00:13:48 · Seema Amble, Praveer Melwani
[00:13:48] S1: You have, like, an investment committee that exists, like, interaction between all of them [00:13:52] S0: through the whole Exactly. So so that's exactly what they'll end up demoing. Or rather, they'll they'll sit there kinda going back and forth and saying, okay. This is what Praveer would say, and this is what Dylan would say, and this is what, like, you know, Sean would say, our CRO. And so you you kinda get a... You get you get, like, the the areas of the document that you either want to go and plus up, do more diligence on, what are common sets of feedback that I need to go and make sure I ask the questions over, and then it reduces the cycle time that we're all sitting together on. So, you know, there's a lot of these areas where, you know, we're both feeling like it's a tool that helps you augment, it's this tool that helps you go and replace workflow, and then it's this tool that helps you seek to understand and learn. And I'm constantly surprised at how quickly we've been able to riff on things, and for me it's actually put me in this really optimistic spot of saying like, I'm gonna... I I think we're gonna be able to attract and build the world's best finance organization because so much of what people want is the ability to learn and feel like they're being invested in and surrounded by people that wanna go and, you know, become the best at their craft. And as much as I can do to go and effect and enable that change, the better I think we're gonna be able to go at retaining and growing and grooming that talent over time too.
00:14:59 · Seema Amble, Praveer Melwani
[00:14:59] S1: And are all those tools... Did you build those right with the LLM directly, Or are you guys also looking at off the shelf AI tools that you can partner with? And would you rather build or buy? [00:15:13] S0: Yeah. The ones I was describing, you know, it's a mix. Like, some some of them folks were sitting in codex and, yeah, I think that's what the team is doing on the on the corporate development team. They're sitting in codex and trying to, you know, work their way through something. We've also seen folks, like, sitting within, you know, some of the workflow automation pieces and building skills around agency... ChatGPT, and so that's how some of the credit and collections pieces were were being built. And we've also started to see folks being able to embrace it and build it on our tooling as well. So we, know, we have a tool called Figma Make, which allows folks to be able to prompt through natural language and create, you know, an application, a website. You can then host it. You have, like, a, you know, back end behind it. And so, you know, some of those instances are starting to crop up too. So I'm actually excited about that because then we can then we can ideally flip into this world where we can give product feedback as well, which, you know, I I know everyone would love.
00:16:02 · Ivan Makarov
[00:16:02] S2: I think we're in the era right now in the period in the cycle where a lot of slop is being called out. There's been a lot of conversation too where we went from unlimited token to people just calling it slop in meetings, or I've heard one CFO talk about how using AI just led to unlimited amount of docs that are being created every day, and you just can't find anything, because everybody can write a doc now, or 10 docs in a few minutes, it would take you a week or two. Were there some of the areas where, and then naturally created also the conversation around hiring. Do you hire younger people that are more versed in these tools, or do you hire people more experienced, that may be both, or do you hire more senior people that know what the world class looks like? And that's what worries me a little bit, is that we will forget what the world class look like. Just like in the pre AI era, you know what a really good model looks like, like, especially with some experience and having created models. Where do you think this is going? Like, have you had to call slop on some
00:16:59 · Praveer Melwani
[00:16:59] S0: of the stuff that's being created right now with AI? Sometimes. Yeah. Yeah. I I think, like, you know, I think we... We've also done a good job internally about building policies around, like, branding when you've written something with AI. I think, like, you know, one of the things that's really important, I think I was sharing it earlier, like, you should not be outsourcing your thinking. You know, anything that you're putting out there that you're asking someone to consume, it's gonna take them time, effort, and energy. And so I think you wanna make sure that, you know, you put a similar bar to go and review, read, summarize, make it simple. You can start in a different place. I'm totally cool with that. But I do think it's on all of us to make sure that we're, you know, being really thoughtful about how we we use each other's time as well. So, yeah, I mean, I've been in instances where I, you know, things look a little bit half baked and you really poke around, you're like, do you do you really understand what's going on over here? But I've been... Also been in instances where I'm like, oh my gosh. Like, your speed of iteration here is, like, pretty damn incredible by virtue of how you set it up. Like, a couple weeks ago, was sitting in a... We're going through 27 planning. I was sitting in a T and E review. We're looking, you know, we're looking at our T and E portfolio and saying, okay, where where do we have the opportunity to go and make changes to policy? Or maybe where where are we not upholding policy where we should? And our corporate finance team, they did a really good job about stitching together information that lived in different data sources. We used Navan. We used SIP. We used RAMP. And there's a lot of fuzzy match that you have to go through in order to, like, really get that information working. Like, was someone actually traveling for for an off-site? What if they lived in San Francisco and were just visiting? Like, how do you actually wanna go and, like, tag and treat that data? Like, 95% of the challenge there is just, like, getting your data in the right spot. And these models are good at making at making these assumptions for you. And I I didn't need something a 100% perfect. I just wanted something that was directionally, you know, in the right...
00:18:43 · Praveer Melwani
You know, in the similar vein for where I'm going that gave me enough information to understand, you know, what are the areas I really need to crank the dial on. It was really interesting. They both, like, outputted a set of suggested changes that we needed we needed to... We could make. I gave... I remember giving feedback that morning. Thirty minutes later on, like, some of the construction of how I actually wanted things to look like, like, it was all visualized for me when we got in the meeting. Was like, holy shit. This is, a really fantastic use case here. And they did it in a way that, like, you know, I don't think they would have been able to do it by hand. So, you know, as as much of that slop, you know, I think can happen. I also think some magic can really get unlocked if you if you if you kind of, like, use the right tool for the right purpose and understand, you know, what is what is good enough and what needs to be, like, really, really picture perfect.
00:19:27 · Seema Amble
[00:19:27] S1: A couple of comments to that. I mean, a key unlock is the data. Right? Now with AI, you can consume all these sources of data and quickly do the fuzzy match, get a a directionally correct or precise, you know, reconciliation across these sources. Whereas before, right, you had to run the SQL query and had to... You had to wait and then and see it... And and then and then essentially, you had to be your own BI analyst on on top of it. The second piece is there's this muscle that's being developed around the automation. Right? And so even if there is slop... Like, for example, I was talking to one of our growth stage c... CFOs who was saying that in an... I think a month or two, have been, you know, 40 or 50 agents created by their team, which obviously are not all being used. But it was getting everyone in this motion of of, okay, if I don't... If I'm doing something manual or tedious, I don't have a great tool for it. Let me build. And so a lot of those, maybe 90%, I'm... I would guess, of those tools and agents are not being used anymore. But it got people to think, okay. Instead of doing this manually, let me think automation first around it, which I think is almost as important as getting the, like, the the agent built.