Name: SupERPower Hour Session 4 — Building with Claude Code + MCP
Uploaded: 2026-04-17T18:31:03.458Z
Duration: 1 h 16 s
Description: SupERPower Hour Session 4 — Building with Claude Code + MCP

Transcript for "SupERPower Hour Session 4 — Building with Claude Code + MCP": Hello. Hello. Happy Friday. I'm John Glasgow, founder and CEO of Campfire. We are the AI native ERP for modern teams, and this is session four of our superpower hour. In this one, I've got an incredible guest. We're gonna be building with Cloud Code and using the MCP. And as promised for session four here, this is gonna be more of an advanced session, and so we're gonna see a bit of, building on the first three sessions, what it looks like with a CFO and how they're actually using all of this today in SEAT. One quick two two quick slides for you all. If you missed some of the earlier sessions, they're all recorded. Highly, highly recommend you to check them out if you register, for today. You should have received them all in your email. Just go ahead and, look up Superpower Hour Series. They're on our website. We started with prompting and vibe coding, more of an introductory perspective. Then we worked with one of our customers, Replit, and we built an app together in Replit. And then, last week, we did a more advanced task. We built skills together. We were working on our desktop. And now I'm honored to welcome up, Nigel Glende, the CFO and COO at Masterworks. And, we're as noted, we're gonna be building with Cloud Code and MCP, and there are some more sessions on the way. You will be receiving, an email, on some more sessions. But in the chat, would love to while we're kinda getting started here, would love to understand where everyone's joining us from. Put your location. And if there's anything you absolutely wanna see today, go ahead and share that as well. We'll be sure to really make this an interactive session and make sure we're covering things that are important to the audience. Nigel, thanks for hopping up, and thanks for joining today. Awesome. Glad to be here, John. Excited to do this. What what prompted me to bring Nigel up, I was in their office in in New York City, and I walked over to the accounting department. This was maybe nine months ago, and there was no software up. It was literally all terminals. And this is before Vibe coding took off. Everyone was writing Python and and and SQL and working in databases. There's no front ends on a lot of the the apps that they've built. They're purely working, from the command line to to run their software. And I I felt like when I go over to our engineering team and I go over to Nigel's accounting team, literally, everybody is was writing Python. It felt like the future of accounting. And now that bio coding's taken off, I think we're kinda seeing the world. Nigel and the rest of us are starting to converge a little bit. And so I think, like, bringing Nigel up, we're just gonna do a live kinda session of watching him work, seeing how he thinks, how they really build. You know, we've all seen the thirty minute vibe coded cash flow forecast at this point. Nigel's gonna, you know, take us through, like, how kinda real scalable finance workloads, are operating, for our for our team, at scale. And, the one thing I'm gonna give a shout out, masterworks dot com, and it Nigel was gonna have to share his email address. But, Nigel, do you wanna give a quick shout out? What what is Masterworks? And, quickly on your background before we hop in. Oh, yeah. Sure. Am I coming through okay? Yes. Greg. So, hey, everyone. Nigel Belende. I'm the, CFO and COO at Masterworks. I, I was an investment banker by background, for a long time working across really niche, specialty finance businesses. And then my my career intersected with the art market, oh, boy, about ten years ago when I helped build, a business, around art lending. Think about it as like mortgage mortgages for really expensive paintings. And, that led me to start working with our c, CEO at Masterworks, helping make art investable. So this is we've been on this journey for about, eight years now. We have $1,200,000,000 in total assets under management. You think about Banksy, Basquiat, all these all these great paintings you see whether in museums or Christie's and Sotheby's, we turn them into single stocks. And we've done that for a portfolio, an art collection that's now, 500, pieces, and growing. We're doing way more stuff around events and museum loans around the collection, which is super exciting. And what you couldn't might not be able to tell outside looking in is we've got, like, all kinds of really uniquely shaped, gnarly back office problems to actually make this make this all work. So it's why the the finance team has turned into a bit of a internal, skunk works. And AI and Campfire have just been this huge accelerance to all of the learning we're we're doing on the team. And, it's I think it's important context because the complexity of the operations of all of the art management, each one has its own SPV or call it its own kinda entity. There's a lot of cash reconciliations. There's a lot of, you know, activity that had to be done at scale because it's just too much volume to be done manually. And so I think that's. why Nigel's team was essentially forced in the early days. I you know, I think, five five plus years ago, you guys were really on this journey, to automate and find ways to to handle the workloads at such scale. And before we get in, you know, I know you recently chatted with with Morgan Stanley, and I think there was some interesting insights here that you had shared, a couple of your thoughts on where the world's going. I think as someone that's, you know, caught living in 2028, I would say, like, where do you see the finance world going? Where do you see AI workloads going? I know it's changing daily. Opus four seven, we're all still digesting every new model. But, do you wanna just share a couple of the couple of the points on how you think about it? And then we'll also. Yeah. No. I think journey is the right, it's the right way to think about it. And when it come from an operating seat, I think the first thing I think about is is buy versus build around around all of this and what makes sense. I mean, not the core DNA of Masterworks is, you know, we do a lot of software building on the front end for Masterworks, but it's certainly not building financial software. It's not building financial systems. We've been sort of forced into that given, you know, the unique characteristics of our business. Meaning, hey. Like, we're business that's trying to do 500 different mini IPOs that are $2,000,000 each. You know? There are sure. There are lots of vendors out there that that do that, but none of that scale. So, so we've had to really adopt these tools in a big way. In terms of AI overall, my hot takes, we're not gonna buy a system of record. We're not gonna try and recreate our ERP or CRM. Right? I mean, I think you need you need, great, you need a great system that sort of scaffolds that out, that has a lot of the core business logic that's gotta be secure, that's gotta be multi user. Sure, it's fun to build little prototypes of that, but that's something we're we're gonna continue to work closely with with Campfire and and other partners. Where where we see a lot of a lot of experimentation now and hence why you see, like, the the infamous cash flow dashboard is it's just that. It's like dashboards, connectors. People have, like, uniquely shaped problems of their business, and they wanted to look at it in a very specific way. And all these tools are amazing at customizing specifically the type of information you need. Now the challenge, I think, is anyone's had with any of these tools is how do you deliver basically context in the background information on your business in in the right way to these tools? Like, that's that's still a gap, and there are lots of different ways to, to to get at that. But I would say dashboards and and business intelligence and, like, surfacing, you know, what the state of the business is, in more real time ways, that's that's where we're spending a lot of time. And then, secondarily, like, orchestration. How do these things all work together? And that's, like, that's the real, like, messy part that we're in right now. I mean, what what I'll show you is, you know, a lot of master is, like, building these, like, Python scripts, which do very, like, specific things, and it's usually trying to pull data out of one system and push it to another. Oftentimes, there's a human that's orchestrating that in the middle. Like, I wanna be in a position where well, it's not gonna be, like, fully autonomous, but the orchestration can be much more AI assisted where people on the masterworks team are then dealing with, okay, validation. Is stuff right? Are they prioritization, dealing with edge cases, dealing with, factors that are, like, the very edge of the business versus they're more, like, rule based things where we can either create very deterministic logic or we can very narrowly scope what we want, like, a language model to do for us, to get some of those, like, those edge cases out. Amazing. But without further ado, I'm gonna stop sharing, and I'll hand it over to you. Yeah. Well, hey, guys. Anyone that's, worked with me and maybe maybe some of the former master workers on the phone kinda know about know about the fire hose. Am I am I screen's coming through alright? Yes. We can see it. Great. Entire screen share. So it's always kind of, like, it's dangerous when you do this. Like, maybe my, like, wife or something, like, text me in the middle of it. So, anyway, this is so we're doing it live. So people that have worked with me know about, like, the fire hose edition. Like, we just I I like to just throw a lot of different things at people, and, we kinda we kinda see what see what sticks here. These these Ghibli things are I guess these are like this is like old news, these things. But, CFO, recovering banker, and I'm a real, tool junkie when it when it comes to these tools. And and and as much as I love driving outcomes, for the master's business and others, there's, there's just a real satisfying, element to just getting something to work with these things. So people that have worked with, they know about the infamous Excel 2,003 formulas book I have. I think when I learned about Accelerace is when, you know, array formulas is what what really kinda changed changed my life. Okay. I wanna I'm gonna take, like, three minutes and do really quickly just, like, how I think about using these tools, overall. Okay? The biggest thing, that is I I think what's most important for these AI tools to be useful or not useful is how you deliver context to them. So a lot of talk about context engineering. And anyone who's worked with these tools knows that, like, the the more robust context you can provide, the the better outcomes you will get. Now oftentimes, I just call that brain dump prompting. Right? You can narrow it. Don't tell everyone's, like, had the the experience now of, like, voice typing, or talking to their computer. I'm a big fan of that. You don't have to correct anything you're doing. AI is very, very fault tolerant. But something I talk a lot about the team, and and it's a little bit of a mindset shift, is you remember something called the x y problem. The x y problem is like a known thing in engineering where someone is asking a question about their attempted solution to something rather than just asking what it is they're actually trying to solve. And that's, like, what's what's it's it's a helpful framework thinking about AI, and sometimes when you think about how you're trying to prompt it, sometimes take a big step back, right, and actually explain what it is you're trying to solve and is the objective and then work with it to iterate what those, what the kind of different paths to the solution are. Because these models may come up with different ways to get at the solution than what your, purported, what you think is the solution to. And, boy, when you're stuck with AI, you think something makes doesn't make sense, use these things as teaching tools. Just throw more AI at it. Have it have it explain something to you. John said this. I think there's a big convergence right now between, like, tools for finance people and tools for engineers. Right? I think at heart, everyone on the phone here like, everyone on the line here, like, we are systems builders, and we've been building systems with, like, a relatively limited toolset. Right? We're working with Excel. We're working with our ERPs. All great. Right? You're gonna find with these tools whether it's sort of a Python, it's called code, it's SQL terminal. Right? And not that I expect everyone to be working with those, like, primitive primitive layers, but a lot of the, a lot of things have been built in this engineering world, I think map very, very nicely for finance people. And, like, it's a fancy way of saying, like, things are a real pain in the neck for finance people because usually there's some, like, tool deficiency because we have to do something Excel. Like, that workflow challenge was already solved for engineers. Like, engineers, could they build stuff by nature or, like, already solved it for themselves? One one quick question, Nigel. It seemed like Excel used to be how all of us vibe coded in finance. Like, that was our version back in the day with the with the formula book there. Is your team using Excel at all? I think some of us, it's like, what's the future of Excel? Obviously, maybe you have investor reporting where you got to deliver a model for, like, a a fundraise or something. But is there you just showed where, like, it used to be Excel, now it's like Python. Is there any use case that you see Excel on your team today? I mean, yes. Yeah. I mean, listen. I know fintech has been, like, ringing the bell on the the or the death knell of of Excel for a long time. I mean, Excel was the original killer app that made the PC thing in, like, the early eighties. It's an incredible piece of software, and and I don't see that going away. I think let me just give you my example of my workflow. What I'm doing a lot more of now is and I'll I'll show you here in a bit, like, prototyping things out in in languages or or or in, frameworks that are very easy for these language models to manipulate, which means, like, code. Right? So if you just think of like, I can embed all the logic of some financial model in a spreadsheet. Right? That's like all a spreadsheet is. Right? It has a bunch of financial logic and then it, like, it has this, like, nice visual, like, design affordance for it. I can I can translate all of that same code all that same logic into code? And you see with, like, Python's, like, pretty pretty readable. I mean, that's the whole point of Python is, like, it's it's meant to be, like, kinda human human readable. And, and then you're and then Excel becomes more of, like, a sharing tool. Right? It's like I wanna like, I print something to a PDF, now I print something to Excel. And these models, like, I I used to think they were kinda crummy with with, with spreadsheets and crummy with with, PowerPoint. Like, that's not really the case anymore. Like, I think these things are pretty good, and the gap between it being, like, okay and, like, really good is usually, like, how you prompt it and create those skills. But, like, the the raw ability to manipulate is definitely is definitely there. So I don't know. It's it's it's it's definitely part of the toolkit now. That's good to still. Yeah. I mean, we're we're seeing a material drop almost. almost no Excel use spreadsheet use internally. I mean, let's let's dive into what we what we build here. You know, we got a, fee share automation. Like, how did you build it? Talk us through, and then, you know, at some point, let's just let's just get building. Yeah. Yeah. Let's let's do it. I mean, first things first, with these tools, and I'm not gonna do a lot in the in in getting, in getting up to, and and kinda getting you up to speed in Claude. It's like, use the use the terminal. Like, it's not as scary as you think. You wanna be, you wanna be in the terminal. Pull it up. One of the great things about doing the terminal using Cloud Code is you can spin up as many separate sessions as you want. So you see what I got here? I've got, like, all of these different sessions of Cloud going, and they all have, like, a little bit of a a different task. So rather than working on, the UI, we were kinda doing things more, sequentially. You wanna be doing things, kind of more, definitely, definitely in, in parallel. So let's let's take a look at what we have here. So I've got the, the Campfire MCP set up. And I've actually done two things. I have the Campfire, MCP set up. And then I also have I actually just gave Cloud Code the Campfire API. I was like, hey. Just let let's look at the API. Let's give his let's give Claude as much, like, capability to, to manipulate Campfire, as possible. So, like, I wouldn't recommend everyone do this at home, like, with their own books, and, like, this is obviously, you know, just a test. So let's, let's, take a look at what's available in Campfire, in sorry. In the demo in the demo account. Okay. So we're gonna have it, and, you're gonna excuse all my typos here because, again, Anne is very fault tolerant and knows what I'm talking about, as we as we go. So running running apparel. So it's gonna get, so it's calling Campfire here. We're gonna start to get, or it's gonna start to get rolling. While that while that's cooking, let me give you a sense of what else what else I'm using. So I set up a a GitHub repo. Okay. This might seem sort of scary. What is GitHub? What is repo? So I see, like, Google Drive. Right? It's like Google Drive. It's like where a lot of your, like, work can live. And what I'm saying, like, what what Masterworks is doing a lot now is a lot of the finance workflows are, like, living in one, like, GitHub repo. And you can just think about that is, like, it's a Google Drive, that can be collaborative where multiple users can basically, like, push and pull. It's designed for code and, like, mostly because we're using clogged code here. But it can be do it can be done for, like, any type of, any type of, file type. So while we're pulling that up here so here's where we go. Okay. So it's found our entities. It's found our departments. It's found our customers. And so this business, I started looking at it. This is a demo account here. So, this business is, looks like a SaaS business. Let's pretend I'm coming in here and I'm like, I see a phone, I like just got hired and I'm just trying to figure out like what on earth is going on with this business so far. Alright. So this looks like a kind of a SaaS business. I gave it kind of a fun name. I called it the goat and grass business. I called it like a like a landscaping like a goat munching landscaping, subscription business. But basically, it's basically, it's kind of a SaaS business. So so let's take a look at what we've got. So it's found our contracts. It's found our budgets. No. It got none loaded. It's done the p and l. It got our balance sheet, AR, APH. So right away, we can tell, like, the MCP is, like, really good at just, like, pulling out all of the data. So it's it's a lot of, like, what you can do in Ember. And Ember is, like, is is super useful for that. Here, we have a little more flexibility because, of course, we're gonna be able to connect it with with kinda different, different tools here. So why don't we just have this let's take a look at the balance sheet by month over the past year. And. So we're gonna have that kinda kinda look. this is fascinating, Nigel. I don't use the the command line much. I think for the group, like, when do you go to APIs and when do you go to MCP? Like, obviously, MCP is the new the the hot new one that most of us are building on. I haven't spent a lot of time building on APIs since you're more advanced. I think what would be the use case? I I here here's the reality is that they are they're they are one in the same thing. Right? The API is just it's just a, something that all these applications have on the back end that, that other software applications can talk to. So all the MCP is is MCP is like a little wrapper around the API that just tells the model what it does. That that's all. So so the it's it so you can kind of use them interchangeably. And, also, like, the MCP is just like a it's like an easy on ramp for the models to understand, like, what APIs are kind of, available. Does that make sense? But, ultimately, you're hitting you're hitting the back of Campfire the same way. Okay. Right? And I don't know if you know, but, like, from an accuracy perspective, is there a difference? Like, is it more deterministic on an API or is it kind of the you're you're you're essentially indifferent other than maybe token consumption? No. You I I think you're you're indifferent. I think what you need to be a little careful sometimes is, like, the MCP, it's obviously, like, it it it it all goes back to context. Right? You've got a the models have to, like, know kinda what they're looking at, and the MCP usually has, like, oh, here's this here's this API endpoint. Oh, and here's some, like, information you need to know about it and how it how it works. Now what's cool about Claude is, like, you could I read it. The docs for KFR are great. Like, just give Cloud Code the whole docs. It'd be like, hey. Like, here are all the docs. It's got this, like you even got a a great, like, LLM TXT file, like, you know, it just explains what all the endpoints are. So I I'm just sort of teaching Cloud Code, like, how to interact interact with these tools. And, obviously, I'm in a demo account. There's, like, a lot of affordances. Right? I mean, you wouldn't want just, like, Claude to, like, just go hog wild, but this is a little bit more of a showcase of, like, what what is, what's possible. Does that make sense? That makes a ton of sense. Yeah. I'll I'll I'll ask a question after this next one. here. Just. one sec. couple things. So alright. So we pulled up our balance sheet, and what's great about this is, like I don't know. Of course, everyone looking at this like, hey, Nigel. You're, like, living in 1980, like, looking all this stuff on on a terminal. But, like, the terminals are really good at just, like, printing numerical information, like, right off the bat. So if you're in, like, a quick exploration phase, like, I need to understand something very specifically about the business, You know? There's something that screwed up. This doesn't look right. This doesn't look right. Claude is gonna be very good at being able, at being able to, to do that and and exploration. So it's giving me a couple so it's even giving me a couple, a couple takeaways. Right? So this business got up and running. Let's see. Got up doorman until August 2025. Is that right? Yeah. Well okay. I see. We got our service revenue. Came in August 2025. Cash jump, AR build, AP. Yeah. And this is a demo account. There probably some things around this p and l that are a little little strange, but we're gonna keep, we're gonna keep we're gonna keep rolling. Oh, actually, it's even asked us, do you wanna save this as an Excel file with proper IB formatting and dig it to another okay. Yeah. Why don't okay. Yeah. Let's, let's create that XLS output. Okay. This is something let me bounce over to skills here for a second because I wanna give you my, like, hot take on just, like, how to how to deal with, like, skills generally. So skills are just like little, you know, prompts that, that explain how you want something done. Like, I am super opinionated, right, about, like, how I want, like, Excel files to look. And that's just because I, like, I had this drilled into me as investment banker. Right? Everyone said, like, oh, you gotta get the right colors and people will have very strong opinions about, like decimals and formulas and all that kind of stuff. So so just, like, just teach Claude that. Right? So so now I've got this financial modeling skill. Oh, and by the way, like, if you have a problem, like, how bringing this skill, like, just ask Claude to, like, make the skill for you. Right? And just, like, brain dump. This is what I wanna see. Give it an example of a spreadsheet. This is exactly what I want. Right? So so here it's got all of these things. Never embed hard codes. I want stuff as a single sheet. I want your input column to look like this. I want your general column progression to look like this. I want your header area to look like this. This is how you're gonna deal with periods and dates. This is how you're gonna do the color coding. This is all this is all number four. This is stuff I would, like, drill into new analysts all the time on the team. Right? And now we're just doing the same thing. We're just, like, teaching Claude, or teaching Claude, how to, how to do it. So we're gonna let that we're gonna let that keep, cooking here, and it's just pulling and then you're gonna see how how that, how that pulls up. So let's do something else. One one, one pattern that's been really helpful, and I'll show you how we build it, is, is creating a little creating what I call creating what I call actions. Right? A pattern is you've got something in Campfire or you got something in from some some systems. I wanna pull that out, and I may want some, like, judgment around it. I may want, like, an LLM to look at it or make some assessment to it, and then I want that system, like, surfaced, surfaced to me. Right? So so I created this little, CFO, CFO summary. And and these GitHub actions are just what I think about is they're just like a little, they're like just like little workflows that will just run, at a particular time for you. And I used to I used to build these things, by the way, all the time in, like, Zapier and n a to n and, like, make and stuff like that. And those are those are great. But, but I may I may not want, I don't wanna have to go through all, like, kind of building all that stuff in in, myself. Right? I can we can work with Claude to build that build that directly. So here here I am. Now I have this, like, weekly CFO summary. Right? Revenue and profitability, how are we doing? You're in excellent shape. Revenue over the last nine months. Oh, we've got, like, a really, really good gross margin. That's the we got a crazy good business here. Cash collection. Moderate concerns around cash collection. We have this amount of accounts receivable. It gets this amount of cash. So your ARR, is not look at this. There's a big big gap here. So what what it's doing we can kinda see how this, this, this built. We can actually maybe go to our explorer here. Explorer is actually still building that that spreadsheet. Is let's create another GitHub action that, updates me on the AR that updates me on AR. Okay. Let's discuss. Okay. One thing you wanna do is you wanna use, like, plan mode. So if I hit shift and tab, you're gonna see down here it says accept edits. Right? Auto mode. Right? Plan mode. Okay. So we're gonna do it on plan mode. Everyone's had the experience of, like, using Claude, and then sometimes it just starts, like, doing stuff and starts building stuff. And you're like, no. No. No. Wait a second. Like, don't let's talk about this. Let's understand what you're doing, and then we can start doing work. So plan mode is like stop. Let's, like, agree what the plan is, and then we can and then we can put, and then we put something, and then we put something together. I didn't what's it sounds like the best use case for plan mode is, like, when you're gonna go perform a right action. Like, if it's gonna go like update a database, update, you know, update Campfire or some some app you build, like, we really wanna be careful. Is that the right way to think about point? think that's I think that's right. And I was actually doing that here in this, like, account I'm gonna show you this, like, accounting repair. Here, I'll show I'll show an example. So I was looking through the, there's a lot here. Let me let me I was looking through this demo account. Right? And I was asking it, like here. Let's go up here. Bear with me here, guys. Bear with me here, guys. Maybe this is a lot lot here. Don't be afraid. Don't be afraid. We're gonna we're gonna get to this. I was what I was doing what's going on with the balance sheet now with a massive negative deferred revenue liability. Like, I was going through like, I looked at the balance sheet, and one of the first things I did was like, hey. Does anything look off here with these books? And it was like, oh, yeah. I see you've got all these contracts. Right? And where, where would it be? Okay. We've got all these contracts, which are in revenue. Okay. We got all these contracts, but our contracts aren't really flowing through, into into revenue. And, like, it wasn't, we didn't have it like a deferred revenue line. Like, none of that was actually happening. Right? So that was a great case where it was like, okay. Let's create it. It creates a Python script. You go through a plan mode, and and let's see if I can actually find you where the plan mode is here. I did. I'm just going through this, going through this in in case. Right? And it was going through and actually, okay. Here's what needs to happen. Repost monthly rev rec j e's net result. Let me find the cash account and then execute. So what it was doing was actually, like, working with me and then patching up all these all these different all these different JEs. So you just think about it like another another partner of the team. It can write journal entries, as you need. Right? And all those journal entries actually, if I look at some of these, transactions, right, and I look at some of these oh, yeah. We're here. Hold on. Cash management transactions. And I should see an audit log on some of these. Actually, that's through that's through Nigel here. You're gonna look and some of these actually will say, like, Nigel via the by the API. Does that make sense, John? It it makes a ton of sense. We got a question, like, how did you upscale your finance team to use GitHub? You know, we looks like us, you know, elementary school vibe coders are still, like, just building local apps and maybe we're kinda doing a simple hosted app in probably an unsecure environment. You know, how did you get everybody to the current state? Did they come in with these skills, or did you just retrain. everybody? Yeah. A part of this is, like, getting people excited with, like, the it's getting getting people excited about, like, new tooling. Meaning, like, hey. Like, you're doing something that's, like, really this is, like, really pain in the ass way because you're using it like an like, like, Excel or you're, like, pushing uploading CSV. So, like, there's a there's a better way to do all of this stuff. So first of all, it's just, like, hiring for it because I knew because of the, like, the specific, like, back office challenges we had at at Masterworks, like, oh, man. And and I think, like, everyone in finance, like, what Masterworks well, I'm not getting any dev resources. Right? We're using dev for, like, the website and all, like, growth parts of this. It's not it's not coming to finance. Finance you on your own. I think everyone's everyone's kinda there. So I wound up, you know, hiring a lot for it. Right? I mean, I think we've got we've had, and there may be folks, you know, on the on on the phone here that that are are, you know, present and former master workers can can chime in there. But, oftentimes, we find people who would, like, have been in accounting, that have some, like, interest or proclivity around, like, around coding. Right? And, so that's a start. We've had people on our team that were, like, accountants, then they would, like, try to go be a software developer for a little while and then, like, round tripped back. And we, like, we love that type of hire. So part of it is, like, like, teaching people on on, like, the art of the possible. And then and then these AI tools are like man, they are incredible teaching tools. And that was the first that was the first, use case I had was, like, was, like, AI's tutor. Right? It was just, like, take and and I think even in my, like, I I created, like, a custom GPT that was just, like, teach Python for Excel nerds. Right? So I do something in Excel. I put in, you know, I create these schedules. I create a balance sheet. I create an income statement. I create these, like, like, assumptions. Like, just tell me what the same analogies are in Python. And then you realize, like, oh, you know, it's very readable to the through the same thing. So, I don't know. Does that answer your question? Anything else? Sorry. I'm not looking at the chat, so you're gonna have, to you're gonna got the chat. Just just keep jamming here. I'll let. you off another task, and then we'll ask another question. Yeah. Yeah. Yeah. And, we're kinda bounce bounce bouncing around here. So rebuild our balance sheet. Okay. I that's can you see that? That's actually not bad. That's actually exactly how I wanted it to come. Right? It's using my denies blues. It's using exactly so, like, that's that's, that's pretty good. So this is the type of thing where, like yeah. There is. I'm just I'm touching Excel a lot less. Like, I'm just touching Excel for, like, tweaks. Like, oh, is this assumption right? You're using it sort of like a canvas. I just feel like the days have actually, like, oh, I'm gonna sit and, like, build this model from scratch. Yeah. Like, that those days those days might be No more work those days workbooks, at Masterworks. Right? No more blank workbooks. might be No. No. And and bear with me. I I'm even I sort of I became a Google Sheets convert, by the way. So I like even my, like, pure Excel skills are getting getting a bit rusty. Oh, I like and especially Excel and Mac. Like, I don't know anything about Excel and Mac. Like, I I moved over to, I moved over to Mac a couple years ago and have never, have never looked back. Okay. I was gonna create oh, okay. We were over here. We're gonna we're actually gonna do something. Right? Okay. One of the coolest things about Quaca is, like, clarifying questions. I love clarifying questions. It's like it it's it's one of those, like, magical parts of using these tools. Okay. So remember I was here. I was just trying to create this, this little, like, AR update. And anyone can can it could be an update for anything. I'm just using AR, but it could be something, like, very specific thing. So what's the angle for the AR update, discharging with the pull, what the LOM says, and what an update means? Dealer collector's call, aging snapshot okay. Let's do, like, an aging snapshot and trend. Where should it land? Now I I think you could oh, look. You can actually do, what's really cool, you can multi select too, which is kinda cool. Why don't we just do all of these, all of this kinda update? Right? So Slack so I actually built in the Slack webhook. It's not as scary as it sounds. Like, you can, you just like working the back end a little bit. I I won't I won't, like, waste time, looking at it, setting that up for you. But, it it's just a way to, like, basically have Slack as another integration, you have here. Okay. So let's have this happen daily. So what's the angle? I want this aging snapshot. I want this in I want you to give me a markdown file. I want you to give me an HTML artifact. This basically means, like, I want something that looks like a website. I don't want this Slack, I don't know, Slack Slack webhook. Okay. We do a lot of stuff like this. And if you go back to this presentation, I was gonna say, like, this is like this right now is, like, what our Slack looks like at, like, at Masterworks. Like, this is actually we have this process where we have to update we're updating the values on all of our paintings, and we do that. So there are 500 paintings. You gotta do 500 appraisals, and then the navs all get updated. And so now this is all, like, a completely, like, Python orchestrated workflow, and and Slack becomes basically the the output about, like, when it's when it's running and, what what its progress meter is. So there's that. This is a cool one with that's more Campfire related. We Matchworks uses we use Goldman transaction banking, which doesn't, you know, it doesn't really integrate with anything. So we gotta create, you know, custom integrations. So now we have these, these, syncs. So every time we have to create a new account for a new painting, it automatically, it automatically syncs and integrates into Campfire. So we see that, automatically. And, actually, here's an example of, like, a journal entry, we just had created. So these are, like, more automated journal entries that I had to create with, some, like, more unique parts of our business, so how revenue, revenue is created. And so we actually have this few shared journal that's that's showing up as, like, Masterworks finance bot. So all this is right now is this is these are, like, Python scripts that are living in that, like, GitHub workflow. Like, they might be triggered might be triggered by a person who is to run them, might be triggered on that command line. That's where we're kinda moving towards. It's like, hey. Let's put more of these maybe in, like, a UI, and then Slack will kinda update people and as to the, as to the state, what their current state is. Okay. Let's go back, let's go back here. So, Claude's give us now a plan. Grass and Goat Global has two schedules GitHub actions today, this daily CFO summary. By the way, I'll update this. Oh, right. This is DFO CFO summary, daily OPEX dashboard, which we haven't talked to. Okay. AR is the strongest c f CFO level story. So or third recurring action that, needs is this AR, AR daily. Tracks trends, narrates what what changed, and gives nine to a live artifact to open during the webinar. Okay. Here we are. This this one actually knew we were doing a webinar. Okay. So what it's gonna do, it's gonna create this, like, one Python script, and it's gonna create this file called a YAML file. I'm gonna show you at the end, like, how I start, using, using this around, like, budgets and models, and it'll be, like, the kind of the the last thing I work on because I think that's, like, that's that's kinda where I'm spending a lot a lot of my time, now. But you can think about a a a a YAML file is just like, it's like an assumptions page. It's like an assumptions tab, basically. And, what it's gonna do is it's gonna, this little automation is gonna grab stuff out of grab my AR out of Campfire. It's gonna look at that. It's gonna send it to Claude. We're just gonna use Haiku, which is like the super cheap model they have. Right? It's gonna have a little prompt that says, like, hey. Look at this. Here's what I care about. Give me a little update and post that to Slack. And, actually, I guess, we we asked it to do, this other stuff. So it actually will would tell you exactly what it's doing. Okay. So it's gonna pull these open invoices. Right? It's gonna compute the aging buckets. It's gonna pull the last nine days of revenue, complete our days outstanding, read prior snapshot. It's gonna give me an LLM narrative. It's gonna write this HTML. It's gonna post to Slack. By the way, guys, like, this is, like, this is all from this one line. Like, this is we we got to this. So this is kinda where we are with these with these models. Okay. Open questions. Okay. That's let's just let that sort of cook for a second, and, we'll see if, we'll see if that actually works. We've we've got, a you picking up. Yeah. We've? we've got a question, around, like, hallucinations and accuracy. You know, how do you in Excel, all of us would check the formulas and you can kinda you can see under the hood. Here, obviously, it's a little harder to are you, like, spot checking? I think some folks seem to be moving away from just, like, asking it to do financial work via MCP, and they now have it right Python, and the Python's performing the work. I think. where are you guys on that spectrum of, like, That is that's exactly it. And it's look. The validation is no different than what you would do when you're working with an analyst. You're working with junior people on the team. Right? It's exactly the same. Right? You've gotta create some visual affordance for you. Right? You're gonna get that model. You're gonna send it to your investors. You're gonna have a little check line underneath the balance sheet that says, like, does my balance sheet balance? So you're gonna create all the same little, crutches and scaffolding, for Claude to be able to validate, to be able to validate itself. So listen, this is where, Excel still is is very helpful. Inspecting, like, the Python code itself, like, that can be that can be useful, and it's actually not as, it it's not as, like, terrifying to actually think of. I'll, like, get into what these things actually look like. But, this I I created I and I'll get into this. I created a little, I created our financial model, but I like I did it in Python instead of instead of in, instead of in Excel. Right? And what, you know, what what winds up happening then is, like, you can see this is, like, mostly pretty, I'm trying to see one that's, like, really, like, easy easy for y'all to, like, kinda kinda follow because some of them some of them are are, like, harder harder than, harder than others to, to follow. But most of the logic is, like, very, very readable. So we're, like, grading labels. We're getting these are where our cohort labels are. This is our cohort matrix. Our our MRR by month comes from here. So it's like and and, obviously, without getting all the logic here, like, it's the same it's the same, like, it's the same as if you were trying to audit, like, Excel formulas. I find Excel formulas are, like, a crazy hard to audit. Right? Everyone's been in in that in, in that boat. And by the way, taking taking Excel formulas and just copying and pasting them, like, into Claude as a way to to, to audit is, is is super helpful too. And also Claude itself, like, you can give it a full working model, and, it can actually parse that pretty well. Because I actually just realized the other day, this is, like, a totally, like, random fun fact, but, like, Microsoft Office files are just zip files of XML. So what Claude does, it actually just unzips the doc file and has a bunch of, like, XML, and it just parses and looks through the XML. I just thought that was really cool. I never never actually knew that. Woah. Okay. So look. Here is our daily AR update. Alright. So it's got our AR, days outstanding, open invoices. By the way, so this is this is like was basically just one shot with this. So huge amount of validation. Right? So I'd be looking at this. Let's tie it back out to what we have in Campfire, etcetera, etcetera. I I still find, like, this commentary. Commentary is like, it just needs to have enough context. Sometimes the commentary can be be a little, kinda be a little irrelevant. But here now our AR update is, is firing. So and we should actually see it. Let's see here. We're in here. If we look at our actions, is it there? No. No. I don't actually even see it up there. Oh, maybe because we haven't we haven't, actually actually pushed it pushed it up here. Open the Atlas demo and outcomes dashboard. Oh, man. What is it? Wait. Hold on. Let me see what we did. Okay. Oh, open, next steps. Commit the new files. Okay. So let's, yeah, let's commit those to main. Alright. Which just means, like, we create all these files, and now we're gonna, like, push it up to GitHub. We're gonna, like, put it in our put it in our, like, cloud store box. There's a whole, like you can get a crash course on, like, how to work with, like, GitHub. And and, honestly, I go back to the idea that it's, like, when in doubt, like, just have Claude explain it to you. Claude does a very good job of explaining to you. Okay. Reports, AR. Okay. So let's go to our AR report. By the way, you see I'm working in these, like, in this, local file, and people are probably thinking like, oh, man. Like, how is that gonna work and scale? Like, you can you can we use Google, workspace at Masterworks. And if anyone uses, like, the, like, Google Drive sync, like, Google Drive sync will just mirror your Google Drive on your local machine. So you don't have to be like this, like, project folder. It doesn't necessarily have to be on your local machine. It can be into a synced drive that works with, with Google Workspace, and that helps with with with collaboration. So let's take a look at where this is. This says, Campfire. Campfire. Let's see. Campfire AR report. Oh, okay. Oh, that's a that's an AR report line. Hold on. Where is this HTML? Reports? Oh, reports. Oh, here we go. Oh, Yes. okay. So everyone's probably seen stuff like this before. It's just giving me an HTML version of it, but we're, like, we're up and running. Commit the new files to the workflow dispatch. Let's see here. And if I refresh this, maybe this has come up. I don't know. This doesn't come up yet. Alright. Well, like, committed I don't want me to push her. Okay. Oh, yeah. Yes. Let's do that. Alright. It's like a whole science to, like and people who have any engineering background are probably sort of chuckling at me, but but, there's a whole thing. I'm sort of a caveman when it comes to it comes to sort of get actions. But eventually, it'll it'll get, get up there. Quick question, Nigel. How are you securely. sharing this with your team? Like, are you publishing it, on a, like, a AWS environment? Is it just kinda, like, being shared on a local, kinda, offline server with folks? Like, if you were to take something like this and you want the whole finance team at Masterworks to be able to see it, What's your it. We will yeah. We'll loop we'll loop it. Like, right now, like, seems all you're looking at is you're just looking at HTML file. It's just, like, sitting on my local drive. So. this is not available anywhere. The GitHub repo is private. So none of this is exposed exposed publicly. So this is probably where you're gonna wanna loop in your engineering team. We've got a, we've got an enterprise wide sort of admin, admin dashboard. So we can host it on our same, in the same place where we're hosting all of our other corporate dashboards, and it's all in a secure a secure environment there. So I think this stuff I I think the moral of the story with those kind of stuff is, like, this is great for, you know, you're almost like a you're like a product person. Right? You wanna design the product and and and the PRD, like, exactly how you want. Like, what do you want it to look at? Does it make sense? You can validate it. Right? You can create these create these prototypes. You can there's a lot you can do, like create create the GitHub workflow. But, ultimately, you're gonna you're gonna loop in you're gonna loop in some from your end's team, just to get it to get it hosted, hosted, securely. And then how about audit? Like, I know you guys are building so many apps yourself. So do you have any any, like, audit challenges or any best practices that you wanna share with the group? Yeah. Yeah. No. We we have when we first I I remember having an audit discussion several years ago, because our auditor was like, woah. Wait a second. You got, like, the finance team are writing these scripts, and they hadn't they hadn't really seen anything. My general view around a lot of this stuff is is the logic is, like, way more is way more auditable. Right? Because and especially now. Right? I I more auditing how everyone said I've been on the phone. Like, it's impossible to audit a spreadsheet. Like, spreadsheet I mean, there are mistakes in spreadsheets all the time. Right? If I find a hard coded variable, like, hard coded number, like, in in in a spreadsheet in a formula, like, makes my blood boil. Right? So so code, I think, is inherently more auditable. And, and also, like, the models are good at if you wanna take that code and then and then extract out, you know, it's read me files or documentation. Like, it's all it's all pretty good at that. I I mean, I have my own views about, like, where audit is going generally. I mean, the auditors are eventually gonna catch up, and they're gonna have their own tooling, tooling around this where I'm just gonna be giving I'm just gonna be connecting camp up to the audits, you know, to an audit bot, PWC or whatever, and they're gonna do their work. They're gonna do what the work that way and hopefully a lot less painful for for all. of us. We have big four auditors building into Campfire, and they're actually instead of doing, like, a sampling of data, they're doing, like, full population audits now. I'm like an autonomous, like, agent to agent. So I think where you're where you're describing is is what we're starting to see. It's very early, but it's, it's on the way. And we know we have about nine minutes, Nigel. Any of the things you wanna share, that we should make sure we cover before we. Let let me just do, like, a lightning round at the end, of of one of the things because I wanted to, I wanted to to show you one other thing I I've done here. I really like the idea. See, I've got this section called budget. So I've started to actually basically decompose everything I would put in a, you know, like an Excel file, into into, like, just a little, like, Python module. Right? And I think, what, what's great about this is you can what I do is I have these you can think about these, like, config files. These config files are, like, like assumptions. I'm gonna take everything in my file. I'm gonna take everything in my model. I'm gonna create all these assumptions. Right? So I got assumptions about, like, cohorts. Is it for, like, acquisition of a new new customers, retention, pricing, cogs, my headcount. Right? And then instead of, like, going in Excel and, like, plugging all this stuff one one by one, think about all these different p n py files. It's like, it's like tabs. Like, they're like tabs in your model. I got some for our actuals. I get some for the assumptions. I get some from the balance sheet. I get some cost forecast. Right? And I've and then now all of that all of that logic, I've started to translate into into code. Right? And the reason that's valuable is because, like, the quantity is just it's so much easier for Claude to just, like, manipulate the code, and I can iterate on it so much faster from, like, oh, man. We forgot to, you know, add in this business segment, or I wanna model this particular thing this way, or I need this type of sensitivity. Right? So now you have something that's, like, way more kind of interactive, that that you can, that you can work with. And, and so this one so let's see actually, okay. So let's let's run the current model and give me the summary output. We'll kinda do that, do that here. But what I've done here is so I have all this, and this is, like, great. I I may have an output. It'll give me an output maybe in the form of, like it'll run the model and oh, I don't have oh, that's why it's because it's on my, it's on my it's on my machine. But the model will actually let's see if I have a version of this, investor model. Alright. So so I went through and, like, created that whole model. And so you can think of, like, all of the logic actually sits there in in Python. But then to your point, like, I gotta send this to an investor. Right? So clients become very good at saying, like, okay. Take take what, take what you just, we created. We iterated it. Right? I think, if I actually look at, it's Masterworks. It's the San Francisco thing. It's actually running here. Let's see here. Oh, it's actually running here we go. Print. So run complete. So this is actually our model. It's running it's not printing out really well, but it's, like, our revenue, our cogs. It's got margin. It's got air. So it's, like, all the logic all the logic is there. Oh, it's actually written these things. It's actually written so it's it's outputted. It's outputted all this stuff into a couple of CSVs, and then it's created this in it's created this investor model. So this investor version of the model, like, it's a totally it's a totally working, it's a totally working model here. Right? And it's like, I've, you know, got it formatted in the way I want, and I want these nice little blue lines, and I want grid lines, and so so everyone's got their own, like, design sensibilities around around these types of things. But, but it's actually a working model all the way through. Right? And I wanted this stuff in red and etcetera etcetera. And maybe and maybe to your point to the question that came up earlier, like, how do you do, like, validation and checking? Like, have it add, like, put in check lines. Have it have it have it do check lines the same way you'd be checking in, checking in analyst analyst work. So that's that's a pattern I think I'm really, I'm really enjoying, creating. And let's see. I thought we hold on. Oh, yeah. I actually created here. Once it's in Python, I used something called streamlet because maybe if I wanted to iterate on all of this, it's like I don't want I'll I'll just refresh this. Right? I don't wanna iterate on like, I could just chat and have it, like, reprinted back to me, but maybe I want something that's, like, a little more, interactive. Right? So Streamlit is, like, is basically a business well, business valuation tool, sorry, business, visualization tool where I can take all this stuff from, Python and just, like, output it into this, like, nice little dashboard. So I've got, like, these MRR by cohorts. I've got my AR decomposition. I've got my PNL. And I think I think these all, like, these all should update if I slide. Some of these gross retention, net retention. Oh, yeah. Some of these some of these will start to move as as they start to move these is start to move these dice. I've got different cases in here. Actually, I just have a base case that's running. I've got my SaaS metrics. I've got CAC LTV, magic number, CAC payback. I've got our cash. Very nice business. We like we like this cash. Got our burn runway. And then I can and then I can export this. Export, like, to, to an investor investor excel. So this is something where you can start. It's like, say you have somewhat really gnarly giant model that's in, like, a thousand different tabs. Like, dump that into Claude. Say, hey. Like, let's create a Python module with this and and and use these YAML files and and stuff. It create the Python module. Then you have the engine actually works in code. You can iterate on with code. You can iterate here with something like this or you can or you or you can just talk to it, saying, like, these these don't work. I need to see this. And then and then, Claud, is when it comes time, if you need because we can create a dashboard if we need, like, another one of these, like, nice little, dashboards here. Like, this is these are an example of, like, dashboards we create at, at Masterworks. Like, this is an example. This is just all, like, fake data, but something we we create and we host we host internally. But this isn't this is, this is static, of course, where you have something here, is is a little more live. We've? got a we've got a question, Nigel. This is super helpful. Thank you. Thank you. Before we run out of time here, someone asked, like, you were on NetSuite before Campfire. You know, why make the move, particularly given, like, both have MCPs? Maybe this is related to your kinda hot take at the beginning, but, why why go from NetSuite to Campfire given, you know, everything we saw today? Yeah. I I think the master's team, we struggled in kind of working with the API, and part of that could be a part of it could be a knowledge and skills gap on on our part. Part of it is I just I I don't wanna pay another consultant a $100, to sort of figure figure that out. I think in in camp for us came about at a time where we we realized that, like, a a lot of the friction in getting to the level of automation that we wanted was basically bumping up against the, you know, like, against, NetSuite. Like, it was where there were some CSV upload. There were some, like there were a bunch of, like, manual steps that were causing a bunch of friction that we realized, like, ah, if we just, like, rip and replace this one thing, I think we'd be able to unlock a lot of efficiency gains. And, and the timing and the timing was great with what John and them are building. I think we had a lot of alignment around, around the design philosophy, around how they were using, Ember, how they were designing the app itself. It was just, honestly, it was it was a nobody's bingo card to do an ERP transition last year. But I will say and there's a little commercial with John, but but it's totally true. It's like the first ERP transition, like, our team was actually excited to make, because I think they they saw the potential for, basically, all of this sort of code level stuff to really, like, you know, really, play very nicely with the ERP rather than being more of a rather than being more of a struggle. Yeah. And, I mean, just to echo that, I thought your team would be purely asking for more API endpoints. I was like, wow. These people are never gonna use our AI. I've been impressed that to the point you made at the beginning, a lot of your AI workloads are moving into he he referenced Ember. That's Campfire's own agent platform that lives within the ERP. So from an audit perspective, from a from a full context window perspective, it just has has all that richness of data. And so I think Nigel's team has been, we've been pleasantly surprised. They've been, like, moving over a lot of the workloads into our own custom agent platform. And then anything that is outside the system that doesn't make sense across platform, They're kinda focusing their own efforts on things that, Campfire's agent platform is not. picking up for them. So Totally. And there's a feedback loop with with John and his team. I mean, we'll build something out. We'll all have, like, Vibe something in the weekend. I'll be like, oh, like, build a custom agent for that. I'm like, oh, yeah. That's we probably should do that. well, I know we're wrapping up on time. I wanna thank everybody. Yes, for sending the recording out. If If you didn't get any swag, you know, go ahead and and drop us a note, and Nigel's hat is the swag from the first session. And, Yes. thank you. Thank you, Nigel, and thank you. Thank you all for for joining us today. Thanks, everyone. Alright. Take care. Have a good weekend. Bye now. You too.