take

advantage of it. 00:13:34.000 --> 00:13:34.559 Maybe, yeah. 00:13:34.639 --> 00:13:34.799 Yeah. 00:13:35.279 --> 00:13:40.559 You could become the um like the capital uh the capital building of the AI district. 00:13:40.960 --> 00:13:43.279 It's really it's really science science fiction. 00:13:43.440 --> 00:13:44.639 I mean, Waymo's amazing. 00:13:44.879 --> 00:13:55.039 Back in, I mean, now it I think what's been shocking to people is that the growth of Waymo in San Francisco versus even even folks like like Lyft. 00:13:55.360 --> 00:14:01.600 I mean, this was uh this was a a post in the um, and I can even I don't know if I'm able to share my screen, but let me try. 00:14:01.840 --> 00:14:10.799 This was the chart of the estimated market share of gross bookings of the different San Francisco operators of rideshare. 00:14:10.960 --> 00:14:15.519 And so you have Waymo, which is down here, Uber, which is up here, and Lyft. 00:14:15.679 --> 00:14:18.159 And so of course Uber is still the dominant player. 00:14:18.240 --> 00:14:22.399 It's an enormous business that's been entrenched for for a decade plus. 00:14:22.559 --> 00:14:29.120 But I was shocked to discover that Waymo had in in just a few short years overtaken Lyft. 00:14:29.360 --> 00:14:31.759 And it's gonna continue to go in this direction. 00:14:31.840 --> 00:14:34.399 It's almost like feels like in an inevitability. 00:14:34.559 --> 00:14:36.240 And I think that's just the power of technology. 00:14:36.399 --> 00:14:50.159 Like sometimes technologies are so good that what they call in the legal, the legal world res ipsa, which means it's so it's evidence that is so clear that it's self-evident, doesn't even require the the attorney to to explain it. 00:14:50.240 --> 00:14:56.639 You know, it's clear to me that's that cars will be self-driving by and large, especially for the rideshare use case. 00:14:56.799 --> 00:15:03.600 Now, the hard question is when how long does it take before driving your own vehicle becomes illegal? 00:15:04.000 --> 00:15:06.399 Because eventually that'll become the case too. 00:15:06.639 --> 00:15:10.080 And it will start out sort of like it'll be like this. 00:15:10.240 --> 00:15:19.919 If you get into a car accident and you were and it turns out that you were driving yourself and you weren't letting the computer guidance system manage it for you, you're liable. 00:15:20.080 --> 00:15:21.120 It'll be like being drunk. 00:15:21.279 --> 00:15:26.720 If you get into an accident, you're found out to be drunk in the United States, it doesn't matter whose fault it is, it's your fault. 00:15:26.960 --> 00:15:34.799 If it turns out that you are self-driving when you had an autonomy system that you disabled, it's going to be your fault if there's any accident. 00:15:34.960 --> 00:15:36.799 And then eventually it'll go even farther. 00:15:36.879 --> 00:15:38.639 It will be, hey, it's just not legal. 00:15:38.799 --> 00:15:41.919 Just like how it's not legal to ride a horse on the highway. 00:15:42.159 --> 00:15:48.320 People would there was a period of time where cars and horses coexisted, but that that period that time is no more. 00:15:48.480 --> 00:16:04.960 And so there will come a time, and I think it will be probably 30 years away, maybe more, where you can't really use a manual car, like a self-driven car on public roads because of the sort of disproportionate risk it poses to people when autonomous vehicles are 100% safe. 00:16:05.120 --> 00:16:08.720 And I think eventually we'll see the same thing across a lot of different industries, right? 00:16:08.879 --> 00:16:13.840 It might, you might look at it and say, How could you ever have a human doing invoice reconciliation? 00:16:14.080 --> 00:16:20.000 It's so prone to error when the system itself is provably perfect or near perfect, that kind of thing. 00:16:20.159 --> 00:16:23.200 So I and I think that's a and I think that then I'm an optimist. 00:16:23.279 --> 00:16:25.600 I think people often ask, well, what am I going to do? 00:16:25.840 --> 00:16:34.960 I mean, societally, like my hope is that we can find ways to free people from the the work that they don't like to do and help them find things that they do like to do. 00:16:35.039 --> 00:16:45.440 And if we're really creating you, you know, we're creating wealth for people through, you know, automation, you know, making sure that we just distribute it in an appropriate way and folks are are made whole, right? 00:16:45.679 --> 00:16:55.120 You know, if we if we're gonna say, hey, we're gonna gain the sufficiency, but I I think that I my hope is that people are we're gonna we're heading toward a more utopian society in you know five, ten years. 00:16:55.360 --> 00:16:56.639 Yeah, I mean, let's hope so. 00:16:56.720 --> 00:17:01.440 I I think it was interesting the graph you showed between Lyft and and Waymo and their kind of intersection. 00:17:01.679 --> 00:17:05.039 Do you think that's almost what's gonna happen with kind of procurement technology? 00:17:05.200 --> 00:17:11.839 Because I imagine I could be wrong, but you know, when Waymo first launched and you know, people, the members of the public could use it. 00:17:11.920 --> 00:17:19.599 I imagine a lot of people booked one because it was like this is new, it's almost, you know, I I don't mean to use this word disrespectfully, but like a gimmick, isn't it? 00:17:19.839 --> 00:17:21.839 It's a self-dreating, let's give it a go. 00:17:22.160 --> 00:17:34.240 And kind of the more it's used and the more positive experience people have, they tell their friends and their friends tell their friends, and you know, it has that steady ride rise in which it now has kind of overtaken Lyft. 00:17:34.319 --> 00:17:53.119 Is it a case that you know when people can harness AI solutions, you know, at to start with, it's maybe seen as, oh, we can get the tools to do this, but you know, once people kind of see actual tangible benefits that that will rise and

and kind of be human-led applications will kind of form or will have those intersections again. 00:17:53.440 --> 00:18:00.640 Yeah, I think that I think that like as you can imagine, things will happen faster where people have agency. 00:18:00.720 --> 00:18:05.599 And so I have choice about whether I want to take Uber, Lyft, or Waymo. 00:18:05.839 --> 00:18:08.640 And so I can make that switch in one day. 00:18:08.799 --> 00:18:17.200 And in fact, in San Francisco, I made the decision recently to start using Tesla Robo Taxi because, well, one, it's clearly subsidized. 00:18:17.440 --> 00:18:22.880 There's no way it's like$3,$5 for that kind of ride, but it's a beta program. 00:18:23.039 --> 00:18:27.119 And I'm an ex-Tesla employee also, so I was curious to try it out. 00:18:27.200 --> 00:18:28.160 And I was really impressed. 00:18:28.240 --> 00:18:31.839 And so now I've already switched off of Waymo to Tesla Robotaxi. 00:18:32.000 --> 00:18:33.599 So the switching cost is very low. 00:18:33.759 --> 00:18:40.160 And so in applications where people can make the shifts quickly, I think you're gonna see things will happen faster. 00:18:40.400 --> 00:18:43.119 In enterprises, they're a lot slower to change their tools. 00:18:43.279 --> 00:18:58.000 I mean, I think every company that I've met with over the last decade has uh has been in, has either been about to start, is in the middle of, or is still struggling with finishing uh an ERP transformation from a certain three-letter company. 00:18:58.240 --> 00:19:02.000 And so, you know, sometimes these things just take a really long time. 00:19:02.400 --> 00:19:13.599 And I think that the organizations where they have the right incentives in place for the executives and the employees make really good decisions, but companies are not so they're not, they're very complex organizations, right? 00:19:13.680 --> 00:19:14.559 They're complex beings. 00:19:14.640 --> 00:19:19.920 Everyone's got both what's best for the company and what's best for for me and my world, right? 00:19:20.079 --> 00:19:25.279 And I think those kinds of conflicts, when they come into conflict, they they lead to the wrong technology outcomes. 00:19:25.440 --> 00:19:32.240 You know, I have this other, this other philosophy that I've I've shared, and and I actually, it's sort of funny as we're talking. 00:19:32.400 --> 00:19:38.079 I pulled up a talk that I gave at ISM uh last year as part of trying to answer the questions. 00:19:38.240 --> 00:19:42.480 I wasn't prepared for your questionnaire and you asked me, what were my predictions last year? 00:19:42.640 --> 00:19:48.480 And I was thinking, okay, let me just look at the predictions and see if this is how many of these came true. 00:19:48.720 --> 00:19:51.039 The first one was AI won't take your job. 00:19:51.200 --> 00:19:53.039 Someone who now knows how to use it will. 00:19:53.200 --> 00:19:54.079 That is still the case. 00:19:54.240 --> 00:19:55.839 I mean, I that's exactly what

I'm saying. 00:19:56.000 --> 00:20:00.960 Someone who is able to use AI, who is very serious about it, is is going to find a lot of benefit. 00:20:01.279 --> 00:20:06.799 They talk about an intro to AI and then agentic, which was at that time very, very new. 00:20:06.960 --> 00:20:09.920 And and what is the difference between Gen AI and agentic AI? 00:20:10.000 --> 00:20:12.000 Well, Gen AI, it's reactive. 00:20:12.079 --> 00:20:14.640 You do a prompt input and then it generates an output. 00:20:14.880 --> 00:20:19.039 Agentic is you put a prompt input and you get actions as your output. 00:20:19.200 --> 00:20:21.759 So it's deciding, it's doing things, it's perceiving. 00:20:21.920 --> 00:20:33.119 Now, LLMs are like a very common foundation for both generating text, pictures, code, music, and generating actions when they're in a framework that requests actions. 00:20:33.359 --> 00:20:38.480 Now, the questions of where those things live, I think that the philosophy is still the same. 00:20:38.640 --> 00:20:42.400 I guess the question is just where have they found real practice? 00:20:42.640 --> 00:20:50.720 And I think that so far there's been this idea, and this is what I wanted to really show you, and what I call the AI capability overhang. 00:20:50.880 --> 00:20:54.880 And so this is the ability of AI technologies, and this is over time. 00:20:55.039 --> 00:21:05.440 And what I found is that there, that AI progress as a technology, at least over the last year at that point, had made a lot of progress, and then AI products themselves were very disappointing. 00:21:05.680 --> 00:21:13.680 And I think that this leads to a it leads to a lot of confusion because people will say, okay, I can do this in Chat GBT. 00:21:13.839 --> 00:21:18.799 Why can't we do this at scale in our enterprise procurement solution? 00:21:19.039 --> 00:21:24.799 And then they'll try to get some kind of AI solution, and then they'll find that it it sort of overpromises, underdelivers. 00:21:25.039 --> 00:21:38.079 Now, the fear that I have is that in a year, and I've felt this shift a little bit in a small way, people feeling concerned that AI has overpromised and underdelivered, and now they'll look at it and say, okay, AI is a false promise. 00:21:38.240 --> 00:21:48.319 And I think agentic has been one of the biggest I don't want to say it's been an overpromise because there are it's like one of these things where there are good actors and there are bad actors, right? 00:21:48.400 --> 00:21:59.039 Like when I go to conferences, a hundred percent of vendors talk about in their booth, on their panels, AI and agentic AI for procurement. 00:21:59.279 --> 00:21:59.759 I was gonna ask. 00:22:00.079 --> 00:22:07.279 You about this later on, you know, I was going to bring out the fact, you know, every time we speak, you, you know, you discuss your problem with this term AI native. 00:22:07.519 --> 00:22:14.319 You know, um I want you to kind of continue by also was wondering if you could tell us, you know, how do we hold this to account this year? 00:22:14.480 --> 00:22:22.079 How does 2026 be that people can can kind of bash can kind of hit back at someone who says they're I native and be like, well, prove it then. 00:22:22.319 --> 00:22:26.400 I think that the easiest thing is to is to see it in practice, right? 00:22:26.559 --> 00:22:33.599 I think that that's that's the that is the that is the you know, if you want to test someone's metal, that I'm you know, I'm from New York, I grew up in New York. 00:22:33.920 --> 00:22:37.519 We're from what we call the show me culture, and so I want to just see it. 00:22:37.680 --> 00:22:44.079 And you know, it's very easy for someone to put together a presentation or a deck uh or to talk a really big game. 00:22:44.240 --> 00:22:48.400 And then the hard thing is, and it's even easy to do like a one-op demo. 00:22:48.640 --> 00:22:52.160 What's really hard is when you're like, hey, I don't want you to be able to prepare. 00:22:52.319 --> 00:22:57.839 I want to just give you my data right now, and I want to see what your system is capable of doing. 00:22:58.000 --> 00:23:00.480 And I think that's where a lot of tools sort of flounder. 00:23:00.799 --> 00:23:10.000 And I think I've been sort of surprised there have been even companies that have come out, like there was a company that announced this big like 50 agent like blitz. 00:23:10.079 --> 00:23:14.880 They were just like, we're releasing 50 something different agents that do all sorts of different things. 00:23:15.039 --> 00:23:19.279 It's a great marketing picture, but then none of them were functionally good. 00:23:19.519 --> 00:23:33.759 And then people are just really struggled to like it's like, and the problem is if you like say we're gonna do this and then it's like not working well, then people are gonna say, okay, let's put pause on implementing technology in our business for like two years because we had such a bad experience. 00:23:33.920 --> 00:23:42.240 So the first to market people really do have a lot of responsibility to determine, you know, what's the perception for a much longer period of time, you know? 00:23:42.480 --> 00:23:47.519 And um, you obviously you pulled up that the presentation you did at kind of the start of um 2025. 00:23:47.759 --> 00:24:04.240 You mentioned about the whole germinative versus agentic, because it kind of felt towards the end of 24 and kind of how into 25 kind of agentic started to pick up, but mostly kind of 2024, you were hearing a lot about you know genative AI, gen AI. 00:24:04.400 --> 00:24:10.400 And obviously, you know, last year Ghana said it kind of ended to the kind of trout of disillusionment. 00:24:10.559 --> 00:24:14.160 You know, why did you think that Gen AI kind of ended up going that way? 00:24:14.400 --> 00:24:18.400 Well, I don't think that Gen AI is in the trout trough of disillusionment. 00:24:18.640 --> 00:24:33.119 I think that the difference between

Gen AI and agentic AI is that Gen AI is reactive, and so you have you input a prompt and you output, and what the system outputs is a generated, a gen, a generative output. 00:24:33.200 --> 00:24:37.440 And that generative output could be music, code, whatever it might be. 00:24:37.599 --> 00:24:44.480 And so what Gen AI requires is that the user understands their use case and that they go and they seek it out, right? 00:24:44.559 --> 00:24:48.240 I have to input a prompt in order for the system to do something for me. 00:24:48.400 --> 00:24:55.599 With an agentic system, what you're saying is I want the user to input a prompt, and this could be a prompt that happens at the time of design. 00:24:55.680 --> 00:25:02.000 It doesn't have to be the individual user, and the output is going to be actions like an RPA but on steroids. 00:25:02.160 --> 00:25:03.519 That's the best way to think about it. 00:25:03.680 --> 00:25:16.720 So I think that in organizations where they're very tech forward and like high likelihood of adopting technology, what we see is that people are, you can just tell because people will in their own work be using Chat GPT. 00:25:16.880 --> 00:25:22.559 They'll be using it to like write emails, to run analysis, to help them brainstorm on strategy. 00:25:22.960 --> 00:25:27.920 If they're if they are using it in their own life, I think it's very likely they bring it into their business. 00:25:28.079 --> 00:25:35.200 But then at the system level, uh, I think agentic is the place where you really see a lot of I would say like workflow type value. 00:25:35.440 --> 00:25:37.759 I wouldn't say we're in a trough of disillusionment. 00:25:37.920 --> 00:25:49.759 I would say that I think that's the fear that we're headed towards is that people overpromise, underdeliver, and then in 12 months from now, people look at it and say, okay, we have to stop focusing on AI. 00:25:49.920 --> 00:25:52.480 We have to focus on our core business problems. 00:25:52.559 --> 00:26:12.319 And the reason that I feel that this could come, especially now, is because when I have a conversation with a chief procurement officer or VP at any of the Fortune 500 global, global 2000 businesses we meet with, uh, they will often say, you know, we have uh our CEO is tasking us with developing an AI strategy. 00:26:12.480 --> 00:26:14.960 And so what do you guys offer in terms of AI? 00:26:15.200 --> 00:26:17.039 And I say, well, it depends. 00:26:17.119 --> 00:26:18.960 Like, what do you guys need solved? 00:26:19.119 --> 00:26:19.359 Right. 00:26:19.599 --> 00:26:21.200 And they're like, we don't really know. 00:26:21.359 --> 00:26:28.799 And so that's actually like a big disconnect because technology looking for a problem like is never never yields strong results. 00:26:28.880 --> 00:26:34.480 It always has to start with what's the problem we're trying to solve, and let's work backwards to find the right technology. 00:26:34.720 --> 00:26:43.759 The point of buying an AI solution or implementing AI is not just so you can say we have AI, it's it's to try to solve a real business problem. 00:26:43.920 --> 00:26:49.680 And so I think people have gotten away a little bit from the question of, hey, what are the actual objectives we're trying to achieve? 00:26:49.839 --> 00:26:54.160 And then thinking about AI as one of many tools that they can use to solve it. 00:26:54.319 --> 00:26:54.640 Right. 00:26:54.720 --> 00:26:57.200 And now sometimes AI is not the right answer. 00:26:57.279 --> 00:27:06.160 You know, if I think about payment systems, I honestly just don't want AI or generative anything near them because you know what, a payment system is a deterministic system. 00:27:06.400 --> 00:27:08.240 I don't want there to be agents. 00:27:08.319 --> 00:27:10.319 I don't want there to be literal agency. 00:27:10.559 --> 00:27:13.119 No decision making should happen by the system. 00:27:13.279 --> 00:27:16.880 That routing number, that bank account number gets sent that amount of money. 00:27:17.039 --> 00:27:18.640 It's ironclad, you know? 00:27:18.880 --> 00:27:24.640 So I think that I think that actually it's not, it is not a blunt force instrument that people should be using all over the place. 00:27:24.799 --> 00:27:25.759 But I think it's a tough thing. 00:27:25.839 --> 00:27:30.559 I mean, what we have is that there are boards of directors who who say AI is important. 00:27:31.279 --> 00:27:34.720 I have to ask our CEO, what's the what's your AI strategy? 00:27:34.960 --> 00:27:39.440 They then go to their VPs and lieutenants and they say, What are you doing with AI today? 00:27:39.599 --> 00:27:41.599 And then they don't have really good answers for it. 00:27:41.680 --> 00:27:47.839 So they're going and meeting vendors and they're asking, well, what do you do with AI that maybe we can plug into our environment? 00:27:47.920 --> 00:27:50.559 But that's actually the reverse of how it's supposed to be. 00:27:50.799 --> 00:27:59.119 You know, you don't go to the doctor and say, like, tell me, you know, tell me if I'm what medication should I take before they diagnose you with an illness, right? 00:27:59.279 --> 00:28:09.920 And so I think that's that's where I start to worry, okay, if people are buying AI solutions, but they don't have a clear problem in mind yet, then they're not going to see the ROI or outcome they expect. 00:28:10.000 --> 00:28:14.240 And then eventually what's going to happen is, like many things, they cool off, right? 00:28:14.400 --> 00:28:16.400 You know, there's there's a hype cycle, it's very clear. 00:28:16.480 --> 00:28:18.480 You know, there's a step change in technology. 00:28:18.640 --> 00:28:20.960 People are surprised by how performant it is. 00:28:21.039 --> 00:28:22.799 There's a ton of hype built around it. 00:28:22.960 --> 00:28:23.599 Everyone wants it. 00:28:23.680 --> 00:28:25.920 They're very disappointed by it and it doesn't provide. 00:28:26.079 --> 00:28:30.240 And

then there's like a, you know, a two to 10 year winter, right? 00:28:30.319 --> 00:28:32.160 Uh, we saw this with self-driving. 00:28:32.240 --> 00:28:33.200 We saw this with VR. 00:28:33.279 --> 00:28:39.440 I mean, VR, if you remember, Oculus Rift was acquired by uh Facebook, what was it, like 10 years ago? 00:28:39.599 --> 00:28:41.279 Everyone was super hyped on it. 00:28:41.440 --> 00:28:45.599 And then it kind of languished with like some niche of gaming. 00:28:45.759 --> 00:28:54.720 And then now, only now, like sort of you're starting to see like Apple Vision Pro and the meta ray bands cresting back into the mainstream. 00:28:54.960 --> 00:28:59.920 But Google Glass was, I remember even when I was in college, Google Glass was already like a thing, you know. 00:29:00.079 --> 00:29:02.880 You know, I think these things do go in cycles, they really do. 00:29:03.039 --> 00:29:15.440 And so I I think it's I think it's inevitable that when there is this much hype, eventually there's going to be disappointment because I just know that 80% of the vendors out there are not are not really what they what they promote themselves as. 00:29:15.759 --> 00:29:27.839 Yeah, I I mean I wanted to talk to you about, you know, how do you make this year the year that kind of you know AI, you know, delivers on its promises, but just a more case of people need to know why they're getting AI involved. 00:29:28.079 --> 00:29:29.279 Is it just as simple as that? 00:29:29.359 --> 00:29:33.839 Or do you think that you know AI can do a bit more to kind of you know sell itself? 00:29:34.160 --> 00:29:40.880 Well, I don't think I think AI has already sold itself as a con as if AI is a thing, but it sold itself extremely well. 00:29:41.039 --> 00:29:44.400 I think that what I would advise people is always to start with the problem. 00:29:44.559 --> 00:29:46.880 You know, what are you trying to get done in your business? 00:29:47.039 --> 00:29:48.880 What are your major initiatives? 00:29:49.119 --> 00:29:51.440 And then think about how do we get it done. 00:29:51.680 --> 00:29:54.480 And I would always start with like there's three steps in the process. 00:29:54.640 --> 00:30:02.480 One is do it manually, then build a process that people can do it, and then build a process that automates the people from doing it, right? 00:30:02.640 --> 00:30:03.599 From needing to do it. 00:30:03.759 --> 00:30:08.079 And so, you know, there's no, there's not there's not magic necessarily there. 00:30:08.319 --> 00:30:11.279 If a person can't do it, AI likely can't do it today. 00:30:11.440 --> 00:30:13.279 Do you remember the Segway scooter? 00:30:13.680 --> 00:30:14.480 Do you remember that? 00:30:14.559 --> 00:30:15.519 The two-wheeled thing. 00:30:15.759 --> 00:30:19.359 So this was like the hottest thing back in the early 2000s. 00:30:19.519 --> 00:30:25.039 Dean came in and invented this thing, and there were all these great tech tech innovators who invested in it. 00:30:25.200 -->

00:30:34.079 Steve Jobs and founder of uh the founder of Sequoia, and uh and folks like John Dore from Kleiner Perkins. 00:30:34.240 --> 00:30:37.279 And so uh everyone said this is gonna be replacing walking. 00:30:37.359 --> 00:30:39.920 It's like the most incredible technology, blah, blah, blah. 00:30:40.079 --> 00:30:49.119 And it comes out and it's like a$4,000 self-balancing scooter, and people thought it was very cool, but nobody suffers from that. 00:30:49.279 --> 00:30:57.200 Very few people suffered from whatever non-existent pain point they were solving enough to justify buying it for themselves, right? 00:30:57.440 --> 00:31:10.720 And so, you know, again, I just think that if you start with a cool technology and you don't have a clear problem in mind, then you're gonna land yourself with, you know, you're gonna land yourself in in uh in what I would call like tech purgatory. 00:31:10.880 --> 00:31:18.240 And if you start with a really clear problem and then you find the right technology to solve it, then you will create an immense amount of value. 00:31:18.480 --> 00:31:25.359 Another example in the cryptocurrency space, you know, what a what a huge amount of hype cycles that have happened over the years. 00:31:25.599 --> 00:31:29.440 People are always asking the question, man, what can we do with blockchain? 00:31:29.519 --> 00:31:31.599 You know, what can we do with with crypto? 00:31:31.680 --> 00:31:33.200 It's such a cool technology. 00:31:33.359 --> 00:31:52.799 You know, maybe we can put it into trading cards or we can put it into NFTs, we can put it into uh, you know, we can put it into uh maybe, maybe your uh your milk on your shelf will tell you how cold or warm it was based on like a fully authenticated blockchain temperature sensor. 00:31:52.880 --> 00:31:53.440 You know what I mean? 00:31:53.519 --> 00:32:01.680 There's a million things that people try to do with blockchain, but when I think about it, what's the best use case for for blockchain? 00:32:01.759 --> 00:32:02.559 It's currency. 00:32:02.640 --> 00:32:05.440 It's a distributed transaction ledger. 00:32:05.599 --> 00:32:11.359 There Bitcoin is the best, best use case of of blockchain, in my opinion. 00:32:11.519 --> 00:32:15.680 And I think that all the other ones have been very difficult to prove out, right? 00:32:15.920 --> 00:32:18.880 And so I think it always has to start with, and and that's a real pain point, right? 00:32:19.039 --> 00:32:31.680 People in in countries with a lot with a lot less of a stable currency wanting a store of value that they can transact, not have to deal with their local inflation, be able to send payments agnostic to borders. 00:32:31.759 --> 00:32:34.319 Uh that is a real set of problems that people face. 00:32:34.400 --> 00:32:36.400 Actually, ironically, less in the United States. 00:32:36.559 --> 00:32:42.640 Like we don't we need blocked, we need Bitcoin the least, ironically, of almost any people on earth, I'd say. 00:32:42.960 --> 00:32:45.279 But uh it's a real problem that they're solving. 00:32:45.359 --> 00:32:47.359 And that's why Bitcoin took fire. 00:32:47.440 --> 00:32:50.400 And then people started to think, well, what else could we do with blockchain? 00:32:50.640 --> 00:32:53.680 And the answer so far has been like very little, I think. 00:32:53.759 --> 00:32:54.000 You know? 00:32:54.319 --> 00:33:02.960 I mean, what's it like leading a company like Light Source in a time such as this where you know technology is evolving at such a kind of rapid, rapid speed? 00:33:03.119 --> 00:33:08.079 I mean, obviously nothing as cool as the Segway scooter, but yeah, really exciting time. 00:33:08.240 --> 00:33:12.160 And, you know, I'm sure this 2026 will be kind of more of the same. 00:33:12.480 --> 00:33:14.400 Yeah, it is it is interesting. 00:33:14.480 --> 00:33:17.200 It is a very fun time to be a company at our scale. 00:33:17.440 --> 00:33:29.680 There is, I feel like so lucky because you know, if I were to sort of design like my dream company, right, to plop onto the earth in 2026, it would be something like Light Source. 00:33:29.839 --> 00:33:57.839 I'd be like, all right, I want to bring together a really talent dense, not super small, but a moderately sized but high growth team with a huge amount of talent density focusing on an enormously commercially relevant problem, which for the most part has been direct materials, which is the, you know, I would say the big unsung here of the procurement universe that we have addressed very directly head-on, like the elephant in the room that makes up 80% of spend that nobody seems to be looking at. 00:33:58.000 --> 00:34:09.199 I would also love if, you know, it would be amazing is if every single competitor that was built 20 years ago, their technology became irrelevant in like a matter of two years. 00:34:09.440 --> 00:34:10.880 Well, that would be amazing too. 00:34:10.960 --> 00:34:18.559 And that's what you know, the AI revolution is that things that were developed 10 years ago look ancient by comparison to today's standards. 00:34:19.119 --> 00:34:21.760 The advantage that the incumbents have is sort of three things. 00:34:21.920 --> 00:34:26.639 One, they have they have an already built-up technology base, which is becoming dated. 00:34:26.800 --> 00:34:31.119 They have resources and they have a lot of people and they have a reputation. 00:34:31.280 --> 00:34:50.000 And so resources and people, it used to be a huge advantage, but I believe now that a team of 100 can do just as much or more than a team of 10,000 if they're really given a unified direction that they're all rowing towards, and they have the kinds of AI tools that they need to extend their capabilities. 00:34:50.239 --> 00:34:59.440 And then the final thing is that AI has been a huge tailwind for us as a business, not just in our product, not just in the way that we work internally, but also externally. 00:34:59.599 --> 00:35:06.079

If you talk to any uh you know CPO today, they're being tasked, they're being tasked to bring AI into their environment. 00:35:06.239 --> 00:35:15.440 And the answer they can't come back with is hey, we're we're gonna stick, we're doing uh something more with the legacy provider, because that is it's clear that that's not the innovative solution. 00:35:15.599 --> 00:35:23.440 And so folks are interested more than ever in meeting, you know, Silicon Valley-based uh high growth companies because they know that there's something there, right? 00:35:23.679 --> 00:35:27.599 And so, you know, I feel I feel very excited about the place that we're in. 00:35:27.679 --> 00:35:31.519 I'm very excited about the time and the world that we happen to exist in. 00:35:31.840 --> 00:35:35.119 But the thing that makes it hard is that the world is changing so quickly. 00:35:35.280 --> 00:35:38.960 I, as a leader, need to stay extremely close to the latest developments. 00:35:39.119 --> 00:35:48.719 I've not found myself needing to read research papers at any point in my career since I left grad school until now to really understand what's happening in the space. 00:35:48.880 --> 00:35:51.840 I also find myself using AI tools all over the place. 00:35:52.000 --> 00:36:09.440 Like I want to use Replit, I want to use Figma, I want to use Clawed Code because these things are are other, there are other people who are right now grappling, very smart people in our in our one mile radius of us, who are grappling with the question of how can we make use of this incredible technology suite. 00:36:09.599 --> 00:36:19.440 And it's interesting to see the ideas that they've brought to their product and what are the ones that we can borrow or exceed or things that would work better actually in the context of procurement than coding. 00:36:19.679 --> 00:36:28.159 Now, the challenge for our team is that this means that we have a very clear North Star, but navigation that is changing every month, right? 00:36:28.320 --> 00:36:31.440 And so that's always the balance is you know, it's like playing a game with soccer. 00:36:31.519 --> 00:36:34.079 You want you tell the team, hey, here's where the goal is. 00:36:34.239 --> 00:36:41.679 But then when the ball gets under the field, we have to be able to move around, run slightly different plays, and be agile. 00:36:41.840 --> 00:36:46.639 And I think that's the way that it it always is for startups, but I feel especially like that now. 00:36:46.800 --> 00:36:49.039 You have to be prepared to change your strategy. 00:36:49.280 --> 00:36:50.400 Yeah, of course. 00:36:50.559 --> 00:36:55.920 And you know, I think if you want to be successful in the AR space, you have to be ready, you know. 00:36:56.000 --> 00:36:58.639 And I think your approach is is probably the best to have. 00:36:58.800 --> 00:37:05.280 You know, you're you're ready and open to listen and learn from other people and you know, take advantage of of all the tools that you have. 00:37:05.440 --> 00:37:07.199 Like you said, you know, it's there. 00:37:07.599 --> 00:37:12.480 I think everyone should be kind of using the tools in their life to you know to their to their own benefits. 00:37:12.639 --> 00:37:19.599 I'm also glad that you mentioned kind of direct materials because obviously earlier this year you debuted your manifesto. 00:37:19.840 --> 00:37:24.480 That was one of my highlights of 2025, being able to pick up that that little book. 00:37:24.719 --> 00:37:33.440 So, you know, for anyone who who perhaps doesn't know, could you give us the cliff notes version of of what that was, how it came about, and and kind of your kind of core message in that? 00:37:33.679 --> 00:37:34.159 Yeah, of course. 00:37:34.239 --> 00:37:36.719 I actually have a copy of it here with and of course. 00:37:37.039 --> 00:37:37.760 Yeah, of course. 00:37:37.920 --> 00:37:40.320 And so this is a Spectre Scale manifesto. 00:37:40.480 --> 00:37:43.440 It is the a book that actually we published earlier this year. 00:37:43.679 --> 00:37:45.519 I am very proud I wrote it myself. 00:37:45.599 --> 00:37:52.000 Uh, and it's all about the the uh the history of procurement technology and how we ended up where we are. 00:37:52.159 --> 00:38:00.079 And really specifically, like the core of the manifesto is this idea that direct materials has been underinvested for decades. 00:38:00.239 --> 00:38:07.199 And that when we look around at the market environment, the vast majority of solutions are focused on indirect XYZ. 00:38:07.280 --> 00:38:19.199 And even the construct of S to P, uh source to pay, being the canonical tech landscape for procurement, ends with P2P, which is the procure-to-pay system, which is indirect, not the ERP. 00:38:19.360 --> 00:38:34.800 And so there's an entirely different set of technology solutions that you actually need that are that bear a very, very scarce resemblance to the indirect world when you really start to think from a first principles basis about what those teams need when you're thinking about direct materials. 00:38:34.880 --> 00:38:42.400 And it's about pulling from the engineering systems, PLMs, uh aligning on something we call specs or specifications. 00:38:42.480 --> 00:39:04.079 Now, this could be CAD drawings or could be CAST numbers for a chemicals business, and then running through the full supply lifecycle management eventually to payment, and then even post-payment managing things like quality and index adjustments and risk, uh, things again that are not very well managed in the P2P systems today, if they're touched at all. 00:39:04.239 --> 00:39:09.039 And so we really wrote the spec to scale manifesto as a way, it's sort of broken into two parts. 00:39:09.280 --> 00:39:11.519 One is looking back, and the other is looking forward. 00:39:11.679 --> 00:39:13.519 And how do we end up in this position? 00:39:13.760 --> 00:39:19.440 What's the history of procurement technology that we end up in this

sort of weird local minimum? 00:39:19.599 --> 00:39:24.400 And then a philosophy about spec to scale, which is what is the ideal set of processes? 00:39:24.559 --> 00:39:32.480 Forget about the technology, but what are the actual things and activities that people are doing in direct materials that are orphaned today? 00:39:32.639 --> 00:39:36.559 And then thinking about, okay, what is the right, the most ideal solution? 00:39:36.719 --> 00:39:43.760 And we try to build light source in that vision to check all of those boxes so it fits like a glove for direct materials teams. 00:39:43.920 --> 00:39:49.119 And of course, there's a little plug for us at the end, but I I really believe that this is a great, I think it's a great read. 00:39:49.199 --> 00:39:52.639 It's really fun, it's actually well informed, very honest. 00:39:52.800 --> 00:39:56.559 Uh, I'm pulling no punches, it's a little bit controversial at certain points. 00:39:56.719 --> 00:40:05.280 And we we, as you know, DPW was it was like it felt like the the manifesto city over there was just like plastered with our messaging. 00:40:05.360 --> 00:40:06.639 So it was just really fun. 00:40:06.719 --> 00:40:09.599 And I'm gonna try to get a hardcover version published. 00:40:09.679 --> 00:40:25.039 Right now it's all soft cover, so no, I I think it's great, and you know, perhaps unfairly bakery and gets labeled with this whole boring tag and what so it's good to have a voice like yourself who's you know not afraid to you know throw some punches and and you know stir things up a little bit. 00:40:25.360 --> 00:40:33.679 So I'm sure it will be seen at kind of events across 2026 and you know, all hopefully see that kind of hardcover edition come. 00:40:34.000 --> 00:40:37.119 I know I'll certainly pre-order one if they're available to buy. 00:40:37.519 --> 00:40:38.159 Absolutely. 00:40:38.559 --> 00:40:39.039 Absolutely. 00:40:39.199 --> 00:40:39.360 Yeah. 00:40:39.519 --> 00:40:59.920 So uh kind of looking back at uh at the last 12 months, you know, one thing I seem to see, and it could just be a nature of my job, is there was a lot of acquisitions, larger companies buying up startups, buying up smaller companies who perhaps come to market with an interesting kind of AI tool, and they've kind of acquired it to kind of boost their own AI capabilities. 00:41:00.320 --> 00:41:04.800 From your kind of CEO perspective, what is the thinking behind that? 00:41:04.880 --> 00:41:17.760 Um why a a a larger company would perhaps swallow up those startups who come to market with a with an interesting AI tool rather than look to either build on it, s build their own, improve it. 00:41:17.920 --> 00:41:23.440 Why is that a case that you know if you're a CEO, what what do you think the kind of thinking is behind a move like that? 00:41:23.920 --> 00:41:26.719 Like a CEO at a at a at a company or at another. 00:41:27.119 --> 00:41:37.760 Yeah, obviously you you yourself is uh I was wondering if you could just give us your perspective perspective as you obviously are a light source CEO, perhaps you think kind of differently to myself, you know. 00:41:37.840 --> 00:41:44.159 I just wanted to know what is the thinking behind it, or you know, you're more and more kind of uh welcome to answer it as yourself, Spencer. 00:41:44.480 --> 00:41:51.760 I mean, are are you talking from the perspective of someone who's like the CEO of a Fortune 500 company or are you asking for like a like the CEO of SAP, for example? 00:41:52.079 --> 00:42:06.480 Yeah, you know, from uh a procurement company, a procurement technology company who will, you know, look to acquire a startup who have come to market with a with an interesting AI tool rather than kind of use it as is is inspiration. 00:42:06.719 --> 00:42:18.079 Do you think it's just a case of they've got people in the door already who who know how to make these tools, let's get them in, or is it just a case of so people will acquire companies for you know one of three reasons. 00:42:18.400 --> 00:42:21.280 One is they they acquire the customers. 00:42:21.519 --> 00:42:32.320 Like if you think about something like, you know, I don't want to say any specific companies, but let's just say uh, I don't know, let's think about like you know, Surtuo as an example. 00:42:32.480 --> 00:42:40.639 You know, I think that there's an I think that there's a way to look at that acquisition, which is, hey, they're gonna buy the buy the product and they'll enhance their platform. 00:42:40.800 --> 00:42:50.480 There's another way to think about it, which is they'll buy the product and then they'll cross-sell, uh, they'll cross-sell, you know, the broader Coupa Suite to existing Surtuo customers. 00:42:50.559 --> 00:42:52.880 And that might be the place where more of the value is. 00:42:53.119 --> 00:42:55.840 The third reason for acquisition is to acquire talent. 00:42:55.920 --> 00:42:57.199 They call this Aqua Hire. 00:42:57.440 --> 00:43:05.360 I am of the belief personally, and for a reference for everybody who's anyone who's listening, Light Source is absolutely 0% for sale. 00:43:05.519 --> 00:43:12.239 Uh, we are eventually going to be on an acquisition hunt, but uh, but not for anything in particular. 00:43:12.400 --> 00:43:16.239 Uh, but we're but we're not we're not building ourselves as an acquisition target. 00:43:16.400 --> 00:43:18.480 I mean, we've we've grown so quickly. 00:43:18.639 --> 00:43:24.239 We're so lucky to be backed by incredible investors, and so we're we're the opposite of a need for cash. 00:43:24.480 --> 00:43:34.719 But I think that if you build a company with the intention of being acquired, I think it will never happen because you will make short-term bets that don't have long-term value. 00:43:35.280 --> 00:43:43.440 And you also will typically have like an acquirer in mind, and you don't actually know if their timing lines up with your timing. 00:43:43.599 --> 00:43:45.039 You can't control

them. 00:43:45.199 --> 00:43:47.199 And then you also don't have a lot of leverage. 00:43:47.360 --> 00:43:55.679 For me, what my dream is for our company is to build the definitive procurement operating system for direct materials and to bring that company public. 00:43:55.760 --> 00:43:59.760 And I believe that that company could be as large, if not larger, than a business. 00:44:00.239 --> 00:44:03.440 Like a Salesforce, which is a$250 billion business. 00:44:03.679 --> 00:44:12.960 Now, when I think about like will like acquisitions in general, like, you know, when I think about we've had actually two opportunities to acquire other businesses. 00:44:13.199 --> 00:44:19.440 And um, one of them we considered and one of them we really didn't. 00:44:19.599 --> 00:44:26.960 But it's funny, like in both instances, when I talked to our CTO and said, hey, they have this product, we can acquire it. 00:44:27.280 --> 00:44:33.360 His view was, well, actually, it's more work than it would be for us to build it from the ground up. 00:44:33.599 --> 00:44:39.440 So someone has to have really, really impressive technology in order for it to be something that you want to acquire. 00:44:39.679 --> 00:44:46.719 Because chances are it's built in a different code base, it's built in their own languages, it's not connected through to your systems. 00:44:46.960 --> 00:44:57.920 Companies, we know certain companies in the procurement space have been built up through acquisition and they'll wrap the business with like a label that it's an all-in-one integrated solution. 00:44:58.000 --> 00:45:06.719 But then when you get behind the nuts and bolts of it, it's like you have to click a button and enter effectively a completely different environment in order to use these other functionalities. 00:45:06.880 --> 00:45:09.760 And so it ends up being a very disjointed situation. 00:45:10.000 --> 00:45:14.239 If somebody were to have a piece of IP that's like pharmaceuticals, right? 00:45:14.400 --> 00:45:17.840 Like where it's like, hey, we don't actually care anything about the operations. 00:45:18.000 --> 00:45:22.400 We just know that the molecule is patented, we can make it, and it's been tried. 00:45:22.559 --> 00:45:24.639 It's a much more clear acquisition. 00:45:24.800 --> 00:45:39.760 But if we were to buy a company, it would really be either one of two things, like buying their revenue, like, hey, they already have$10 million of ARR, and that can accelerate us and we can cross-sell to their companies, like light source instances to grow that. 00:45:40.079 --> 00:45:46.480 Or maybe they have some really core kernel technology that we absolutely need, but that almost never happens. 00:45:46.719 --> 00:45:52.000 Or we really like the team and we can uh we can hire them, but that would be too expensive to buy a company for. 00:45:52.079 --> 00:45:56.480 So that's why aqua hires tend to be when companies are like really not doing well, right? 00:45:56.559 --> 00:46:06.480 So if you think about like another coupa acquisition, I won't say the name, but they made an acquisition recently where it was clear that the company was just, it was, it was just stagnant. 00:46:06.719 --> 00:46:09.519 And they probably bought it for pennies in the dollar. 00:46:09.599 --> 00:46:19.760 And they were able to say, hey, we made an acquisition, maybe we get some customers, and but more importantly, we get to acquire the team and like bring on like a few really great people onto our organization. 00:46:19.920 --> 00:46:23.280 And is it worth spending$10 million for that? 00:46:23.440 --> 00:46:25.039 Like probably, you know. 00:46:25.280 --> 00:46:29.840 I think especially now when like you can build software so much faster using AI. 00:46:30.000 --> 00:46:34.480 I think it's becoming less and less important to be able to acquire pre-written software. 00:46:34.639 --> 00:46:41.280 Like, I think even in 10 years, I think people will look at coding interfaces and be like, how did we ever write code by hand? 00:46:41.440 --> 00:46:45.360 Like right now, you write code, like a lot of people are writing code with English, you know? 00:46:45.599 --> 00:46:47.039 So thank you so much. 00:46:47.119 --> 00:46:52.400 I was kind of really interested to kind of get uh uh a look at it from the other side. 00:46:52.480 --> 00:46:56.960 I feel like so much how just we just report on, you know, Company X has bought company Y. 00:46:57.039 --> 00:47:02.800 And you know, it's nice to kind of look behind and you know, kind of see what drives people back to out searchers. 00:47:02.880 --> 00:47:14.960 I also didn't know and there was a case that it isn't just a uh a kind of plug and play kind of thing that you know you acquire one company and you know you lift up their resources, their programs, their tools, their solutions, and you know, you can't just embed it into yours. 00:47:15.039 --> 00:47:21.920 It's obviously a much more uh difficult process, and you know, it all depends on and on on how the solution was uh was kind of built and whatnot. 00:47:22.000 --> 00:47:23.440 So no, thank you so much, Spencer. 00:47:23.920 --> 00:47:31.119 So uh looking ahead now, obviously we're 12 days into 2026, you know, second full week back for a lot of people. 00:47:31.440 --> 00:47:36.719 You know, from your perspective, uh what do you think this year has in store for AI and procurement? 00:47:36.880 --> 00:47:44.639 Do you kind of see Agentic continuing to kind of its dominance of the space, you know, LLMs coming uh back to the forefront? 00:47:44.719 --> 00:47:48.639 You know, I'd I'd love I'd love to get kind of your thoughts on what was session at all. 00:47:49.039 --> 00:47:50.000 Yeah, it's a great question. 00:47:50.079 --> 00:47:55.920 I think that last year was the year of experimentation, and I think that this year is the year of implementation. 00:47:56.159 --> 00:48:03.760 So last year a lot of companies were doing experimentation,

trying to understand the landscape and trying to build uh build proofs of concept. 00:48:03.920 --> 00:48:12.880 And I think that this year will be a year where companies are actually starting to really incorporate AI and agentic systems into their long-term uh digital roadmap. 00:48:13.039 --> 00:48:19.679 I do think that there is a risk of, of course, of like people being disappointed at a certain moment, but I don't think that's gonna happen this year. 00:48:19.760 --> 00:48:22.559 I think that could happen in two to three years from now. 00:48:22.719 --> 00:48:26.320 Like anything with this much hype, it's just very hard to match it. 00:48:26.480 --> 00:48:31.440 And by the way, if everyone was light light source customers, I think they would not have that risk. 00:48:31.599 --> 00:48:40.400 It's that I know that there's, you know, 80% of the companies out there just are not are not gonna be able to deliver on the kinds of promises that they've made. 00:48:40.559 --> 00:48:52.159 And if that's the case, then I know it's gonna, I know the conversations in two years, people are gonna say, hey, Spencer, like this sounds really great, but we ran a pilot with XYZ, and you know, it didn't really work so well. 00:48:52.239 --> 00:48:55.119 So we're not, we don't really have the appetite to try something again. 00:48:55.280 --> 00:49:00.639 And then I'm gonna be in the position where I have to convince them, hey, look, like I know that you ran a pilot with those guys. 00:49:00.800 --> 00:49:03.760 Those guys are not, they're not technologists, right? 00:49:03.840 --> 00:49:05.199 They're not Silicon Valley based. 00:49:05.360 --> 00:49:09.440 I think the other thing you could do, honestly, just like look at the CTO of the company. 00:49:09.679 --> 00:49:19.039 If you look at the CTO of the company or you look at the engineering team on LinkedIn, I would stalk them, you know, not like not like in a creepy way, just like look through the LinkedIn of the company. 00:49:19.199 --> 00:49:39.119 If those, if the people in the team are all coming from like Waymo, Google, Facebook, Meta, uh, you know, uh, Airbnb, Microsoft, high-caliber organizations, then chances are that they are going to be able to deliver really impressive AI solutions because they're staying close to the latest technology and they're coming from places where that kind of stuff is developed. 00:49:39.280 --> 00:49:48.239 Like our CTO, Adon, for example, came from Google X and Google Research, which was the birthplace of the Transformer model, which is now the T in Chat GPT. 00:49:48.320 --> 00:49:52.400 It's like the basis of every LLM is this model called a Transformer. 00:49:52.559 --> 00:50:03.280 And so it's funny, like there are people who are like talking about transformers, and he'll sort of like he'll kind of cringe because it's like, you know, I was in the team that developed this, you know? 00:50:03.519 --> 00:50:16.559 And so there is just an element of like, you know, being ahead of the curve that I think involved is involved, you know, if you bring really, really high quality people to the table from these elite institutions and and have access to knowledge early, you know? 00:50:16.800 --> 00:50:21.840 So I think that actually is the third canary in the coal mine that you can look at is you know who's on the team. 00:50:22.239 --> 00:50:22.559 Perfect. 00:50:22.639 --> 00:50:22.880 Yeah. 00:50:22.960 --> 00:50:25.920 I think you know, my prediction for this year was kind of similar. 00:50:26.079 --> 00:50:37.360 I I, you know, I thought more companies will kind of take the plunge and and and move from pilot to you know uh an outdoor adoption program and actually moving forward to it. 00:50:37.440 --> 00:50:46.960 So fingers crossed, you know, I think we always kind of see at the end of, you know, I at the end of last year there was uh clarity report saying, you know, people are still stuck in stuck in those kind of pilots. 00:50:47.199 --> 00:50:52.239 So you know, how do you think in 2026, you know, people move away from from the pilot program? 00:50:52.719 --> 00:51:10.800 So I it feels like we've kind of said it all throughout this this last hour that we've you know that there seems to be big enterprises kind of in their ways, people are messing around with AI, but they can't get their pilot program off the ground, they can't move forward to it, they can't implement the software. 00:51:11.119 --> 00:51:21.119 Uh and you know, it's you you're worried it's gonna lead to those conversations this time next year where people say, you know, they you know there's a fear that people become disillusioned with it, you know. 00:51:21.360 --> 00:51:33.760 So how does 2026 become a year where procurement teams can, you know, take advantage of this day to day rather than it become this never-ending Yeah, it's a really good question. 00:51:34.000 --> 00:51:36.800 Now, I think it depends on who I'm speaking with. 00:51:36.960 --> 00:51:47.199 So if I'm talking to an individual buyer category manager, I would say make sure that you have access to LLM tools that you can use safely within your organization. 00:51:47.440 --> 00:51:51.119 At home, use Chat GPT, use Claude, whatever you want. 00:51:51.280 --> 00:51:53.280 At work, you have really sensitive data. 00:51:53.519 --> 00:52:02.719 Chances are your IT team, if they're doing their jobs right, will be giving you access to an LLM tool, whether it's co-pilot, chat GPT, or something else. 00:52:02.880 --> 00:52:09.440 I would say the number one thing you can do as an individual is become like just start using AI in your day-to-day work. 00:52:09.519 --> 00:52:19.840 Just start using Chat GPT, become addicted to ChatGPT, just use it for everything, and then you'll find that you can automate large amounts of your own work and just you can do the work of like 10 people. 00:52:20.000 --> 00:52:22.480 It's like having an army of AI interns. 00:52:22.639 --> 00:52:23.440 They're not

always right. 00:52:23.519 --> 00:52:27.599 You have to like sort of curtail them, but it's it's definitely an accelerant. 00:52:27.760 --> 00:52:33.440 If you're a CPO or a CEO, then you have to really work to partner with your IT team. 00:52:33.679 --> 00:52:43.920 I have found by and large that IT teams that have a dedicated procurement and supply chain member, I think serve those customers much, much better. 00:52:44.159 --> 00:52:48.480 Sometimes IT can be a bit on an island, like a silo away from the business. 00:52:48.639 --> 00:52:56.800 And then they will be in their own world creating their own AI digital roadmap strategies that may not be as connected with like the needs of the customer. 00:52:56.880 --> 00:52:59.920 And the customer here would be like procurement, like the CPO. 00:53:00.079 --> 00:53:18.000 And so what I would ask, if I were a CPO, if I were chief supply chain officer, I would want to have an IT resource that reports to me and dotted lines to the IT, the CIO, because I find that in those organizations, they tend to have much more agility to get tools that they want that actually end up getting used. 00:53:18.159 --> 00:53:20.639 I think there's a lot of risk alternatively of shopware. 00:53:20.960 --> 00:53:49.199 And then what I would do personally is I would run, I would really map out like what are my processes and what are the goals that I want to achieve, and then triangulate that against like first start with the nuts and bolts stuff, like the brick and mortar, like blocking and tackling, like what are our workflows, what do we need to do, and then figure out what can be automated, what should be automated, and then go and take a pretty broad perspective on the kinds of companies that I'm willing to chat with about it, you know, and then bring them into the fold. 00:53:49.280 --> 00:53:54.400 And then I think it's just like this like conversation, proof of concept, contract. 00:53:54.559 --> 00:53:57.679 Now, the thing is you have to be ready to actually contract. 00:53:57.760 --> 00:54:06.000 Like I think there's another thing, which is it's very easy to run pilots and proofs of concept and then put things in your wish list and not revisit them for six months. 00:54:06.239 --> 00:54:10.880 But then you will be delaying progress for the business by six months to a year uh by doing that. 00:54:11.039 --> 00:54:14.559 So really say, hey, we're gonna, we're gonna do, we're gonna try something. 00:54:14.719 --> 00:54:17.840 If we like it and it works, we're gonna buy it and we're gonna implement it. 00:54:17.920 --> 00:54:24.159 And then of course, like the AI landscape is changing in two, three years, you can always reassess, transition to something else. 00:54:24.320 --> 00:54:26.800 But I do see a lot of testing in the market right now. 00:54:26.960 --> 00:54:31.519 And then if I were a CIO, I think that they have a really hard job right now. 00:54:31.679 --> 00:54:32.480 That's my perception. 00:54:32.639 --> 00:54:40.159 I think it's really difficult to be an IT leader because everyone's looking at you and saying, we have these digital systems and they're not good. 00:54:40.320 --> 00:54:44.159 Like, why is ChatGPT so much better than the tools that were being given? 00:54:44.400 --> 00:54:45.679 You weren't born today, right? 00:54:45.760 --> 00:54:48.079 Like you're sort of an you're sort of, it's like the president. 00:54:48.159 --> 00:54:52.559 You're you're you uh you're running the economy that your predecessor designed, right? 00:54:52.800 --> 00:55:00.880 And so, you know, you have you have systems and data and you can't break, you don't want to break the ship while you're while you're while you're while you're captaining it. 00:55:01.039 --> 00:55:05.119 And so to some degree, you you know, you have to maintain the systems that already exist. 00:55:05.280 --> 00:55:11.920 There are decisions that were made before your time or before AI really existed that you know you're living with today. 00:55:12.159 --> 00:55:13.920 But I would be open to new solutions. 00:55:14.000 --> 00:55:31.840 And I think that a lot of times what it makes sense to do is some of the best companies that we've seen strategically what they do is they move away from storing data in siloed systems and they move towards storing data in a unified data architecture, like a snowflake type system. 00:55:32.079 --> 00:55:34.800 And there's many, many alternatives as well. 00:55:34.960 --> 00:55:44.639 Uh, but if you store all of your data in a data lake and or some kind of really strong data warehousing system, it doesn't really matter at some level which point solutions you're using. 00:55:44.800 --> 00:55:49.920 Those are just business logic and they're drawing from and repopulating like a unified data system. 00:55:50.079 --> 00:56:07.440 And I think that's much better, to be honest, and much more flexible than having all of your data, like having like 20, these kinds of this set of data in tables and SAP, this set of data is in tables and in Coupa, this set of data is in tables in Light Source, this set of data is in RPLM. 00:56:07.599 --> 00:56:11.360 If every system is just instructed, push it to a unified data layer. 00:56:11.519 --> 00:56:13.679 And by the way, this is not a procurement data layer. 00:56:13.840 --> 00:56:25.519 This is a data layer that exists across the full organization, including HR, including sales and service, et cetera, just different data tables, then I think that you'll have a much more unified system. 00:56:25.599 --> 00:56:29.599 And then your switching cost is also much lower as new utilities come up. 00:56:29.760 --> 00:56:31.440 I'm trying to think about a good example. 00:56:31.599 --> 00:56:38.559 It'd be like, you know, it's like, I don't know, like it's like replacing your computer, but then keeping your electrical system in your house the same. 00:56:38.639 --> 00:56:40.159 You know, you plug in a new computer. 00:56:40.320 -->

00:56:40.800 It's okay. 00:56:40.960 --> 00:56:44.480 You can upgrade your computer without changing the wiring in your house, you know? 00:56:44.639 --> 00:56:48.400 But really smart IT teams like think this way and they develop things this way. 00:56:48.480 --> 00:57:02.719 And and then some are are really behind the ball and you know, frankly, got swayed by sales folks at some point to do the wrong thing and get locked in with a vendor, and they're spending millions and millions of dollars on technology that's shelfware, and that happens all the time too. 00:57:02.880 --> 00:57:03.920 Businesses are imperfect. 00:57:04.159 --> 00:57:06.800 People great people also make bad decisions sometimes. 00:57:06.880 --> 00:57:07.440 It just happens. 00:57:07.519 --> 00:57:08.320 I do all the time. 00:57:08.480 --> 00:57:22.559 So yeah, no, you uh you absolutely spawn you know, you can kind of get stuck in those loops, and you kind of help we've all been kind of seduced by a salesman before believing the product we're about to buy is gonna change our lives, and unfortunately it doesn't. 00:57:22.800 --> 00:57:30.159 I'm gonna round things off now with kind of uh the last question, almost like looking for something controversial for you to finish on. 00:57:30.400 --> 00:57:35.199 You know, what do you think will be the most shocking development in AI for 2026? 00:57:35.519 --> 00:57:38.639 I I have I have one take that I think is gonna happen this year. 00:57:38.800 --> 00:57:43.280 And this and I don't know if maybe it'll be 13 months away, who knows? 00:57:43.360 --> 00:57:49.840 But here's my here's my prediction is that I think in the next year we're gonna see a one-person unicorn. 00:57:50.000 --> 00:57:57.119 And what I mean by that is a unicorn is a a unicorn in Silicon Valley is a company that is valued at a billion dollars. 00:57:57.440 --> 00:58:17.519 And I believe that within the next year, there's gonna be some kind of news story about a company that is valued at a billion dollars, where they do an exprose and they discover that it's actually just one person, one guy or one gal in their garage running with like 50 15 computers. 00:58:17.760 --> 00:58:20.719 Each one is just kind of doing a business function. 00:58:20.880 --> 00:58:36.480 They have a sales and SDR agent system, they have an account executive agent system, they have a payment processing agent, they have, you know, a software development team that's all being run through Claude Code with a separate agent that's like reviewing the code. 00:58:36.639 --> 00:58:44.800 And it's gonna be like one person just really good at using and orchestrating AI tools that builds like a billion dollar company. 00:58:44.960 --> 00:58:48.320 And then I think it's gonna be uh I think the floodgates will be open. 00:58:48.400 --> 00:58:51.440 I think it'll be a hard world for venture capital at that point too. 00:58:51.599 --> 00:58:53.760 But but that's uh but that's my prediction. 00:58:53.920 -->

00:58:56.719 If I had to just like choose one, one man unicorn. 00:58:57.119 --> 00:58:59.360 I mean, that would be absolutely incredible. 00:58:59.519 --> 00:59:03.280 Uh is this you kind of secretly revealing what you're doing in your off time? 00:59:04.880 --> 00:59:08.559 No, but I do I do uh use these tools. 00:59:08.639 --> 00:59:12.320 We're we're we're unfortunately way past the point of being a one-person ban. 00:59:12.480 --> 00:59:16.320 And we've been a two-person band, me and my co-founder since the very beginning. 00:59:16.480 --> 00:59:19.920 So that that has unfortunately passed that moment has passed for us. 00:59:20.159 --> 00:59:26.880 But but on my nights and weekends, I do build stuff, you know, and I even created like I created this site. 00:59:27.039 --> 00:59:30.960 I was sort of interested in seeing if I can actually use these tools, right? 00:59:31.039 --> 00:59:33.280 I want to make sure that I'm staying close to the technology. 00:59:33.440 --> 00:59:35.760 So I created this app called Job Source. 00:59:36.000 --> 00:59:45.440 It's basically an application that I built using AI, and the tool itself uses AI to help you find procurement jobs. 00:59:45.599 --> 00:59:48.159 And so it's procurement career intelligence. 00:59:48.320 --> 01:00:06.559 You click find a job and it goes out into the world and it pulls a bunch of stuff, uh, salary data, career portals, et cetera, and constructs for you an environment where it shows you all of the possible procurement jobs that are available on Earth. 01:00:06.719 --> 01:00:10.000 You can filter by locations and skills. 01:00:10.159 --> 01:00:16.880 And we have an AI matchmaker where you can put in your you your profile, LinkedIn profile, and it will match you with the best jobs. 01:00:17.039 --> 01:00:18.400 You don't have to search on your own. 01:00:18.719 --> 01:00:20.159 Like procurement. 01:00:20.480 --> 01:00:23.199 Yeah, like exactly, like Tinder for procurement. 01:00:23.360 --> 01:00:31.599 And then you can form things like wish lists and send these to yourself, and you can view, you know, view the jobs and see what they what they include. 01:00:31.679 --> 01:00:38.239 And I don't know if you have access to the information, but do you know if if anyone has is gotten a procurement job because of because of this tool? 01:00:38.559 --> 01:00:39.440 You know, it's interesting. 01:00:39.599 --> 01:00:44.320 I just launched this like a month ago, and this this is an example of a side project. 01:00:44.480 --> 01:00:51.119 Nobody really from Light Source was involved in it, other than the fact that there's a link to Light Source, a banner ad at the top. 01:00:51.280 --> 01:00:52.639 And I'm also interested in the stats. 01:00:52.800 --> 01:00:58.400 You know, how many jobs, what's the density of job locations that are being captured here? 01:00:58.639 --> 01:01:01.920 You know, what are the top hiring locations, the top hiring companies? 01:01:02.079 --> 01:01:06.880 ABB apparently has like a ton of procurement jobs open right now. 01:01:07.119 --> 01:01:10.800 The seniority levels, you know, job postings over time. 01:01:10.880 --> 01:01:14.719 You can see that weekends are a very unpopular time to post jobs. 01:01:15.360 --> 01:01:19.760 So, and then light source skill is required preferred for all of them. 01:01:19.840 --> 01:01:21.440 That's our, but that's just cheeky. 01:01:21.519 --> 01:01:22.400 That's my own thing. 01:01:22.639 --> 01:01:26.159 So, you know, and and and and the the point of this is not, I think there's two things. 01:01:26.239 --> 01:01:29.760 One, it's like, of course, if you're looking for a new job in procurement, where do you go? 01:01:29.920 --> 01:01:31.360 There's thousands of websites. 01:01:31.519 --> 01:01:34.400 It's all pulled together here for you using AI. 01:01:34.559 --> 01:01:40.320 But the second thing is that I myself as a CEO don't have time to develop other companies in my free time. 01:01:40.480 --> 01:01:47.039 In fact, you know, I'm 300,000 of my time is committed to Light Source, more than I could ever dedicate. 01:01:47.280 --> 01:01:50.960 But it's still important for me as a leader to stay really close to AI tools. 01:01:51.039 --> 01:01:53.679 And so I built this using code generation platforms. 01:01:53.760 --> 01:01:55.440 You know, I was really playing around with a bunch of them. 01:01:55.679 --> 01:02:04.639 I wanted to figure out can I do the work that would have otherwise taken some team of engineers a week to put together and do it in, can I do it over Thanksgiving in like a day off? 01:02:04.800 --> 01:02:06.320 And that's what uh that's what I did. 01:02:06.400 --> 01:02:09.679 And so I think it's really important for leaders to stay really close to this themselves. 01:02:09.840 --> 01:02:12.239 Like if you want to have it always comes from the top. 01:02:12.400 --> 01:02:17.920 You know, if you're experimenting and you're staying close to the technology, you can enforce that on your team and you can inspire them. 01:02:18.000 --> 01:02:21.440 And if you're not staying close to it, then it's very hard to know what's happening. 01:02:21.599 --> 01:02:24.800 The art of the possible is like will be out of your reach, you know. 01:02:25.440 --> 01:02:27.519 Well, I think that's the that's a great place to leave us, Ben. 01:02:27.679 --> 01:02:31.440 So uh honestly, I could chat with you all day, but sadly that's all we have time for today. 01:02:31.519 --> 01:02:33.440 Uh, thank you so much for joining me today. 01:02:33.599 --> 01:02:34.320 Thank you, Aaron. 01:02:34.480 --> 01:02:35.599 Really appreciate the time. 01:02:35.760 --> 01:02:39.280 And uh thank you all for for listening to the Procurement Magazine podcast. 01:02:39.360 --> 01:02:45.280 Uh, be sure to subscribe on whatever platform you get your podcast from so you can know the moment we have another episode for you. 01:02:45.440 --> 01:02:47.199 Until then, bye for now.