The Future of Finance is AI-Native: Jonathan Sanders, CEO of Light, on Organic Software, AI Agents, and Rebuilding Business Finance

0:00Hello and welcome to another episode of Riding Unicorns. Today we're incredibly excited to be joined by Jonathan Sanders. Jonathan's the CEO of Light, a company on a mission to redefine business software with an organic approach to AI. So Jonathan, you've had an incredible background with Bain, Plio, and then you were the founder of Juni as well. So we'd love to dive into some of the operational side of building companies, exploring tools and systems. AI obviously is going to be a topic of discussion. So we'd love to find out more about some of the workflows around that. But I'll start by understanding a little bit more about you and your plans.

0:38Were you always destined to be a founder? Was that always your goal when you started out? Thanks for having me, Hector and James. It's a pleasure to be on. I think my journey to become a founder started very, very early on when I was about 15, 16. and I had this urge to change the world. I sort of saw the world working in one way and I thought it's not a great way and I want to change it. And so I actually initially started in like politics and I was doing that for about two to three years until I realized that you can dedicate your entire life to this. And if the stars doesn't align, that one thing that you've been crusading your entire life might never happen.

1:20And so like just the odds of you being able to effectuate change in a democratic political environment for good and for worse is just it is really really hard it's really really hard and then i needed to figure out okay like what do i then do and and i i went to to london i'm originally from from denmark despite my surname i have british grandparents and so i went to london and back then like the financial center in london was like the hottest thing in europe and almost in the world like the london financial center was like the thing we also had new york and you had a little bit of singapore but but like london was pretty much up there right and so you could have like an outside impact on how the world works via that and i was fortunate enough to get a great job in a bank back then that had done really well coming out of financial crisis and then it handled that upswing really badly so i joined credit swiss and it was doing really well when when i joined and then later it had a lot of troubles and unfortunately went bankrupt.

2:23I left much, much, much, much, much before that because I basically found out that moving money from putt A to putt B is not how I wanted to spend the rest of my life. And I had worked in a small startup in the medical sector when I was a student and I thought I want to do that. So I went to business school, found a partner. We talked about like starting a company together. And since then we sort of started it and then some family stuff and migration log got in the way. But since then, for a little bit of a detour for a bit of time, I went into the startup sort of game. First with Playo, that was a lot of fun.

2:59Then with Junie, that was a lot of fun. And now lastly with Light. And all I can say is that if you have an undying will and desire to change the world, one way of doing it is to become a founder and do it. and if you're thinking it could be cool, do not do it. It is, in many ways, it's also a terrible job. You get all the stresses in the world on you. And to be honest, I wouldn't, I don't hope that any of my kids become founders because I know what it does to your mental wellbeing and the amount of stress you're on. But it is one of the truly magical ways that you can effect your change in the world.

3:34I had this idea, I wanted to do something very differently in how like finance works. And then you try to bring that idea to life. and now there are employees and there are people wearing our merch in the streets of San Francisco and London and Copenhagen. And in a way, you know, like that idea has now become somewhat reality. And I think that's the closest that I've ever experienced coming to magic. Jonathan, awesome to have you on the show. And speaking slightly of magic, you've talked about this sort of concept of organic software. And I think as a result of really AI, you know, there's a new era of software where rather than the user kind of having to adapt to the software, it's the other way around.

4:12So the software adapts perfectly into the user's workflows. But can you give us like some sort of really concrete example of what you mean by this and maybe apply to the work you're dealing with these finance workflows? Just to set the scene, I've been obsessed with AI since I started my career as a quant. So in the bank, we had like, we were using machine learning techniques. I could see how powerful those tools were in terms of like producing alpha. And then when you join the business side of things, when you join like companies and startups, you can see how very unsophisticated almost any business is when it comes to using stats.

4:50And so like, even like simple regressions is out of this world for most business purposes, right? So I always knew that there was a great application of that into the world of business to basically be a bit more sophisticated in how you look at data and how you automate your data. I was fortunate in that we did basically automated investment decisions, which is some of the most sensitive decisions that you can make, right? Like your investment decisions, either you directly make or lose money and that's on you and we automated that. And that is basically part of the current world is that you basically outsource the decision-making authority into a machine.

5:24So I came from a school of thought where no, no machines can do very sophisticated and highly leveraged and highly risky things for you. And they could basically manage their money. And that's about as much power as you can delegate to a machine as possible. But we did that before AI. And then when we started Light, we actually started to work together with OpenAI. This was back when they only had API models called GPT-2, or maybe there There was a three, I can't remember, but it was just, it was a couple of months before they started at GPT. And one of their ideas back then was as much smaller companies.

5:57So you were speaking to people actually building it. And one of the ideas from that is that basically the way that they thought about it is a student assistant that learns from you and sometimes make mistakes. And that is from call it GPT-2 to where we are today. But the core idea has been with us ever since. And so one of the ways that it does it, and it does that from the very beginning, is that you basically it learns over time like you get in like invoices on a pdf and they're messy and they need to be coded and once they've been coded the next time something similar comes from the same supplier it gets the same coding as as you did it before and now that is like a small automation where it's very easy for the user because they just correct the student once and then the student learns from next time and then it knows okay i should do this differently next time but that's not the bar anymore.

6:47The bar and the possibility now is truly magical. So we have an always-on agent that runs in the background. And what that does, it's called Astra. What it does, it looks at all of the data that goes through your system. And then it creates a summary of what it observes. And then it writes that into a table. And then once a day, it looks at the summaries of all of the runs. And then it creates insights from those information patterns. So you We basically go from data to information to insights on that classical sort of enrichment step. And through that system, it started to tell me things about my own company that I had no idea that I would know.

7:23So what it would tell me was that we had an offsite in Italy three weeks ago. And at the beginning of the offsite, it would say business travel to Italy has started because you could see people booking flights in multiple different jurisdictions and then flying to the same area and then spending money in the same airport and then moving to a different location and spending money there. And so it actually told me that, you know, the offsite is going strong during the offsite. It told me like, it's still ongoing. I can see all of the spend. I can see, you know, some of the receipts are missing because some of our sales guys went to the bar and got a little late, right?

7:55And so you don't always do that. And then I could tell it, you know, hey, you know, next time this happens, if they don't do it in the same day, just freeze their cards. And then it started to just send notifications to everybody and start freezing their cards so they couldn't use it anymore. And then at the last day, it told me, you know, the offsite return has started because I can see people buying like Ubers and Lyfts and like all kinds of stuff and like different airports around the world. And then the next day it told me that now it was ended and there's no more travel activity. When you say it told me, what is that?

8:23What is that output? Is that like an email or how do you get notified? Yeah, exactly. So at Lite, we try to be everywhere where our users are. So we have a mobile app, we have a web app, and then we have a Microsoft Teams and Slack integrations. I get a daily report of what is happening and I get like a notification if ever anything material happens between the daily reports. And then it just gives me a summary. It's like, you know, Italy offsite continues. And then it gives me some of the things that have happened. And then I can tell it, you know, hey, you know, if JJ has all of these missing receipts, maybe I need to make a rule about it.

8:56And so I could just ask it if any of them haven't uploaded receipts within 24 hours of the purchase, please notify them. And after 48 hours, just freeze their cards. And then it started to do that as part of his actions. And for me, that is very easy because the UX is just me having an always-on assistant that I can just tell things. It's like a chief of staff that works 24-7 that basically enforces things for you, but also updates you on things. And I think that is slightly magical, but the truly magical thing is that it understood that there was an offsite and it was going on and nobody had told it that.

9:28But it just deducted that from looking at all of the pieces. It's like, okay, all of these plane tickets going to Milan and all of the cabs and all of the drinks in that same cafe outside of Milan. And then I can see all of the travels back and all of the Ubers and Lyfts and whatever TFL and whatever people do to get back to where they're from. From that, I can summarize the dues that people are traveling back, which makes sense. It's just, I didn't think that it would get to that, but it did. And then I could ask it, how much did we spend on the offsite? And it could tell me without me needing to categorize transactions and putting them in an offsite bucket.

10:04It just knew from that insight, from that information that it gathered every day, that we had this event. And from that, it can start to tell a story, which is ultimately what we're about. We are storytellers. And one of the beautiful things about AI is that we can actually enable it to tell your story using your numbers. So you can understand, oh, okay, travel was 80 % higher last month due to an offsite that accounted for 99 % of that. And if you get that out of the box, instead of needing to traverse through every single transactions, make sure it's categorized in a certain dimension so you can produce a report and you can look at those dimensions, right?

10:41Which is the classical way of doing it. But you can basically have the AI help tell your story so the CEO and your board can understand it. That is truly magical. And that is where we're going to. You asked like applicable areas, right? One of our clients, they were like, it would be like super cool if you guys could like validate VATs of all of our customers. You have a customer and you got to figure out if they got the right VAT code, if it's valid. You know, they might have a VAT code, but it might be deactivated, in which case you should charge them VAT, et cetera. You should only charge them zero rate on cross border if they have valid like domestic VAT, blah, blah, blah, blah.

11:16blah. And so we actually got the same agent to basically look through all of your draft invoices, look at the customer, get the VAT number, and look up the online VAT database and check that it's valid. And then it produces as part of the daily report. Me today, I also get like, oh, okay, we had these two customers in light that had missing VAT numbers. I believe it's this and this, and then finance just updated it. It's incredibly cool, Jonathan. And I mean, it is literally like magic compared to where we were a couple of years ago and how do you do all of this is it a combination of llms imitation learning how do people set it all up what's the integration process of connecting everything so that the data is all flowing through the system yeah so fortunately out of the box you get all of the data sort of flowing through light because you have all the sort of the use areas in light you have bill payments contracts you have cards in the core platform.

12:13And so the AI can just leverage, basically, you can think of it as having access to a wider selection of tools, and it can do things like freeze cards, and it can do things like publish a contract, or it can do things like Google some sort of thing, and then use that to understand how to categorize certain expenses, or notify you, or figure out that, okay, we have an intercompany allocation we need to do at the end of the month, and this is my allocation key, and now I'm automatically creating journals for these intercompany allocations. So what we've basically done is we gave it different tool ability, and then it comes seeded with some memory.

12:48But the beautiful part is that each company will have different things that they want to chat with the assistant about, what they want the 24-7 CFO to do. And so one of our clients that I was working with yesterday, they are paying some sort of regulatory bodies on a monthly and quarterly cadence. And as part of that, one of the most urgent thing is that to make sure that the agent notifies them if they have paid. And more importantly, escalates if they don't pay on time, because if you don't pay on time, you get fined and they just automatically fine you for any missing payment. And so one of the things we did with them was basically set up the agents with a schedule of all the vendors that are regulatory bodies in the platform.

13:29And then from there, one, it can keep itself up to date on any changes in the payment windows by just doing a monthly search for that regulatory body and figuring out, okay, has there been any changes? Like usually it's on the 20th, but isn't on the 15th. Keep that up to date, but also just inform them of those payments. And so the beautiful thing about AI, and I think, you know, AI and LLMs as a fundamental engine, we are only using like 10 to 20 % of the capabilities writ large in Scala. But it is really like figuring out how to deploy that. How do you design a system that learns, that is adaptive, that is self-healing?

14:05So you don't need to go and do a lot of configs all the time to make it do what you want to do. How can I have the conversation be the core config between you and the software? So the software molds itself to your needs instead of you needing to deploy consultants or spend a lot of time becoming an expert in the system to make it do what you want it to do. And that's absolutely where the users get wow moments. And we talked about magic. That's where it feels like magic. And that's how you keep customers happy, right? But I wonder, Jonathan, just changing tack a little bit, tell us about some of the misconceptions around building a business.

14:41I mean, you're now an experienced founder. You've also, you're experienced to other things as well as founding. But tell us about some of the misconceptions around sort of building a company. I think overall, it's a good thing that the idea and the concept of starting a business is maybe today is cooler than it was 10, 20 years ago. And I think it's a good thing overall that there are more entrepreneurs and there are ways of doing it. But I think one of the misconceptions is that it is maybe a little bit over glorified, if you will. And so I think people might have, and I see this in some of the people starting businesses that I try to help, might have a slightly misconstrued idea that starting a business, becoming a founder is a viable, similar alternative to joining Goldman or joining McKinsey.

15:30And it is just not. Is there like a trade-off between purpose and pleasure or something where you're going to get a huge sense of purpose from building a company, but it's going to be painful? It is. Is the purpose in the pain, I suppose. That's what I'm asking. I think there was this question to Jensen Huang from NVIDIA. And what he said is like, I just hope suffering happens to you. Because part of his building that company has just been a lot of suffering. And there is definitely suffering involved in starting a company. But you have your ups and downs. And you have your ups and downs sometimes within 24 hours.

16:08And sometimes within 60 minutes. Swing from one extreme to the other. You get a lot of purpose. So you do get that. And I think you need to have a mentality of my life is more fulfilling if I have purpose in it than if I have pleasure in it. And if you want a life of pleasure, and that is perfectly fine. But then I think, you know, starting a company is probably not the greatest path for you. You should probably just join it after it raised the series B or C. But as a founder, the ups and downs, the stress that you're under, it can be substantial. Can you share one of those moments for you, Jonathan?

16:44Obviously, you've had a couple of goes at this and from the outside, light looks like it's just been on a one way ticket to the moon. You've got an amazing product, great branding, solving the problem that's critical to every business. So what have the hard times been? I think if you speak to any company, any founder, James, like they would tell you the same thing, that there's definitely hard things involved. And I think one of the hardest thing that you discover when you start a company is like how incredibly emotional the journey can be. It's way more high stakes and way more emotional than investments is being a consultant or being in a professional sort of environment.

17:28because you have so much purpose and you're so close to the mission, which is we're building this thing to help this set of people do this, right? And so everybody is super passionate about that. And so that makes it also really hard, you know, if things with certain employees or partners don't quite work out. And that is probably the one thing that, at least for myself, that's one of the sort of hardest thing about the journey is because you got to assemble like an A plus squad. And sometimes you make mistakes, you hire the wrong people or you make the wrong partner choice. Going through that, that can be tough.

18:04And then, you know, you always have this at the back of your mind, even though like the company is, it looks outside and from outsiders, maybe they will say like, wow, I would love to be in that position. But you always have this sort of like fear that, okay, like any moment, you know, something catastrophic, some undefined catastrophe can happen and it will all go badly, right? But I think, again, that is the fear that keeps you on your toes. And that is the sort of the pressure that drives you to like make the hard decisions and ultimately do what is in the best interest of your customers. Because if you take care of your customers and you do good by them, then they will take care of you.

18:44But it starts with your team, your team members, and getting them rallied around the mission. And then number two from there, make sure that that team is now producing excellent outcome for your customers. Yeah, and obviously you mentioned your product gives you a sort of daily report. But what are some of the other things that you personally do on a day-to-day basis that you think help the company move forwards and make progress every day? I would say I started my sort of startup journey probably mid through the whole cloud transitioning that was happening. I think it started maybe five to 10 years before I started.

19:23And then I would say Light is the first company that I'm at where we're at the very beginning of this AI revolution, right? Like this is a complete new wave. And it means you need to redefine how you build things. How do you manage people? Which type of talent do you need? And all of that stuff is changing on an almost six to 12 months basis, which puts a lot of pressure on the organization. And one of the things that I've done in that is I have gone quite deep on using AI tooling to really understand how does it work and how can I use that to be more productive and amplify myself. Jonathan, once you find something that you think is valuable, what's your process for kind of getting the trickle down effect of that through your organization?

20:09I do a lot of showing. So I basically show the rest of the team, okay, this is how I am using it. One, you know, come with ideas into the product. So before I would come with ideas to designers and then we'd do something together. And now I can basically create prototypes and hand it over to engineers and they can actually click through and I can show it to customers and I can show it to designers. Like, okay, these are some of the ideas that I'm having and this is how I'm thinking about it. and wouldn't this be interesting? I think this could, in a very nice way, solve a lot of things in this area.

20:44So that's one thing, but it's also just like presentations. I haven't done PowerPoint slides or Google Sheets slides in half a year and all my presentation are HTML files that I built together with Claude or Gemini. And that just makes the storytelling much richer. You can do that in a different way now. You're not bound to the same objects anymore. And it's even the same with like one-on-ones, with how you manage, your span of control. I think should also change because you can actually do more with less than. And I think internally, I'm trying to like beat the message. And I think that is probably one of the hardest things is to really embrace this new technology.

21:20You got to change the ways of working. And it is especially true for engineering. So we keep having internal conversations and ideas and ways of working, which are very, what you would call AI pilled in terms of automated bug fixing, how to do delivery completely different design engineering and product is all one group at light at this point it's organizationally we're also aligning it towards that because they're all working directly in github and code changes how do you talk about that and how do you get that group to really embrace the new technology and the new way of working and i think we've done a pretty good job of that but i think maybe another company organizations and not just at light you know two three months ago there was a little bit of anxiety going around in the engineering culture at large about like what does this actually mean not maybe as much for my job but for my identity and so we spend a lot of time speaking about like identities change over time because you're now solving a different problem in a different way i mean you're still solving problems but the way you're doing it and the method of you getting there is fundamentally different now since december than it is any time before yeah and it's interesting because it's suited to a slightly different person not not all engineers love the job of prompting yes and and some have found themselves to be total you know experts yeah exactly that's the hard part right that's you know then you got to say okay this is the journey that we're on though you're great in the old world like this is the new world and you got to be on with us for this journey can i just reflect on something else there jonathan i think you said you said like you're actually presenting product features and ideas but with sort of mini prototypes that you've built and i think there's an old concept in venture which is like show me don't tell me so the founder presenting a demo is better than a deck or a document and i think we're now in a phase where like the documentation side of things is so easy because you can anyone can say hey i've got this idea this proposal and write something in five minutes with pulled or chat GPT and create something that goes here's an idea yeah but actually to go just go and build it documentation is just like layers of abstractions further away from like the product that you want to see in the world so if you can just directly build your product and say like okay this is actually how I want it like this is how the data works this is how manipulations to the data this is how the AI part works here this is how the UI is kind of like direction it's going the look, then for everybody, it is much easier to understand exactly what you mean instead of writing it and then trying to be very specific in your words of what you mean and what you don't mean.

23:59Just translate that to code, which is cheap now, and then you can show exactly what you mean. And you can give something that people can click around and give you feedback, or you can give it to your team and they have a very well-specified product directive. It's much more helpful for all involved. It's much more exciting to do. I feel like where people who use AI will find their jobs are much more interesting if you can get over the sort of learning, get up the learning curve. But we need to move on to the other questions. But I do just want to hear your kind of one minute view on what the future finance team of a scale up is going to look like in a post light world.

24:38One thing that is true is that you're going to go from being an operator to more of an orchestrator. but also the demand from the rest of the business are going to like have an outsized increase to follow up with that. So it's not like you can just, you know, deploy light and retire because now if the product helps you do storytelling and then you bring that story back to the business, you can basically empower the business much more. Like finance as its core is a control and support function. And it lives to make sure that, you know, money comes in and there's a control of how much goes out, et cetera.

25:12And then there's reporting. But the core purpose of basically finance is to record all of the economic activity of an organization and tell the story of an organization using numbers. And that is fundamentally what we're trying to do so that everybody in the business understands what's happening in the business. If the things that you're doing has no economic activity associated with it, it is probably of little relevance to the company. There might be a few exceptions here, but overall, I think this is true. What we're actually doing with finance is we're trying to describe our business in numbers so that we all have a joint view of it.

25:48And what we're now doing is basically going from those numbers back to storytelling, starting to like, you know, why are we very optimistic about Q3? You know, well, this is happening on the sales side and this is happening on the expense side. And these are some of the new hires we're making. And this is some of the talent we're brought on. And you can use that to drive so much more decision-making, whether that is in marketing and sales and product development, in manufacturing, like wherever it is in the business, if they have 24-hour access to that same page, you as a CEO and you as a board will now see a company that's much more unified around the same story because you have agents help tell and deliver that story.

26:27So I think that part of it, that need for strategic insight, that need for answers from the rest of the business towards finance is going to increase a lot because that's what's getting unlocked now. And Jonathan, look, we're going to move on to our final two questions. But it's just been a fascinating conversation about where the future of the finance function is going. So our next question is our future unicorn prediction. So if there was another company that got something really interesting about it that you thought could make it all the way to being a unicorn, who would that be? You told me to pick an early stage company, James.

27:03The earlier the better, but you can go for who I'm going to use. I'm not going to pick one that is just about to make or announce unicorn. But I would say one that is very impressive to me, we share an office together so I can see them up close how they're doing, is a company called Wawa Fertility that does clinic management for fertility clinics. And they combine basically CRM and patients and embryos and everything into one, consolidated AI-powered suite. Awesome. Yeah, that's really, really interesting. and not come across them. So it's always great to hear new, exciting companies on the way up.