Teach your team how to use AI productively and securely (Panel Discussion)

Trent McLaren: Today we're moving on to our main event: teaching your team how to use AI productively and securely. This is our first panel discussion of the day and, as I said, the third most registered, so nearly 800 or 900 people registered for this session, which is awesome. A lot of you are going to watch this on playback, but I'd like to start off by introducing everyone on the call. We've got Travis Bacon, Beau Gaudron, James Rose and Gus McLennan. Welcome, all. I appreciate I've actually not brought enough diversity to this call, and that's on me. So if that comes up in the chat, that's on me, and I will take that on. If you're in the chat today and you'd like to be on a panel like this or contribute, we're always looking for good new speakers and panellists, so please let me know.

Trent McLaren: Travis, let's start with you, mate. Why don't you introduce yourself and tell us what you do?

Travis Bacon: Cool. So I'm a sole practitioner based up here on the Sunshine Coast, North Brisbane. Got a small team, offshoring in the Philippines a little bit. Claude's my main tool. Love that guy, he's part of the team, as I'm sure most accountants appreciate these days. I use Claude mainly, but I'm a huge believer in building tools and connections outside of it that don't depend on AI. So using AI to build and innovate is really where I do most of my work.

Trent McLaren: Awesome, very cool. Thank you, mate. Welcome, and good to have you here. Beau, over to you.

Beau Gaudron: Hello, everyone on the call. My name is Beau, and I work with Growthwise in Newcastle. My role there day to day is to look at systems operations, so how systems can talk to each other, and obviously AI comes under that umbrella. It's also making sure of exactly what we're talking about: how do we actually get our teams involved, make sure they're using tools, and make sure they're using them effectively as well. And then, kind of like Travis, what I also do outside of that is try and build things that will let our teams get back to the work they really want to do, rather than the things they don't want to do.

Trent McLaren: Awesome. Love that. Very cool. Looking forward to digging into more of that later. Gus, welcome.

Gus McLennan: Hi guys, Gus McLennan. I run a business called AI Dojo. We actually started life as an AI training business, and we'd built a platform for a large professional services firm down here in Melbourne. I'll probably share some insights from how we got them to onboard and really start using the platform we developed for them, and we've since gone on to turn that into a SaaS product and offer it out to other accounting firms now.

Trent McLaren: Awesome. Very cool. We'll hear more about that through the session. Jimbo, James, welcome.

James Rose: Hey, mate. Yeah, co-founder of Content Snare. We're a platform for getting information from clients and client onboarding. Claude Code has been my tool since about March. I have not been jumping from tool to tool like everyone seems to, and it's funny because now everyone's back on Claude. Anyway, I just don't feel a need for a lot of other tools. When it becomes your command centre, it's insane. It's connected to so many things at this point, and I probably spend 90% of my day in it.

Trent McLaren: Yeah, I've got it running right now, it's doing like 10 things.

James Rose: It's the same, it's doing three things while I'm in here.

Trent McLaren: It's wild. This sounds like I don't do a lot, but I'm normally busy, guys. I'm at my computer, I'm doing work, and I've actually had time to start making meals again, because I'm like, Claude, you're going to do all that stuff, I'm going to quickly whip something up, and then I'll come back. So good. Travis, I'm going to come back and start with you. Our whole premise for today is how do we help everyone and their teams and their firms adopt AI? I'll tell everyone else the rules. Change management is the biggest thing we've all talked about for the last 10, 20, 30 years. GST came in, there was change management, because we went to desktop. We went to cloud, change management, because we had to move software and learn new things. AI's come in. It's the same thing. What I don't want to give you all today is another change management session you've probably heard a hundred times, where you need a champion and you need to do training and all this stuff. That part is key.

Trent McLaren: And you do need to do those things, but I asked the panel not to bore you with something you've heard before. We want to get into the deep, nitty-gritty weeds of the practical things: what are you actually doing that has helped move the needle, the things that are sticking? And the flip of that is, what's not working, and how are we trying to fix that? Now, with that preface, Travis, you've mentioned Claude being part of the team. Do you want to talk us through your journey of adopting it? When did you start, and how has the team's adoption been?

Travis Bacon: Yep. So essentially I'm the powerhouse user, so I'm the one building all the tools and the little connection pieces and whatnot. The goal behind me doing that is to save someone time on a process that is mundane or repetitive. So what we did as a team is sat down and had a discussion around: what is it that you do every day that is repetitive? And we started there. Once we built something, a tool or a connection, or used an MCP or something, what we were trying to do is solve a problem. So we solved one problem, and that was where we started. The problem we started on was BAS. They're super repetitive, they're almost mundane for most accountants, so trying to solve that, speed it up and get some efficiency was the name of the game. That's where we started, trying to speed up BAS. And look, it was a hard slog.

Travis Bacon: Going from nothing to something was the hardest part of adoption. But I can tell you now, the team are sitting there going, don't take it away. I joked with them the other day, I said, I'm going to turn Claude off, and they were just like, don't do it. Please don't do it. So just start with something small. Very small, one piece, and then come back. Have that meeting again and go, okay, cool, we solved that one thing, what next? And just start trying to save time. For me, it's all about creating efficiency and getting something out of it, and for most people that's just going to be time. Save time.

Trent McLaren: When I come back around, I'm going to ask you what things have created the most efficiency for you. If you really want to tease them, you could tell them you're going to move them to Gemini instead and get them sad about that. Beau, I'll throw to you, mate. Do you want to talk us through the last thing you guys implemented? What was that AI thing? It could be Claude, it could be something else. Sorry, I'm prefacing my question because I feel like I know what you're going to say. You guys have traditionally always been very fast with your change, right? Most of your team are tech-savvy. I've been to your office, you love hanging out together, you do big budget nights, pizzas, drinks, all those things quite often. Would you say that culture is just embedded from day one, when people come into the firm?

Beau Gaudron: Yeah, I would say so. Connecting back to what Travis was saying, I think for a lot of us we're naturally ready for this sort of change, because we're already doing things like identifying the different problems, writing them down, and keeping a log of: this was a problem, or this is something we could improve. We've done that for years and years, even pre-AI. So the big change for us was that we already had a lot of that data, and we'd naturally worked on a lot of them. But when AI became something we could use, we were like, oh wow, a lot of these things have been on the list for a long time. Instead of going, well, we'll wait on that one because it's going to be too hard, or we can't do that at this point in time, we've reshuffled it and rethought it, because we can deliver a lot more of it. So now it's more a process of looking at what's on the plate and what's going to make the most difference.

Beau Gaudron: For us, probably one of the most recent changes was that we set up a GPT using OpenAI's API with a ZDR agreement in place, so there's no training, nothing like that going on, to assist in a BAS process. It could use a bunch of Xero data to give that first-pass review to our accountants, to say: here are the things that look like they'll need correction or action inside of Xero. So rather than having to get in and manually look at those things, or go, okay, this is something I don't need to worry about, they already had a note pre-done by AI. And when we talk about having the team involved, the thing that's a little bit unique about me is I'm not an accountant. So when I'm building these things, I have to work very closely with our accountants in the office.

Beau Gaudron: I say, okay, I've done this, and they pull it apart, because they're like, what about this? You haven't done that. This part needs to be done a little bit differently. So I've learned a huge amount about the way they expect the outputs to come out for them to be usable.

Trent McLaren: Yeah, awesome, okay. My follow-up question for you is going to be: given you have that culture, what are the biggest barriers or challenges you're trying to tackle? I'll come back to you with that in a sec. Gus, do you want to elaborate further? You mentioned you started as an AI training consultancy and you've built this AI SaaS. When you're getting firms using the tooling, do you have an approach, or have you seen anything that's working really well for firms trying to get that adoption in place?

Gus McLennan: Well, the biggest case study I always cite is a firm called Blue Rock that we're associated with. We launched the platform, adoption was kind of okay, but it started to plateau, and the CEO came to me and said, what's going on? At the time we were still running old models and AI wasn't that great. We're talking probably back in 24, 25. What we did is organise these training sessions. Half an hour, in person, and we basically started doing practical stuff: how do you prompt, which back then people didn't even know what that was, how do you check the AI, how do you look for citations, all this other stuff. And what we found is that usage in the firm doubled within the month. We were monitoring all the stats.

Gus McLennan: And then it grew by 50% again after that. I don't want to labour the whole change management point, to your directive, Trent, but getting in front of people and constantly reminding them how to use the tool is absolutely critical, because a lot of this is just muscle memory. The more you do it, the more you start to understand how to prompt more efficiently, how to ask questions, how to do exploratory work. So yeah, that would be my suggestion, my recommendation.

Trent McLaren: I'm going to come back to that. There's a piece here on how you're managing and monitoring the adoption. You mentioned usage really took off, so I'm keen to understand what that looks like. On the other side of that, token usage is less of a thing because we're all getting cheap subscriptions that cover a lot of stuff. Any recommendations or thoughts on how people manage and monitor that as part of team adoption? In the chat I can see there are security questions, which we're going to come to, so keep throwing them in there. But the second part is, if we do get the team to adopt it, does that mean my bill's going to jump to $5,000 or $10,000 a month, or whatever it is? I'd love to come back to that when I circle back around.

Trent McLaren: James, I feel like you're in an interesting spot, because you guys have been in business, what, seven, eight years? Started as SaaS?

James Rose: Technically since 2010. Why?

Trent McLaren: Wow, that's longer than I was thinking.

James Rose: Content Snare since 2017.

Trent McLaren: Yeah, nice. You guys started as traditional SaaS, but now you've been able to rebuild and wrap AI around it as well. How has it been trying to help firms adjust, saying, hey, you've been using us like a SaaS product, but now you can start using us like an AI product?

James Rose: The interesting thing is it's very much the same curve as adoption of our product to start with. All the early adopters were really keen to sign up at first, and now we're in the late majority and it's a harder slog to get people to sign up. This has kind of restarted with AI, where all the adopters are like, bam, they're straight in, it's smooth, they work it out. I think for a lot of people they just haven't actually tried it. And you realise the only way we can convince people to do it is just to show them. You click here and here and here, and now you're connected to Content Snare. Now you type in, do this thing, and it's done, built the client request. And people are like, what? Getting people to adopt anything is like that. When they can't picture it or don't want to try it themselves, you just have to let them do it.

James Rose: And I've spoken to people where the way they get their team to adopt things is actually to build some of their own stuff on the side for fun, personal stuff. One of my friends ran a competition across his entire company to build a game with AI. He has a cleaning service scheduling tool, and they all built a cleaning-related game, then they put them on the website, and it made a big fun thing out of it. So I think you need to get people's click moment. Sometimes they do it themselves, sometimes it's sharing how it works, sometimes it's getting them to play around with it for something that's completely unrelated.

Trent McLaren: And the creative part there is the thing I think a lot of people have struggled with. I don't know about the accountants in the chat or on the speaker side, but one of the things I come across most is accountants saying, Trent, I'm really good at the numbers, the business, whatever. I don't think I'm a creative person, though. I can creatively solve a tax problem, I've been doing that for a long time, but how do I take something and be creative in my approach, especially if it's not something I've been comfortable with? I loved your example of building games, and one of the lessons in there is you've got to force everyone to sit down for an hour or two and say, hey, for two hours you've all got access to Claude right now. You can be as dumb as you want, do whatever you want, just to see how it works.

Trent McLaren: And I normally hate that advice, because I don't care if you can do a shopping list or write songs or whatever it may be. But when you give people a non-risky situation, something where there's no consequence if it gets it wrong, it lets people free up a little bit on what is possible. Travis, you mentioned your BAS process before. There are some things in there that are a bit more risky, that you can't get wrong. How did you build your confidence that you could start doing higher-risk things with your AI?

Travis Bacon: Yeah, so one of my guardrails is essentially de-identification. When we put information into AI, it's all around removing client identification from it. So think TFNs, bank logins, IDs, all that sort of stuff, but go the next level and start thinking about client names, addresses and that type of thing. If you get a set of financial statements and you can strip all that information out, that data is now safe to go into AI and start to be analysed. Everyone's going to have their own risk profile on that, so you need to figure that risk profile out yourself. My process was to separate the AI from the actual tool. We built the tool using Python and Playwright. For the AI component we used the Claude API, so we're only sending data off to the API endpoint to use Claude once we're comfortable the data's been de-identified.

Travis Bacon: So just really having that guardrail and that really strong consciousness of going, okay, how do we de-identify this? How do we get this to work and still get the same result, but take that bit of risk out of it? That's the approach we've gone down. I can see there are so many questions about security and risk in the chat, and I think a lot of it, from my perspective as a tax agent, is just having those guardrails set up and sticking to them.

Trent McLaren: Vicki had a question there about how time-consuming it is to strip out the information. You understand why you're doing it, but have you found ways, or are you using agents to strip all that out, to make it less time-consuming?

Travis Bacon: Yeah, so build the tool that does it for you. You can start to look for identifiers like Pty Ltd, so you know any of the information before that is potentially a client thing. If you build a tool that identifies it, and all you have to do is approve, because it gives you five or six different identifications, you've sped up that process. So build the tool outside of the AI to get to that result.

James Rose: I feel like a lot of people get confused on this. I've had this conversation so many times, because people say, oh, but then it's going into AI anyway. We really need to clarify that you can get AI to build your own little piece of software that's completely independent from AI. It's not using AI to do the redaction. You've just got to be really clear on that, because people say, I don't get it. No, you get it to build a tool that does the redaction, and then you run that tool before you put it into the AI.

Travis Bacon: Yeah, creating that separation. It's almost like church and state, keeping them completely separate. That's how we've accomplished it.

Trent McLaren: Love that, awesome. I want to come back to that, absolutely, because it keeps coming up in the chat. The biggest question coming up in the chat is around security. How do we keep our teams doing things securely? How do we make sure our things are secure? There are questions about whether I should connect Xero data, even though there are approved Xero MCPs and all the marketing around it. Look, I could care less about most things, so I throw everything into Claude. I'm on paid plans, I'm there. A lot of people are not like that. They're on the very other side. And I want to help people build a bit more confidence here, because a lot of people are holding back because they're so scared they're going to put something in. How are you guys getting around that, or what's your attitude and approach to the data and security side of things?

Beau Gaudron: Probably the best thing is I'd reword what you said, get around. We don't get around it. We respect the actual TPB guidance on this.

Beau Gaudron: So part of it is the data redaction all of us have spoken about, and James touched on this. I think there's a lot of confusion around exactly what James said: if you're going to redact data, there are ways to do that which sit outside an AI agent. Travis touched on this as well. There are a lot of things you can build with Claude, or with Grok, or any AI agent, that actually sit outside of it and don't use it. And on top of that, you can do limiting things. Travis mentioned Playwright, which is a browser automation tool. If you've got browser automation tools doing specific functions in Xero, for example, you might have a Xero login user that only has access to specific things. So a lot of it gets approached from a cybersecurity perspective: what are the things you're allowed to give up, and what are the things that are non-negotiable that you're not allowed to do?

Beau Gaudron: For us, we went an extra step with some of the data we're using. Yes, we have internal policies that say don't do these particular things in these tools, because we know they train on data, for example, or they can't really be trusted. But with the BAS process I talked about before, we actually reached out to OpenAI and asked, can we have a zero data retention agreement? When that's in place, it means that when the data's being processed, they will not train and will not do anything with the data other than give you a response. So that's another big part of it. Anthropic do have these as well. But I think it's also important to note, and I know you just said it, Trent, oh, I've got a paid account: a paid account does not mean they do not train. That's a very big thing to be aware of.

Beau Gaudron: For our team, and I know you ragged on Gemini earlier, we say, if you're in Google Sheets and you want to know how to do a formula for something specific, use Gemini. Try and encourage people to go, there's a tool available, start using it. The other big thing is, if you want your team to be using AI tools, make sure you have them. Like Travis said, they've got Claude, it's available, they know Claude's the thing they can go to. Tell your team: these are the tools and the allowable exercises within them. I think that encourages use naturally. They're not afraid of it then. They're like, oh, okay, cool, I can go and ask that. And this ties in to encouraging use: we have a Slack channel set up, and every day it posts and says, tell us your AI wins and what didn't work, which is equally important. Then Steph and I review that at the end of the day and go, okay, is this a training exercise

Beau Gaudron: where we need to talk to our team and say, what you were trying to do, you can achieve it, but you've approached it wrong? Or is it something where we've got to go, maybe that's something we build, rather than the team building it?

Trent McLaren: That's a good definition, right? You want teams to have some authority to play and explore, but there's also a line of, hold on, I'll build that and give you the space for it. And I guess that comes down to having good policies: this is what we think as a team you should be doing. We talk about setting guardrails within your AI, but you've almost got to set guardrails within your team of what's possible. Gus, you must come across this all the time. You have so much financial data in your system, you're accessing ledger data and everything else. How do you communicate that and keep everyone feeling comfortable that you're being secure with the data you're hosting?

Gus McLennan: Yeah, well, we've got a technical guide that we issue, and that goes out to lots of prospective customers as well. We let them know where their data's being stored, which is here in Australia. We explain how we interact with the APIs from our various AI providers, and that there's a no-persist policy on any of the data that gets shared. The one thing I'd say to everyone on the call, which I say to anyone using AI, is make sure you standardise and use one common platform for all your firm data, the approved platform. The number of times I've sat in meetings and had people go, oh, I didn't like that, so I've just been using ChatGPT instead. And you say to them, have you got the data sharing settings set correctly? And they just look at you with a blank stare, and you're just not sure what's happening.

Gus McLennan: So use the one approved platform for all of your data. For us, we do the redaction just for TFNs. And I saw some of this in the chat: having to de-identify documents just takes time, and sometimes it's impractical when it's a whole bunch of PDFs and stuff like that. So you use our platform so you don't have to do some of that stuff.

James Rose: You're muted.

Trent McLaren: I was really in the zone with that. I was basically saying that despite the enthusiasm of everyone in this chat today, not all of you are going to know what to build or want to build, and it's actually just easier to find the right provider you can buy that's going to take all of those things into consideration. Gus, for everyone that's maybe not seen what you guys do before, do you want to give a high level of what AI Dojo does, so they understand what kind of data you're housing and the jobs you're trying to help solve?

Gus McLennan: So we are a secure AI platform. We grant you access to models from all the major providers, including some of the open-weight models that come out of the Chinese labs. We enable you to do general-purpose stuff and tax research, all the way up to skill-based workflows to prepare BAS, individual tax returns, all that sort of stuff.

Trent McLaren: Yep. Awesome, okay, great, thank you. Jimmy, you guys have had to deal with security for a long time, and you came in before with your comment on redacting. In the theme of everything we're talking about, do you have any follow-ups or recommendations for the audience on how to work through these items?

James Rose: I've been trying to blast answers in the chat here, because people don't understand that with an app, you can just say, Claude, can you publish this as a web page locally? You can run a web page locally that looks and feels like an app. That's how those games I was talking about worked. So that's how you get started. You say, hey, Claude, I want you to build a redaction tool that I can call, or I can use, that doesn't use any AI at all. I want you to build the whole thing. Here are some sample documents with some fake TFNs and names and stuff in them, so you can see these are the bits I want to redact. I explicitly say, because there's a lot of dodgy redaction out there where it just puts a black box over the top of something: I want the data fully removed from the document. You can put a black box over the top if you want, to show it was redacted.

James Rose: I have a script that does this, by the way. It's actually just for TFNs and stuff, and IP addresses. I just asked Claude to do it, and then I fed the results into another chat and said, can you confirm this data is not anywhere in the document? That's the process I go through. You could try and read the document, but PDFs are encrypted and stuff, so it's a bit hard. And the next step from that is I actually built a command line tool. Now we're getting very technical. Instead of having to go to a browser and put folders in it, a command line tool is something Claude can just call locally on your computer. So I have files that sit outside my Claude workspace, but it can still call the tool that has access to those files, which can redact them and pull them into the workspace Claude has access to. So at no point did Claude read the files with the sensitive info in them,

James Rose: but now it's got access, because the little script pulled them into the workspace. I know we're getting super bloody technical here, but if someone's got the recording of this and feeds it into Claude and says, can you build that thing James just said, please? Like, literally.

Trent McLaren: I'm actually trying to prototype it now while we're all chatting, just to see if I can give you all a copy. There's a good question here assuming that ChatGPT and Copilot can do this. The one thing I would say is I actually don't know that Copilot can do this. It's not a coding tool. Does anyone want to tell me I'm wrong?

James Rose: No, it's more of a chat engine, like ChatGPT.

Trent McLaren: Yeah. Whenever we're talking about building and coding things, the tools we'd probably all recommend, and I gave stick to Gemini before as a joke because I didn't want to say Copilot, since I know everyone probably uses Copilot, are Claude, and ChatGPT has a Codex product. You could also use Lovable. If you were on the speed dating session, Nathan from Ahoy Ahoy talked about Lovable. I call Lovable the beginner's coding tool. I started on that, and it was actually a great way to get started. And then you've also got tools like Replit and Base44, which are probably in that beginner's coding tool category.

James Rose: Is there a Google one? It's got a funny name.

Trent McLaren: Antigravity just came out.

James Rose: Oh.

Trent McLaren: I think that's the one. They've got a few names for stuff. Travis, I'm conscious of time here. We've talked a lot about security and the risks. Tell me about the wins, mate. Tell me about the things that are absolutely humming for you. You said before your team would be in tears if you took it away or moved them to Gemini. Tell me about the wins. How much time saved and efficiency are we talking about?

Travis Bacon: Yeah, you know before, when you were talking about getting Claude to do something and you walk away and grab a coffee? My wife would probably argue that I don't actually ever work, and that Claude does most of my stuff for me. But by building on that BAS prep tool, we went from a BAS prep tool to an accounts and tax prep tool to an individual tax workpaper prep tool. And that's like, think assistant accountant job gone. That's kind of where it's gotten to. So the time saved for staff in getting a job to the point where they can actually start working on it, we're talking a couple of hours a job for the bigger ones. The other thing I'd say is that James and Beau and I have probably gone a little bit far for some of the people just starting out, talking about coding and everything. It's all really good, great stuff. What I'd probably say is, Claude in itself,

Travis Bacon: just the way it operates with Excel, Outlook, the web browser and the actual desktop program, the efficiencies you can get just from a co-working perspective between all of them talking to each other, that on its own is going to save you time. So even if you're at the very early stage of getting started, just start using the program as best you can and start looking at the interactions it can have. That's a time saver in itself. Stop using it like Google, stop just asking it questions, and start giving it things to do. That's a time saver.

Trent McLaren: Do you have one specific job or example you've been able to save some time on?

Travis Bacon: So basically, the lodgements and OFS process. We built a tool that takes all of the annual accounts, the tax returns and everything, and builds out an OFS bundle, I guess you could say, and a summary for the client. It's a click of a button. You don't have to sit there doing it. Get it to look at all of the past stuff you've done to build the document library, and it just spits it out. That's a powerful one, if someone was looking to start building a tool.

Trent McLaren: Yeah, awesome. Beau, any quick wins or tools you guys have built that have created the most efficiency for you?

Beau Gaudron: I've been reading the chat as well, and I know we've probably gone to the more nerdy level for a lot of people. What I would say is, don't be afraid of connecting something like the Xero MCP connector to Claude with your own Xero organisation, or an Ignition MCP server, or a Karbon MCP server. Do those little things and then query your own data, so you can just get used to, okay, this is what I can ask it. Start small. Start by asking a basic question, like, can you identify something within this system? See if it can do that, and then build on it from there. A lot of what we've built over time has been going, can it do this? Yep, cool, it can do that. What's the next thing it needs to do? And I've seen some of the questions about where you save time. This comes into more of a coding tool as well.

Beau Gaudron: One of the things we've built off the back of the Xero API is something that can do payroll-style reconciliation. Think of doing something like an IAS at this time of the month. It can pre-do all of the checks to the point where it can tell us, these are the numbers that should be lodged, and it's a one-click button to go into Xero and just verify it. Before, it was quite a process to export all of that data, make sure it was all right and have it verified. Now we've done it all in code, and that all lives outside of an AI agent. Yes, we used AI to generate the code initially, but now that's a win that's available long-term and ongoing. And that's another reason I'm saying start small, start with the MCP server connectors, start with the things you can query quite quickly.

Beau Gaudron: You hear all of us talk about this stuff and say it's really useful, it's really great, and I'm sure Travis can talk about this as well, but they aren't overnight exercises for some of the bigger things we're working on. I just wanted to make that clear. I think people see some of the things being built and go, oh, that'd be great, but don't realise there's a lot of logic, a lot of testing and a lot of things that go into it to get to that point. Sure, there are wins, but the wins normally come out the other side. So just be aware that you can end up spending a lot of time doing the checking along the way.

Trent McLaren: My high-level lesson for everyone is that someone in your firm is going to have to learn this stuff and get good at it if you want any real direction with it. For those that don't want to, you're going to need to find a trusted partner, a trusted platform that's doing all that thinking and expertise and delivery for you. Gus, I actually think that's where I'd put you guys. You're trying to position yourselves as that place for people that don't want to spend all that time. Would you agree that's a good role for you to take on?

Gus McLennan: Yeah, and I think everyone's been talking about coding and stuff like that, but the entry level for a lot of accountants is starting to just build some skills. For those that aren't familiar with the concept, a skill is basically a package of files that's an instruction set telling the AI how to perform a particular task. I ran a webinar a few weeks ago where, on our platform, with our custom-built Xero connector, you could create a skill that gets the AI to connect to a particular Xero org, pull through all the GST transactions for a particular month and go through a whole classification assessment. I think we had 14 different rules it had to check. And then it spat out a dashboard at the end. So you didn't need to be in Claude Code or anything like that to do that. You can build these skills just through chat.

Gus McLennan: I don't know how it works for Copilot, to be honest, but definitely if you're in Claude or ChatGPT, you can build some of these skills. You never see a line of code, but you can build out these packages, upload them and start to use them. That's really what we started doing. But a lot of people just aren't even there yet, so we've started building them and publishing them and making them available for our customers on the platform. A lot of them have taken them, downloaded our skills and made their own version of their BAS prep skill. And yeah, that's awesome as well.

Trent McLaren: Yeah, awesome. Gus, there's a question for you in the chat, I'll let you take that in a sec. James, Jimmy, as we start to wrap this session, do you have any further feedback or recommendations? Again, there are two types of people in accounting right now. There are those that want to build, which is the small percentage, Travis and Beau and the handful of others that are into it. And then there's everyone else who's trying to figure it out: I'm used to clicking buttons in the UI and doing the thing and meeting with clients, and I've got manila folders all through my office. That's their world, that's their environment. Any final tips to help with this crazy wave?

James Rose: I'm just going to reflect what I said in the chat: just ask Claude. I've found myself saying this so many times now. If you're curious whether it's possible to automate something, just ask it. And not in the chat, again, not in the chat. This has to be in Codex or Claude Code. Say, this is our process, this is what we do, here are some sample documents, whatever, this is what I want you to do, can you come up with a way to do it? Oh, that's not right, maybe you need to do this. Literally, just ask Claude. Honestly, at this point I think the biggest roadblock is having the ideas on what to automate. A lot of the time you can just work through it. And it's knowing how all this fits together. You'll build knowledge over time, and as your Claude is connected to more things, more things become possible. Yeah, that's it.

Trent McLaren: Awesome. I'm going to start to wrap the session, so if anyone's got any final questions, throw them in the chat. I was going to say, if we've gone a little bit too hard and nerdy, I apologise, but then, Cassandra, you wrote that really nice thing saying it's nice to see the nerdy stuff, because we all need to know what scale we're all on, and I do agree with that. Some of these sessions are going to be so far ahead of what you're thinking about, and that's okay, because you need to know what's possible, and the only way you're going to know is if you hear a few people nerd out on this stuff. My ask, and you do what you want, but if you want to take something away to make the time you've invested today worth it, just figure out what's most relevant for you. What's the biggest problem? What's the biggest opportunity? If you've not started, the one thing I said is that at least one person is going to need to understand how this stuff works for your firm. That's the context.

Trent McLaren: You don't need someone at Travis or Beau's level, or our level, to know this stuff inside and out, but someone needs to understand what it looks like in your firm. And dare I say, if this was a few years ago, we would have said the same thing: someone had to be in charge of your cloud accounting strategy and digital strategy and everything else. Gosh, damn it, someone's going to have to be in charge of your AI strategy and implementation as well. I'm saying damn it because that was me avoiding change management topics, and I just can't avoid it. Stan, you had a question on Grok bot. Anyone been using Grok bot? Show of hands. Beau? Like it?

Beau Gaudron: I've literally just started using it because it popped up in my account. Because it's using screen reading and stuff like that, it won't be using any client data, but I said, hey, log in to my LinkedIn and go and have a look around and give me some statistics based on stuff I've posted there. It's been interesting just to watch what it's come back with, and I do think longer term it'll be an interesting tool. I was also going to say, when you were talking about having a champion: if you are that person, and for me I'm generally the person trying to encourage the use,

Beau Gaudron: be conscious of the way you're interacting with your team. When my team comes to me going, oh, I want to do this, don't approach it with a level of fear, which so many people do: oh, you can't put that in there, you can't do that. Say to them, this is what you can do. I had an example recently where one of our team was using Gemini and had come up with a Python script, and he was like, what do I do? It's not doing the thing, because it had generated all this code. And I said, I'll take it and help you learn how to run it, rather than just going, oh, I don't know what that is. Make sure you still encourage that exploration phase, because otherwise they're not going to feel comfortable getting involved in using the technology at all.

Trent McLaren: Good points. Anyone have any other final points before we close the conversation?

Travis Bacon: Probably just two. I want to reiterate what James said: just ask Claude. Seriously, if you've got something on your mind, go in there, even if you use the voice transcript. Just click on the little voice button and talk to it. You'll be absolutely amazed at the results you get back. The final thing is TPB requirements, all that sort of jazz. My go-to phrase is human-led, AI-assisted. Don't let the AI lead you. Make sure you're reviewing everything. You are responsible for everything that goes out the door. So, human-led, AI-assisted.

Trent McLaren: Awesome, thank you. Ladies and gents, it's been amazing. Thank you for joining us, I hope you've enjoyed this session.