Name: SupERPower Hour Session 3 — Agents & AI Updated
Uploaded: 2026-04-14T18:31:03.325Z
Duration: 59 min 58 s
Description: SupERPower Hour Session 3 — Agents & AI Updated

Transcript for "SupERPower Hour Session 3 — Agents & AI Updated": Good morning, everyone. Welcome. Good morning. Good afternoon, everyone joining us from around the world. We're very excited you are here. It's the Campfire team, coming to you live from San Francisco. I'm here. I'm Irina. This is Lauren doing all the magic in the backstage. We're from the marketing team, putting together the vycoding series for you this month of April. We're going to give it a few minutes. We got a lot of interest from people in this series, so, let's give it a minute to get people settled in. So while we wait, drop us a message in the chat. Let us know where you're logging in from. We'd love to know where everybody is. Hi, Catherine from Maine. Welcome. Michael. Oh, so many of you. Going so fast. New York, France, India. May got people from from all over. Local day. Yeah. Welcome. Oh, Chicago. Oh, Sao Paulo. Hello. Welcome. Cancun. Wow. Amazing. Hopefully, you're on the beat or something. So excited you're here. So excited you're here. Thanks for joining. We got thousands of registrations for the Vibe Coding Series, and so, we're excited you're here. We've already done, two of these. It's our third one. I'm gonna start in a few minutes. Alright. Let me start by sharing my screen here. Give me one moment. Okay. Here we go. Let's go. Alright. So this is Superpower Hour. We've been running these every Friday from, for about an hour. The super power hour series is a practical, live session, where we share with all of you how finance teams are actually using AI tools to modernize their workflows. It's an extremely practical and live session. We have done two sessions so far, and, session one was really to get you the to the basics. And then session two, which we held last week, we actually invited our customer, Replit, to build with us, and we built an app together live. So both sessions have been recorded. They're available on demand. You can watch them anytime. And, also, please share them with your colleagues. If they can't didn't make the past ones, they can watch them on their own time, and then they're gonna be able to watch this one also today. We're gonna share with you the link in the chat box the chat box in a couple of minutes. Today, we have, John Glasgow. He's our CEO and founder. He's back on stage here today. We're gonna going to dig into the AI agents and how to how to run them. So, John, going to pass on to you. Can you tell us what are we going to build today? Yeah. It's a great question. And thanks all for joining today. Really excited to be here. As you can see on the screen here, we're gonna start very basic on on, scheduled tasks, and we're just gonna learn how to build a scheduled task in Claude. We'll be on the desktop version of Claude today. And then we're gonna work on a a few tasks, whether it's browsing the web, whether it's going through my desktop, helping me get organized for the week, recommending things, I should be scheduling. And then we're gonna, go look through our vendors and and our contracts and look for things up for renewal as a finance leader, here at Campfire. That's one that's been been helpful for me. And, daily spend report, this is just another example of a way we can have it take a look at items for us, flag things, and then, Campfire's own AI. And we're gonna, of course, talk through a little bit about when when at Campfire do I use Claude, when do I use our own AI, pros and cons of each. And then, Irina is gonna be super helpful on making sure we get the q and a. If we have time, we'll build, build a task together as scheduled agents, off the off the q and a and, just make this super interactive. So this time is really for you. I'm doing this for all of you. And so so many folks have just been saying, like, hey, you know, you're automating so much. You've created so many agents. We would just love to learn, and so that's why we started doing the series. So we're gonna go ahead and get started. So I'm gonna screen share, and we'll we'll kick this off. Okay. And just confirming, looks like we can see my screen. Okay. I'm gonna share the whole screen just to make sure anything that's interactive is gonna be coming through. And the first area we're gonna get started with is scheduled tasks in cloud. I've got a few that I've that I've teed up. A couple things to note. It does require a computer to be awake and so a little scary for all of us, but keeping it awake, you know, maybe you go get, you know, glass of water in the middle of the night and you walk by your laptop and it's on and it's doing tasks, but it it it does need to be awake to perform it. Maybe just for security reasons, you don't wanna do that or we're not quite ready for that yet. I just keep it turned off, and then, when the the machines fired up, it'll go ahead and perform the tasks. But I think the the first thing we're gonna do is, just have it propose something. I think so many of us just don't know where to start. And you can hit the new task button and it's gonna walk you through the structure. What I actually normally do is, I I start with a normal conversation that I'm having. So we're in a, just a regular co work item here and we're gonna ask it. And we're we're just gonna say, like, what is something we should be doing? And now it's prompting, you know, should we create a scheduled task? I I usually start by getting it really well refined before I create a task, for it to be scheduled. So look through everything and just take a look through everything here. I got it here. I'll just go ahead and create something a little more structured. And I asked it to preview before scheduling. And while that's running, I'm gonna start to talk through some of the things that, personally, I have found, Cloud Cowork was not able to help me with. It's actually quite good at going through, like, an email inbox and helping you perform tasks, but, our security at Campfire, doesn't allow Claude to take over, and write and move things around, in our email. And so, you know, really, a a a lot of this, you're gonna find some security limitations if you're if you're in a corporate environment. There is a browser agent that Claude has. It's a Chrome extension, that I that I use and love. And it's going through the Internet and even kinda uploading invoices to procurement portals and helping with some tasks. But Claude scheduled tasks, at least to date, you know, I asked it and, you know, I I think whenever I try and find a limitation, I just ask it. And so, you can see this one here, we're just gonna run now. So can you run it now? But what we're gonna do is take a look once it's going. Weekly aged items. Okay. This could be a good one. Can you show three items? Because I think one maybe isn't enough. You can have to do five or 10. I used to write really long prompts, but when you're when you're doing scheduled tasks, I found that really long prompts, it actually has to reread the whole prompt every time. And so, it does consume a lot of tokens and I was starting to run out of tokens every day, in the account or, you know, on the overages, it can get quite expensive. And so I've actually started to shorten my prompts, but ensure all context is there. You don't need to write clean sentences. You can see things like this, just keeping it short. In the beginning, feel free to just keep it free flowing, but I've evolved to a state where kinda minimal, context is required. And so, we've got three that I really like. So we're gonna go ahead and just create a scheduled task. And again, we can go into schedule and we can use the structured scheduled creator, but I like to just chat with it, get it in a good spot and then schedule something. And that's the same, within Campfire's agents as well. I've spent a lot of time, like, refining and getting things right, and then I'll say, great. Now let's create an agent, once we've really gotten it right. Because historically, of course, I was, like, having a great things, I would check it and then we would have a lot of back and forth. It is gonna ask and it's gonna prompt you if you're connecting third party systems, you're gonna get prompted a lot and you can make it auto approve. I'm still a little hesitant on auto approve, so I usually have it come to me. So we're gonna go ahead and and approve the schedule, and it's setting that one up for us. And while this is running, Rena, anything we wanna we wanna hit in the in the q and a? Yeah. Absolutely. One of the questions we've gotten is about how to stay organized in co work because you're using co work to manage so many skills. So how do you organize yourself? Like, how do you project manage that? You know, do you organize it by month? Do you organize it by skill? How do you how do you stay organized? Yeah. When it comes to being organized, Claude has projects. And so you can go in and you can create a project, and then that's where we can group everything together. And then Claude does do a good job. You can just see we just created the three. I'll just go back real quick just so people can follow along. It proposed three things and then I did have it create a schedule for all of them. And so we now have, there was a couple other ones I had already created, but you can see all three are now here set to run. And maybe my Monday automation idea, we don't want it to run at 8AM anymore. Maybe Monday mornings are pretty busy. So we can go here and, and we'll make it 10AM or when we wanna move it to Tuesday. There's a couple other options. In terms of the model, our internal mantra at at Campfire and and I use this, as well is quality of output then speed then cost. And so I generally keep it running on 4.6 opus 4.6, the most powerful generally available model. If it's a lighter task, you can always move it down. But it in particular, if you're early in your journey, you're not hitting token limits yet, then, keeping in the most powerful model. Otherwise, you may see the duration in quality of output and then, of course, that can get a little frustrating. So we'll go ahead and save. We're gonna run that on Tuesdays even though it's called Monday morning automation. And it can also go through the web. It can browse news. So what we can do here is we're gonna try out the actual scheduled, task builder. So brief me on news. And I typically don't use much of a description. They're pretty self explanatory, but it is required. So now let's say we wanna, like, stay up to speed on what's going on. I am a CFO for an accounting software company and I wanna see, like, funding announcements or things that's going on, and I wanted to just browse the web because maybe some of the publications I subscribe to, don't have it. And we want this to run and we could do daily. That's that's kind of a lot. So we'll do weekly and I got a little too much going on on Monday. So we're gonna go ahead and save this. And so this is the other way to create items by creating that new task And I can always trigger a run now, if we wanna go ahead and see it right now. So we'll go ahead and trigger that. What's another question that we're seeing here? Sure. Let me pick another one for you. Question about, you know, put you know, potentially running out of tokens. Right? You're using Opus 4.6. Do you think it provides enough marginal benefit to use it all the time at the risk of running out of tokens versus SONNET or IQ? I think if you're new to all of this, I would stay on Opus 4.6. And the reason for that is, like, the last again, on, like, quality, speed, cost, like, that kind of order of prioritization. The last thing you want is to be getting, like, incorrect answers or things that don't make sense. For the reasons stated, it uses a lot of tokens and it also is this often the slowest. So if you're looking for, like, quick tasks, if you're looking if you're heavily running it, like, if it was an hourly prompt, you can see you can run them hourly here. Then obviously, you'd wanna be super sensitive. So I think part of it's depending on where you are in your journey. If you're heavily we're on an enterprise plan at Campfire, so I'd spend less time worrying about token consumption. But if you are consistently hitting limits, then, yeah, you're gonna wanna start to fine tune things and call debts more of like a expert mode. If you're if you're newer to all of this, again, I would just continue to stay in in the highest quality model because we're we're on that area we're on that over prioritization around, like, let's just continue to focus on quality. Cool. Do we have time for one more now or do you wanna proceed to Let me just show this output and then let's go go through another question. So, here's what we've got is it it prepared a briefing for me and this week's edition. There's some items maybe we wanna read. There's some sources, VC outlook. Looks like some some funding in the category. And so, you know, m and a trends. And we can, of course, always click in. And then again, we're we're gonna wanna keep fine tuning everything. So it can be frustrating, like, you know, I was even preparing for this session. I was working on a few sample prompts, and I was hitting some limits. It couldn't read some some of the software that we had. It didn't have access, like I said, in our Gmail to create labels. We were hitting some some limits. And so you wanna just ask it how to get through items and how to proceed, and you might need to get creative. But the the key thing is to keep going. So here's kind of one of our prompts where we're browsing the web, and it's, saved as a draft. So maybe we wanna send it to the team. But, yeah, what's next? Cool. I have another one here for you. How do you use or compare this to ChatGPT agents? I it's a great question. I'll defer to folks in the chat. I personally haven't used ChatGPT agents. I know Gemini. I know there's OpenClaw. There's a lot of great options out there. I think a lot of us are falling into what, corporate plan that we're on. Obviously, in your personal life, you can go do anything. But for me, we're on a company wide, cloud account and so that's where I've I've focused my efforts. I personally have not tried out and done any sort of benchmarking against the two. And, you know, John, maybe as a follow-up to that, have you found that there are tasks that Claude is really better and good at or has not or other tasks that he's just not been able to do? There there are definitely some tasks. So the comment earlier about, like, having it browse the web as a scheduled item, I've not been able to do that. We have some security limitations on some of the software here, rightfully so, that it doesn't have access. I would really make sure, like, when you go in, and again for those on the desktop, we can go into settings and we can go into connectors. The, the app store is pretty robust at this point. You can browse the connectors and so this is a good place to to add more. You know, like, we wanna add Campfire. You can just go add add a Campfire account. Most of your vendors will probably be in here. You can always create a a custom connector. And so if you don't see something in the store, then, go to their website or just Google it and see if they have, an MCP, but it's maybe it's just not in the App Store and you can it's a little more work, but, you can manually connect items. And then that will allow you to have much broader coverage on your tasks. In terms of you can see here, it's, you know, running through financial software, and that's really allowing us to, make it much more powerful. So it's connected to our, you know, email account, again, from a read only perspective. But, yeah. It's a great question. Cool. Let me know. Are you gonna proceed or do you wanna answer one more? I'll just keep showing a few more things just for the, for the group that wanted to see what we had built. I know there was we wanted to build the cash burn forecast and so it's gonna go through here. And, this is one that it's just gonna, like, keep reforecasting. You can see it's creating a memo, but maybe we wanted Excel or we wanna slide. You can just give it any format that you want it to be. I just wanna show a few more items, that we've Monday morning spend report. And so we can go ahead and this is the one that it just created. We'll go ahead and kick that off so we can go look at it in a second here. And I will This one already ran. So let's take a look at this one. And I'll just pull that up here. But it wrote it wrote a memo for us, so we're getting a daily memo. Oops. Let me go back. We've got an Excel file that it's building for us, and you can see that here as well. So again, any format that you want, you know, we're getting a daily snapshot of spam. Next question. As I pull up the next tab. Sure. There's been a couple of questions about, you know, safety and how safe it is to use cloud. So I'm I'm seeing two questions. One is, you know, is it safe to connect, you know, your kind of human resources tools and your payroll software, Rippling and Gusto? Can you tell us a little bit about that? We have not connected our payroll software. I think there is a lot of sensitive data. Some folks are working in, like, group projects with other people at the company. And so, obviously, you need to be incredibly careful. To date, we've, of course, connected our Campfire account and then our our spend management accounts. I think to me, spend is, you know, corporate spend is a little less sensitive internally. Of course, payroll data is incredibly sensitive. So we're not we how I actually pull it is, at, out of since Campfire is connected to our payroll provider, then, I'm getting kind of anonymized and somewhat aggregated payroll data, via the Campfire connection and then not going, directly into our payroll provider where it is at the employee level and then much more sensitive in my opinion. So that's the approach we've taken today. And then I think some folks are also still concerned that Claude's gonna maybe do something dangerous with their computer, you know. Yeah. I mean, it it Can you to that? It's it's very real. So, like, you know, I can have Claude go in here. I'll just show you real quick. Can you go into my folder within downloads? It's called organized into those things. When Claude does delete things, I'll just show you here. It'll but you probably seen these on the web that, it does use what's called the terminal. If you've heard of it, this is a terminal. And, it's essentially irrecoverable when it deletes things, when it does it in this format. I think they've worked on making it more, user friendly. But I think to the to the points there, I ask it to preview all work that it's done before, I actually do it. So, like, here's just some sample data that I had put in, and we're gonna do I'm just gonna pause it here. And what what we'll say is, like, can you show me a proposal of how you're going to work? And I always ask it to preview the work. So I treat it like anyone else on the team. You know, it's coming back with, like, an unreadable format here as a Excel junkie. Can you put this proposal into Excel? I'll just have it drop it into Excel for me. And as a reminder, we covered this in the first session. The blue dot here means it completed a task. So earlier, we had prompted it to kick something off. You can have it run multiple tasks at once. So just to quickly show this one, you can see it just ran. And, so yeah. No new data, but we'll go ahead and and hop back to what we're working on. Okay. So that one's still running. We'll take a question while this is running. Sure. You know, John, I think some folks didn't attend attend session one, so they didn't see you set up your connections. Uh-huh. You know? So we're getting questions about, like, how do you set up, your connections ahead of time so that today you are ready to, like, just start running those prompts? It's a great question. Let me just finish this one real quick. So it's got two tabs, organizational proposal and folder tree. And I get a little unorganized on my desktop. I don't know about everybody else. So, you know, it, it proposed what we're gonna do with it. We got bills. Looks like there's a couple folders it's proposing. Here's the folder tree. Don't love that output, but, this all makes sense to me. And then I'm gonna go ahead and great, can you organize based on this? The good thing is you don't need to be great at typing. It'll figure it out. And so while that's running, I'm just gonna go ahead and show you how it's set up. Go into again, I'm on a cloud desktop, so I'm gonna walk through the desktop approach. You can always, just ask it by the way, like, how do I connect? For anything in the cloud when I get stuck? I literally just ask it. How do I connect my Campfire account? Or just how do I to cloud co work. And say I don't even know what an MCP is, which is model context protocol. It's a framework invented by Anthropic, but, essentially, everybody's using it right now. It's a little debatable on that. It consumes a ton of tokens. There are some questions and concerns about its security, but generally, it is the the leading framework in the in the market right now. If you're familiar with APIs, it's essentially an API for for agents. And so it's how agents and agents, connect to each other. And we'll go ahead and ask it, and you can see it's getting confused here. So Campfire is in your app store, I think, which I know it is in the App Store, but let's see if it figures that out. Oh, maybe I should have given a little clarity. Oh, here we go. So I found it. It's in our connected directory. So, you can always prompt it to kick it off for you or and we'll just kinda go through that flow. It's pulling me into a different window, but it's essentially just having me log in. And you can see we're now connected to Campfire. And now let's go look at all connectors. And, so you can see everything that's connected on this account. It's a demo account, but, Slack is probably a good one for folks. And then, like, Gmail, Drive, Sheets, Docs, things like that are probably good ones to have connected. But, when we're like, if you wanna just go straight to the page, bottom left, and then go to settings, and then we go to connectors, and then we have connectors here. You can see what's connected. If you're worried about it, we can go in and configure, and you can see you can manage. So maybe we want to block, like, for a read only perspective or you wanna block write access where it can, like, accidentally delete things in your ERP, which should, you know, be a nightmare. But we can go here and and manage, its provisioning. So maybe it can never look at the balance sheet. And some folks will ask us, like, hey. What if can I ask, you know, certain users of the company maybe can't see, like, you know, payroll data in the system, but they can see, you know, their own budget? So this is a way that you can go ahead and set things up. But make sure you have all your connectors set up and, like, similar to like, let's check out. Let's configure Slack. So again, you wanna go in, like, can it write, you know, does it need approval? When it says need approval, if you saw earlier, it was asking you for permission to perform a task and I was clicking approve. That's what we're seeing on the need approval here. You can always allow or you can never. And so if you are worried about it performing tasks, we will just move it to block for all of these. And then we're in a read only state. Maybe we don't have to draft messages either. Okay. So back to this task, all nine files have been sorted. Let's just go ahead and take a look. And here's that folder. Here's the proposal. And it went through and it organized everything for me. You could see ten twenty seven. This is a live task here in the Pacific time zone. Just looks like it finished four minutes ago. And we got our headcount plan. Now we can always have it just, like, continuously update for us. So awesome. Some people are really nice to AI. Some people don't care. I try and be nice, but, that looks great. Can you do this weekly scheduled task on Tuesday afternoons? And just we'll go ahead and just have it keep cleaning things up. Let's take a look here. While we're going and we're at the halfway mark, any other good questions we should hit? Or do people wanna see another specific task? Anything you've been trying to do that we could maybe just live workshop together? Would love to hear from you. John, I mean, you are a cloud power user. Can you tell us about your Campfire employees? How have they adopted the use of AI in their day to day? Can you give us some examples? Yeah. We've got and to the point earlier when it asked permission, this was that comment when we were setting it to always approve or ask for approval. See what's going on here. At Campfire, I've seen it candidly, Cloud Code I spent a lot of time in Cloud Code. I think it's gotten more user friendly, but I was getting stuck because it had me do some terminal activities and and dash and, you know, installs. Claude Coerc is sitting on top of Claude Code and so it was transformative for me personally when it came out. And so we did an all hands and I took one task for each department, whether it was the sales team, the implementation team, in engineering did not need any help. There were already power users, but everybody else. And I created a a manual task for each of them by either a piece of software, how to create a slide, how to do something, like a project plan for an implementation, and, showed in the all hands everything I built to show that everybody here can get a lot of value out of using AI. And then we launched a show and tell channel on Slack and everybody is showing everything that they're building and we're seeing, across all the the teams. Now everyone's kinda using each other's software that they've built, their slide formats, their scheduled tasks, their skills. There's lots of skills being shared internally, which is, like, a way of kinda prompting it in, like, this, right here. It's called a skill, And you can create, kind of, pre built actions where it's gonna have, like, a full prompt behind it as soon as you kick off a skill. And so I'd say internally, since then, there has been a transformative movement of AI's being used for I'd really push the team for every manual task, meetings repetitive. I want everybody working on strategic tasks. And so how I think about it is our headcount is 10 x in the last year, and if we get 10 x output out of everybody with AI, then we can kinda 100 x, our performance and capacity as a company. And so that's really the framework we've taken. And so, yeah, we're all on an enterprise plan. Some companies have launched token leaderboards. I know at at Meta, other companies in token leaderboards internally where you can see who's consuming the most AI tokens. I more look for called quality of output and not people, like, doing long running agents to to drive token consumption. Of course, there's an added cost if you're at the top of the token leaderboard, but it is associated with, being more AI forward if you're at the top of the token leaderboard. But to date, internally, we have not launched a token leaderboard. It's been more of the show and tell channel, and then that's been a way we've we've driven, adoption internally. And, I would love your thoughts on it. As a new newer camper, I'll put it back to you. What have you built? Wow. What have I done? Definitely, presentations for you, John. Yeah. Of course. I've, you know, sometimes I'm doing some forward looking planning, and I just kind of write it very in a very dirty way in a doc, but I need to present it in a very coherent manner for leadership so that we can make decisions very quickly. And so Claude, helps me organize my thoughts, and that's been a game changer for the marketing. Great. Amazing. Why don't we pause the task? So we we had to browse the web. We had to go through our, all our connected systems, our our, finance stack software and create a cash flow report here. We had to go on to our desktop and perform some some tasks. We had to ask for permission first. You know, we connected some some accounts together. Would love to hear if there's any other burning questions, anything we wanna build together. Otherwise, in the next few minutes here, we can go through, the agent platform built within Campfire and how I'm using it to perform finance and accounting workflows. Yeah. Let me go just go through. I have actually a lot of questions. You know, one of the questions we always get is about how to verify that the output is credible and accurate. Can you speak to that? It's it's honestly one that I struggle with. So I spent a lot of time, looking at, like like, sales forecast here. And so on Sunday, I was heavily using our CRM connector, via Claude and using co work here to have it build very kinda complex, like or not complex, but call it, like, deeper analysis for, like, q two, sales forecast. And, it's great, but I did notice there were, like, some fundamental inaccuracies where it it even said, like, one sales rep was, like, underperforming and I went and checked it and they were actually outperforming. And so it it can be quite scary. I I generally will go into the system itself. So, So, like, I'll go into the CRM in one screen and then I'll have, like, Claude in the other screen and I really check its output. Of course, it's gonna be hard to find everything. So what I'm often doing now is, and I know this isn't related to, like, the scheduled agents, and kind of the task for today. But maybe we I'm happy to just focus on this because it's so important for folks that, I will take the output from two different tools if available. So, like, your spend management tool and your ERP, your your CRM and your ERP, I will like, we from an audit perspective, we did, like, a prepared by client list and I had it pulled out of, Claude. Claude generated the prepared by client for the audit request and I had Campfire's agents prepare it. And then I actually took the raw download of the data out of Campfire, the, like, transaction level detail, and I dropped it into Claude. And I said, can you reconcile both of the files to ensure everything is accurate? And Claude ended up actually saying, oh, data was paginated on the API that I was using and I missed some of the, kinda, some of the data. But once it looked through the full dataset, it was able to send me a reconciliation report, to confirm that it then went back and went through the paginated data, and then it it updated the entire file based on that feedback. So trying to take different systems or raw sets of data and then get in to have it stress test the outputs has been the most helpful for for me. There's a lot of little hacks people claim that if you ask the AI to to take a deep breath. Some people say if you threaten the AI, like, I'll get fired as a CFO if you're wrong. There's lots of kind of prompts that people will add that they say has improved performance on the quality, but just more call it objectively, I I just, like, hey. You gotta tie it out to this. Can you reconcile everything to x, has has been my approach. And maybe just also tied to, like, outputs, do you have any suggestions on document version controlling when teams are collaborating on shared projects? Have you done anything about that before? Yeah. I think having it go into Google Drive, and then having it, like, not versioned but more of, like, update existing versions, has been one. I will say I get when I'm working out of, like, a Excel or PowerPoint, I do have a version control issue just like anyone else back in the day when you're on, like, version one zero one of an Excel file. That issue persists when I'm working, in, like, all, like, non cloud files. So that that's one where yeah. You just have to be careful if you're not working in, like, a cloud drive. Cool. Thank you. Yeah. But we can like, here's one task we can do together that a lot of folks will have, like, like, a weekly meeting that, you have to produce some sort of output and it's typically manually updated. So I'll just go into to to Claude here. Can well, actually, we're gonna take what we just built. So, let me go back real quick. I'll just speed this along for the group. Well, maybe I've I guess we've done a lot today. What I'll do here is we'll take this cash flow, spend I'm just gonna grab the last one. Can you create now, we did this in the first one, but you can create a skill that captures the, the formatting for your slides. If you haven't quite created a skill yet, the easiest way to to go ahead and do this, I will just upload yeah. I'll just upload something. Let's do I'll just can you create a slide with this output and use this format. This is just the easiest way to get started on it is just ask it to use this format. You can then ask it to create a skill with this format and you can, like, reference the skill. But for those of, like, just upload I literally just grab any file that's in the Campfire formatting and just drop it in, and it'll just kinda go to the file and pick up what it needs. So we'll go ahead and have it perform this task, create a slide. And then we you know, again, it's part of the schedule. Maybe you have, like, a leadership meeting at 1PM on Mondays and you wanted to send it to you 10AM on Monday. And so, hopefully, instead of you manually updating slide every week for some sort of recurring sync, this is gonna take over the task. I found it to be quite good at that. And I typically start with the previous version that I did I built manually. I will drop in the previous one and then and I'm like, hey. I want this format. I wrote the commentary. Can you take all of this and just recreate it, refresh it with the latest dataset? While that's running, we'll start to hop into Campfire agents a bit. So we're gonna start to hop around a little bit. But this Campfire agent platform, There are prebuilt tasks and agents that you can have it follow through. Like, I wanna build a report. We wanna build a dashboard, like a cash dashboard or something bespoke to our business. I'll just kick off a cash dashboard. And everything from socks, readiness, and auditing over to, permissioning and controls over to performing accounting tasks, it's kicking off, a full kinda dashboard for us. But similar to how I work out of Claude, I will start when in getting the right format and give it feedback until I get to the right format. So we'll just see what it comes back with. Gonna give it a little bit of feedback, and then we'll create an agent. And then I will have it, like, send me, again, a Monday morning Slack. And I'll just kind of a Slack dashboard that's going into, like, the leadership channel for looking at cash, the company. While that one is running, another task I like to do that I can just show the group here. Here's kinda what I was referencing earlier. I'll speed it along for the sake of time. Can you take the attached slide format and create a slide of just like a a weekly flash? And so it's going and there's obviously some messy, data in here just with this demo account. But I had to create a weekly flash, and let's pull up the flash. And so it took the it took the format I gave it. You know, again, there's kinda some some dummy data, so it's not the best. But, ultimately, it's gonna go into my account. It's gonna create this for me and drop it into Slack channel or kinda drop it into Amber, the the agent itself. But the key thing is it's a recurring task. It's gonna automatically perform this for me as opposed to me having to prompt it every week going into that that team meeting. Alright. So this one's still running. So we will hop into this one here. Looks like Claude's still working too. So we're consuming tokens across a couple systems here. Would love a question. Rita, what's top Oh, yeah. Hey. Here's here's one. What's the hardest thing you've done in FP and A and accounting? Like, the most tricky or error prone when using AI? 2026 plan, I did have it built, with with Campfire and, and with Claude. The most error prone part is AI still will make a lot of assumptions even if you would rather ask you. So I think, like, really asking it to to not make any guesses or assumptions on your behalf and come to you for everything. And then so it'll be like, hey. Like, headcount plan, like, where are they gonna be located? Because we need to make sure we're putting in, like, the correct overhead allocation on the headcount plan. I need, like, multiple offices. So you gotta tell me. So I think having it, like, ask me, like, anyone else on my team, was probably the most error prone part, but then also the most powerful because it was able to do lots of scenarios and it instantly gives me feedback. And, I think, you know, building a I built a 30 like I said, over the weekend, I built a 30 page q two deep dive on on our sales, forecast and all of the risks and all of the reasons why we're gonna crush plan, but also, you know, I think, like, at a at a rep level. And I don't even have it take, like, what the reps are forecasting. I have it come up with its own conclusions based on historical performance and based on its own weightings, of where everything is in the pipe. And so with that level of judgment, there can, you know, to the question there, can be a lot of, like, inaccuracies. And so then you really have to stress test it. You really gotta do what I described earlier where you give it, like, a raw dataset. You really gotta push it hard, and you most of us know our business well enough that you can kinda sniff out, issues, but that is something. And so here's here's the cloud one. And you can see they're essentially the same cloud. There's a little more data that it's looking at on this one because there's some more systems I connected it to. But, it kinda came to the same conclusion, same format, same updated. And that's a good point. Like, you know, when to use Campfire or when to use, you know, in product AI versus, like, third party AI. For browsing the web and doing things like, you know, brief me on the news or organize my desktop, obviously, I'm going to Claude. For really anything that's with financial data. We've seen a little higher accuracy doing it out of Campfire and just because the AI is sitting within the context layer. So I need all sorts of, like, 2026 plan, cash flow analysis, help me deep dive on q one spend. I'm going to Campfire whether it's an Excel file, a slide, or just like a a new cash dashboard, for for tasks like that. So let's hop back in. Created our cash dashboard. Lots of visibility here. And maybe I want, you know, again, I'll get a Slack ping every week. Just for the sake of time, I'll just say I'm happy with with where we landed. So great. Can you create an agent? Send this to me. Stay. I'll do Wednesday night. And we just you know, we're creating an agent now where this will one small perk of the Campfire one is that your system doesn't need to be turned on. So this will run asynchronously when my laptop's off. So I can actually maybe I'm at dinner Wednesday night, but I wanna look at it, like, Wednesday before bed before, like, an early morning with the team. Where the clogged one, if you do want that and you're logged off, then you will wanna turn on the, like, wake up my machine. So just kinda get to know the agents and how they work. But here, we'll go ahead and run now or we'll just schedule. And I know we got ten minutes. So what other questions do we have as we as we start to wrap up? Actually, just to elaborate because you talked about your q two analysis and then, like, forward looking, the forecasting. So people are trying to understand that. Did you give Claude your Excel model to run, or did Claude actually effectively run your forecast model? Claude built the model for me, v one. I've, spent maybe more than a year of my life building Excel files, and I'm done. So I think, Claude and Campfire take a first draft at every Excel as of maybe two months ago, maybe maybe a little less, and then I will fine tune, a draft presented to me. I mean, I'm the only person in the finance team at Campfire. We're north of a 100 employees. And so, I gotta really get a lot out of Claude, given the scale. And so, you know, I don't have time to start with blank workbooks these days. So it's really, like, first drafts are being sent to me, but I'm actually rarely the one editing them myself at this point. I just continue to push Claude and Campfire's Amber agents platform for feedback and and iteration. And then most of the oh, near final outputs are not even being done by me anymore. Yeah. And there was a question here also about how did you give Claude access to your pipeline data? Yep. Yeah. So, back where we were on the CRM connectors, in in cloud here. So I know we connected some systems. I went in, here into settings and to connectors like we did earlier and connector CRM. Cool. Do we have time for one more or a few more? Yeah. Let's let's do it. I'll just stop sharing at this point. I know we got eight minutes max. What else would be helpful? Yeah. You know, Jay here attended our event here in San Francisco in San Francisco. So thank you for attending the event we held with Avacom last night. And while he was there, there was some discussion on automating CRM workflows to improve forecasting accuracy. So he's asking, if our team is using Claude to make sales data more reliable for the finance teams, particularly regarding revenue forecasting and planning and and and what the impact Yeah. So that that, like, 30 pager that I built over the weekend was actually to better forecast. So, like, our CRM will give you, like, default, wait like, when I wanna look at weighted pipe, it has its own defaults, but I will have it use, like, a training sixty day. Like, Claude is now using a sixty day, average on pipe conversion at each stage of our pipe. And so it's, like, very fine tuned on actual business performance. And then at a rep level, like, how is each rep from a conversion perspective? And then it's actually going to that level, which, like, typically, like, it would be too much for a human to do, at least for me. I was at big places like Adobe doing this. And so then you kinda get incredible It's like having, you know, a thousand financial analysts on your team that are willing to perform any any level of granularity that that you desire. I really push it to get to that to that level. So it'll even, you know and some of it's in Excel. I'm actually starting to have it build software more of these days, and the software is sitting on top of the CRM. And so build a forecasting software that is then refreshing daily, based on any activity, and then it does send me, what changed. So maybe there was a closed one, which we always would have known about, but maybe, like, a deal value changed or a expected close date changed. We can always get visibility and ask the rep, like, what what happened here? Why did it get pushed out? So we have seen incredible increase in accuracy in our own forecasting by using AI. Yeah. And related to that, maybe there's some level setting we need to, you know, bring to our audience here because you've said a few times, like, fine tuning the feedback to the agent. So how how do you actually do that? Like, are you provide improving how do we improve your agent in a way so that you have better, like, more accurate outputs? Yeah. For me, it's mainly, I'll look at so, like, just to make this very tactical here, I'll just rescreen share quickly. Like, when I spent a couple hours over the weekend, just very, very tactically, I look at what it came back with and I say, like, I don't like, this format doesn't work for me or can you show more detailed views at another slide with every single vendor over five k? Right. And so I will just continue to iterate with it. And when I say fine tuned, it was literally just, like, constant feedback. So that that deck, over the weekend started at maybe a five pager, and I kept having more ideas. The great thing is you can literally have it do anything, and it has a limited amounts of time. So even if it was, like, I don't know if this is a good idea. Like, I had to build a slide on average rep time from email to meeting and even response times per rep. And it's going through, you know, hundreds of thousands of emails and meetings in our CRM and giving me, like, granular visibility on, like, which rep is the fastest to respond to a new lead. And so I think things like that, I'll just continue to fine tune the deck through the entire workflow. Great. Have some other tactical questions from the audience. You know, people that are dealing with invoicing coming through Gmail, have you done anything regarding that? Like, do you send them to a folder and then have Claude, you know, analyze those? Like, how how have you addressed invoicing? Our invoicing is fully automated, without me having to do anything in our email inbox. Our kind of spend management plus Campfire, there's, like, end to end automation, automatic payment application on paid items, automatic ingestion. So, personally, we have it. I think that is a good task, for for AI. So I would I mean, if I were you, I would create a scheduled task where, like, go through this, you know, AP or AR inbox depending on if it's bills or invoice, customer facing invoices, and just have it, like, work work through them. There's some great vendors out there that can help you with this too. So maybe check the market depending on your volume and complexity. Otherwise, creating scheduled tasks probably gonna be really helpful if it's lower volume. Cool. Let me pick a few more here. Is there anything you've chosen not to connect to Claude? The one earlier around, like, payroll, we haven't. But I we haven't really needed it because I more of look at, like, aggregate payroll, and so that's already in our Campfire environment. So I think I look at it like a department level. So I just get anything I need there. So if I look at, like, toll total for fully burdened department level costs, or by location or kinda any vector, you you know, to date, I haven't done any task where I'm looking at anything at, like, an employee level from a, like, a cost perspective. So that's one that we've not needed. But also from, like, a sensitivity of the data, I have not added either. Awesome. John, I've I've seen your your Slack workspace, and it's a very busy one. Like, do you leverage Claude for trying to summarize what's happening, in our, like, Slack channels? Just try to summarize conversations for you. Yeah. It is it is good at that. I don't spend a lot of time on summarizing. I think at least today, I try and read it all. Claude's pretty good at summarizing, but sometimes it loses some context, that I think is important to me. So that that is a good task, but, not an area of focus for me. Cool. We're coming up to the end of session three of our superpower hour. John, anything you'd like to, you know, tell our audience as we're concluding, this third session? I just wanna thank everybody for joining. I know in the beginning, we saw people from all over the world join. I know it's, you know, late, some places. My my big advice, get started if you haven't already. Ask the AI for help when you get stuck and don't start with anything sensitive if you're not ready. Have it send you a weekly news digest. There's nothing sensitive there. There's no security risks. Just summarize the news. Send it to me Monday morning, and let's just all start on our journey. If you're a power user, great. Let's have it build sales forecast and dashboards. And the good thing is these systems are always improving. And so I think every time I log in, there's some new feature or some new AI harness that it has. So really, like, try and stay up to date is, like, the other thing I try and encourage everyone because the performance is getting way better. The task can do or getting having it browse the web and go into procurement portals for you and perform all the tasks we really don't enjoy doing. It's it's here to help. So thank you all again for joining and hope that was helpful. Thank you so much, and see you all next week. Bye.