Your software stack is going agentic, whether you planned for it or not. In this episode of Unaccounted For, Wayne Findlay and Marvin Galang talk through what happens when the ledger starts acting on its own, and who in your firm answers for it.

Listen on Your Preferred Platform

Episode overview

The ledger isn't waiting for us anymore. Xero's JAX is already live inside the ledger, matching transactions, chasing documents and flagging anomalies. XeroForce is on the way, and practice management and workpaper tools are shipping agents of their own. Most of it arrives in subscriptions firms already pay for, so nobody ever decides to switch it on.

In this episode, Wayne Findlay (CEO, The Back Room) and Marvin Galang (COO, The Back Room) talk through what that means for accounting firms:

• Agentic in plain English: old AI answered questions, agentic AI does tasks. What changes on an ordinary Tuesday.
• JAX shipped, XeroForce is coming: the real question isn't whether AI enters your firm. It's who designs the system: your firm on purpose, or your software stack by default.
• What breaks first: workflows, review or trust, and why trust is the one to protect first.
• Switch off or let run: a simple rule for what can run on its own and what needs a partner's deliberate yes.
• Who's accountable: the firm is. "The AI did it" is the new "the junior did it", and every agent needs one name next to it.

One thing to do this week: open the AI and automation settings in your ledger and practice tools and write three columns: what's switched on, what each one does on its own, and whose name is next to it. If column three is blank, you've found the real gap.

Transcript:

[00:00] We've seen a big drop in the share price for for Xero and into it for instance will they be there in the future we're saying SAS is dead um I don't hundred percent believe in that are you gonna make sure it's secure are you going to pay for liability that gets hacked hey here's my receipts go ahead and record that here's the bank statements go ahead and reconcile that Singapore AJ rolled it out 274 and now it's about 3 80 or three 90 I thought it was just for coding apps and stuff but it's not I don't use Claude Chat hundred percent Claude code

[00:36] I cannot say it's hundred percent secure it's 99% secure you cannot fully trust the result of AI can you get AI to check AI I need to learn more about AI software and where it's going you're the guy I'm gonna interrogate because you know a lot more than me you're listening to unaccounted for well what is unaccounted for it's the podcast about everything the Ledger doesn't capture

[01:05] Gonna cover AI people and what's really going on inside accounting firms I'm Marvin Galang and I'm Wayne Findlay from the back room no slides no CPD's and nobody well maybe maybe we're pretending no we're not no pretending they've got it all figured out how are you oh

[01:27] Uh it's sunny out there we should be out there versus in here I know all good yeah no golf today no golf for you cause your back is sore my back is sore from the weekend playing golf so yeah well Marvin I've got you here today and I'm gonna talk to you I I actually I need to learn more about AI software and where it's going and yeah that's and I and some of the stuff that I don't know is probably stuff that yeah the people that listening to that are gonna wanna know and you're the guy I'm gonna interrogate

[02:04] Because you know a lot more than me I hope yeah but I'm trying to think of a good start point and what this episode is about but I think a good way to start is where the Ledger is going and also where accounting you know the accounting Ledger and how that relates to people where that is going

[02:28] And so maybe my first question to you is what are you seeing in the industry as far as a gentic type the Ledger being able to do its own thing and you know Paris here what what are you seeing in the world with with the Ledger and it being a gentic yeah that's a good question you know 2026 now um

[02:56] Maybe first half of 2025 uh you know I've been involved with another accounting software development we can name names can't we yeah we can we should um a shout out to AJ AJ how are you it's jazz Dot AI yeah um founded in Singapore

[03:17] But you know ever since we've been working on that and the capabilities I saw a this was last year I saw a future of accounting Ledger as being just Ledger okay the double entry debit credit uh repository of numbers and data

[03:38] And accounting software being no front end where you know people would be very much into chat GPT or Claude yup and they would work off chat GPT and Claude and they would want accounting to be working with Chat GPT and Claude but you have a Ledger in the back end so all you have in essence is just a Ledger that sits in the cloud yup and then you have your Claude or open AI will query it

[04:15] Exactly and and that's where you will have you know prepare me an invoice or uh reconcile this bank account it it it just does it it just does everything for you and it's like really talking to you know I have an assistant hey here's my receipts go ahead and record that here's the bank statements go ahead and reconcile that uh but it's it's it's it's a chat yeah and everything is working there and run me my report and then hey you'll you'll get your report uh but you know that was the initial vision but right now it's

[04:51] It has just gone it's gone beyond that but you know with Xero and quickbooks uh enabling their MCP's now what's an MCP that okay so you have your APIs where hey here's a software another software and you know there's a shake hand and they are talking to each other agents don't AI agents don't use that AI agents use an MCP so MCP is hey here's your AI agent here's the layer that interacts and it really translates what you talk about and then works off your account

[05:29] Your software doesn't have it accounting software CRM any software so that layer in between is an MCP now Xero has an MCP with Claude doesn't it is that correct that's correct but it's very limited and I've used it okay so the MCP is owned by the software company so in this case Xero or Quickbooks has the MCP so because they own it they dictate what an AI agent can do with their software

[06:02] With Xero um I think right now at least with Claude I know they have 11 from the last time I check what do you mean by 11 11 uh tools 11 skills 11 things that you can do or an agent can do with Xero that will go expand ah okay good good point Quickbooks had 11 yeah I think two months ago if I'm not mistaken but two months ago they also had 11 and then suddenly now they have about 74 so they increased it in Singapore

[06:38] AJ rolled it out 274 then now it's about three 80 or three 90 so you you know basically saying if you know hey AI do whatever you you wanna do a lot of arguments there you know people may not like it people may like it uh but your the your first question is where are we headed yeah that's where we're headed how far away is that do you think that's like it obviously it's here now in a limited version of it but what where do you see us in three years

[07:13] Four years time it'll be there um I think sooner you know um it's Mark very much market dependent right now yeah and the availability of tools uh if I'm running Xero and I'm trying to put my you know well what's the big question how do the big companies like Xero and Quickbooks

[07:36] How do they where do they go are they like a sales force in the future you know it's totally impacted by AI big drop we've seen a big drop in the share price for for Xero and into it for instance yeah do you think I think that's gonna be you know do we will they be there in the future they will be there at least for me yeah I don't see um why is that what's the reason you still need the core I think I I honestly believe I still need the core

[08:06] You know the core knowledge in accounting and finance is already there I don't think it will be it will lose its value we're saying SAS is dead um I don't hundred percent believe in that people can try to vibe code and develop for me the real question is after you've developed that what do you do next are you going to maintain it are you going to make sure it's secure are you going to pay for liability if that gets hacked we people don't think about that right now they don't think about that and that's a big thing now with AI and anthropic and in open AI yeah

[08:44] About them and their IPO and they were trying to put the safeguards in place and they basically get themselves out of liability but the real liability in developing the software is making sure there is liability in there so if there's another uh hugging face type incident they're accountable for that and that makes sure they develop the AI with you know I suppose uh with that sense in it so there's no um hacks or there's no disruption to people the value

[09:17] Although yeah they're struggling right now and a lot of other companies we're just not talking about accounting software um but eventually there will be a different source of value yeah people will find we'll eventually find the value there and that goes for everywhere even accounting firms you know hey um we're challenged right now we don't you know how do we shift do we need a lot more people do we don't need people uh transitionary I think um

[09:45] Yeah you know you we may not need a lot of resources right now you know a year two years from now I can see they're gonna they're gonna go back hiring they will they will right yeah it will be a circle uh uh circular so but let's go back cause you were talking about um you know are we seeing it now is it three years down the road um as I mentioned market specific where what I'm seeing cause if we go to let's say North America you know specifically United States even Canada

[10:19] Um man I mean Quickbooks and Xero has so much competition with accounting apps coming out in the last or came out in the last 3 to 4 years agentic first AI first you tell me I I hear this a lot and I'm not a I I know the difference but I know a lot of people don't know the difference between agentic AI and generative AI yep so what's it basic difference in layman terms between agentic and generative

[10:52] Um okay analogy as it relates to accounting uh generative AI is a um junior accountant and if you if you're related to the models they're able to generate you know financials uh generate a little bit of analytics but it you do you do a lot of prompting you give a lot of instructions yup uh

[11:17] In order for the system to do that that was the models in 2023 2024 is a super accountant so from junior to super accountant um really more than a CFO if you think about you know yeah you know what the knowledge that the knowledge and and the capabilities so you know that's what they're doing now so it kind of like it gave it you gave it a brain you know I need financials okay maybe you just don't need financials maybe I'm gonna compare this to what's your industry

[11:56] Uh who are your top customers who are your top suppliers what is the profile of your customers uh by the way who are the owners let me find out what they you know their characteristics what's the latest news about the industry and and you know and and and really we're feeding it with context but now it not only that it has brain

[12:19] It has you know hands and feet so a gentic is you give it a goal so if you gave it a goal it would go out and make that happen by using the tools itself that it could find whereas generative is a little bit more layman's it's stuck with how it goes through and does all that correct yeah but and the um whether scary or exciting part

[12:45] Is sometimes it's able to decide like hugging face hahaha

[12:52] Let's try to hack this uh but yeah it it it's able to make yeah you know levels of decisions and that's where guardrails are very very important well that's that's the thing you have to have all that in play yeah we've seen that sometimes when we're coding and you know it's like um I did not tell you to push it and I just want you to show it to me but it made the decision to push it because we're not very specific in terms of

[13:20] You know hey you only work up to here let me carry it from there yeah I can see that so with with Xero if we let's go back to Xero and and it's got AI called Jax yeah so is Jax a genetic or is that more generative type to me it's more generative right now you think it will become more agentic and it will help or do you think the plug ins to Claude will be the agentic side of it okay thank good thing you brought that up um

[13:51] Sometimes while you know in the accounting software that that we built in a couple years ago uh we also have an in app chat agent yeah you know um we and the reason we rolled out the MCP's with hey um 390 tools

[14:13] There's that philosophy that eventually you know people we don't want to lock people in uh the app right yeah they make their decisions like which which AI agent is best for me well Xero is a platform in theory isn't it correct so you should be able to overlay it with exactly different apps yeah exactly one of which is called so that could be the agentic arm yeah but Jackson itself is really generative yeah yes I wouldn't say it's limited because it is useful but and what it can do is is very um

[14:49] Guard rail doesn't it right it's to coding and yeah a little bit more but what do you see yeah I went to the conference in Denver and there was a lot of hoo ha yeah about Xero workforce there was Xero force they're calling it right so how do you see that coming in or playing out in the future

[15:14] Um that that's the beginning of the agenic phase I think you think from them in terms of uh from my understanding of I haven't tested it yeah Xero force somebody described it as the uh a Zapier inside Xero or an N a 10 inside Xero where you build your workflows it's an enabler isn't it enabler yeah right but again it's still in Xero I don't know the

[15:40] You know the technology behind it as far as the AI models um but I think the MCP is really the way to go if they're able to follow let's say you know what Quickbooks has done has slowly increasing it and and giving the freedom to accountants and users to choose your models because which model are you uh

[16:05] Are you using everyday wow I'm Claude I'm a Claude man see that's the thing your your your Claude I'd love to use Grock but then I'm sort of tossing up you know why would I move to grok I see this Grok bot and all this you know right now we're looking for an EA yeah and we're talking about our brain

[16:29] But yeah I did yeah what model I I I think they all chop and change and they all catch up with each other see that's the point next year like okay so 2025 uh self confessed open AI most people start with open AI AI chat GPT chat GPT yeah and then they move forward from there 2026 January yeah I moved to Claude yeah right now I'm still Claude Astra came out uh I resubscribed to open AI and then tested Astra um

[17:06] Tested Sol Sol uh so now I'm keeping both yeah but I'm still 70 80% um Claude so the point is what will be the model next year yeah we don't know so I cannot be stuck in an environment where hey use this because this is the latest and greatest but it's not you know I I want this freedom that I'm able to choose which model yeah and I'm able to equip my agent you know with the tools and then do my work yeah yeah which is interesting and actually just on that cause I know you had a question the other day

[17:44] Around guardrails and using um AI tools like Claude and open AI if you were to use that in an accounting firm maybe explain how you would do that to have that safety of that information in the the testing of data well talking about AI and I know a lot of accounting firms will wanna talk about this and go over it so if you wanna go and learn more go to the show notes or tap on the link and the link domain it's called unaccounted for podcast.com

[18:21] Alright let's let's go back to this podcast episode back to it you know we've spoken with some clients and um you know had a one on one sessions with you know a few others what is very interesting is um it not having the understanding what's the difference as far as security between chat and code so I mean I'm using Claude

[18:49] Uh are you using Claude Chat or Claude code um Claude chat okay uh and doing work papers and stuffs in Claude Chat um okay there's probably a you know a high security yeah risk on that one um like right now at least for me I don't use cloud chat hundred percent cloud code and the files are residing in my computer

[19:19] And I'm just utilizing an AI agent to work on those files hmm uh and you can instruct them hey I don't want any of these to go out on the web I want you to work here I want you to save the files here and you know just work off this uh I cannot say it's hundred percent secure it's 99% secure as long as you have those proper you know but it's private to you it's private to you you don't upload it in the web right yeah I think that's a big big uh

[19:48] I guess misconception I don't know if people realize I I know I've only recently used Claw Code and you know I thought it was just for coding apps and stuff but it's not it's it's bigger and better than that so you can use it a lot yeah a lot better absolutely see it's a good point the you mentioned code and the idea that hey I thought it's just for coding

[20:14] I think with agents being where AI agents being where they are at right now or AI models um it really changed the definition of coding as well yes when you talk vibe coding and you know maybe explain how that works vibe coding is like anyone can code everybody's out coding apps and doing stuff now and it's all vibe coding isn't it yeah and I you know where like I would redefine coding in a sense where hey if you understand workflow meaning I know the trigger I know the process in between and I know the ending

[20:52] I know the result that I want you understand rules you understand controls you understand security confidentiality the tools that you wanna use and then put them all together with the help of an AI model technically you're already a coder yeah you know given given the capabilities of the models right now uh which is sometimes unfortunate for

[21:17] For developers you know um remove all the other questions on security as far as uh insurance or liability but you're already coding you know so how do you see all this coming back to the accountant and how they work with clients

[21:36] With these tools you have now there's a lot more you could do with your clients oh man absolutely I mean even for us you know I'm seeing well unless you're a geek and you're really working on this and you wanna be a developer but even for us um yeah having the choosing the right AI filtering and maybe more technical but choosing the right I call it harness right the right harness and being able to you know to

[22:10] Plug in the right tools you know my calendar uh my chat yup you know everything it makes life a lot easier more efficient simpler more efficient yeah but it and now allows us to spend more time with people yeah instead of doing this report

[22:33] Okay the report's already with me so hey let's talk let's have coffee yeah we can do a podcast well that's we should we should down below at the coffee shop but that's where I think the future lies with the accountants is more in front time or may face to Facetime with the clients solving their problems

[22:55] Every time you sit down with someone come back they've always got problems clients and you can go out and solve that with the help of AI yeah um so I think that's something that in the past you've never had that time yeah I think you're gonna have that capacity where you can do that yeah

[23:18] It's exciting very very exciting I think the future problem would be you know hopefully you you know how to say no because now you're have all these ideas oh hey I can do that for me I'm gonna build this or I can sell that type of service and it's gonna be easy fast but uh stop at some point I'm seeing that sometimes right now like uh you know 10 15 projects at the same time dude you're gonna you're gonna be overrun with projects and this once you get into this world and what you can do there's so much you can do yeah yeah

[23:55] Trying to do it all at once is just gonna lock you down too hard so you need to be able to just work on a few projects that add value right work with your clients and what about the accountant how do you see them utilizing the tools and what they do um it's hard to envision a future where they don't have an AI agent working with them so they it's it's what we talked about in our webinars it's it's that 1 + 1 = 3 it's a having the ability to have an accountant with an agent

[24:32] Like we have our GERI but you you could have someone that helps you on a day to day basis absolutely absolutely it would bring out a different set of skills that's very very important you know well that I think that's probably another podcast for another day about what the future accountant looks like yeah but the the the key thing is people yeah that they're they're gonna be ingrained in that accounting they will be they will I don't think the fear of losing people is is zero

[25:04] That's a myth in fact the value of people the knowledge of the person of the individual really it has to shine right now and the key I think well that those why you need people comes back to trust a lot I think as well with AI how does that how do you see that working yeah absolutely

[25:27] Uh I already yeah it wants to see AI GERI a key element of trust is is people I think are those two tied together in the future yeah well that's a very important element I think with um you know with the use of the of AI technology right now uh I'll just mention one but you know this is a philosophy or a skill that I usually share with people that use AI is

[25:58] Is uh it's an old adage uh trust and verify trust but verify you know we trust but you have to put a way to verify stops and good accountants great accountants are very very good in verifying you know um good accountants when they look at numbers they immediately they're seeing something that's right or something's not wrong not right on the numbers

[26:25] That's the skill you know it's gonna be very very important and yeah I must admit when I wasn't the greatest reviewer of financial statements but I was not that detailed type person it will go down the checklist and you know that's why we've built GERI Review do all that for us but but I would get the set of financials on my desk and I would I could see if something wasn't right

[26:51] From looking at the financials I I could look through and instantly I could see something now that's not right you need to fix that um that sometimes is hard to get from AI yep you well you can you can prompt you can put you don't know that client like some things you know about them or you know put it in context yeah right so right yeah it's a even well context is that's a

[27:20] That's a totally separate that we can do but um even if you put all the all the context in place um you know all the guard rails and say hey you know I don't want you to do this or get me a report or tell me that you why you think you did the you did it right eventually you know at least for us and it's highly recommended somebody will sign off on that an individual will review sign off on this because 100% at this point

[27:51] You cannot fully yeah trust I trust the result of AI but I need to verify and make sure that uh yeah you know that I sign off on this because so far I haven't seen a liability where I'm gonna pay what AI has produced if it if it makes a mistake it's still humans

[28:14] Can you get this is as a aside can you get AI to check AI so a report's prepared by Claude can you give it to open AI to check and peer review it yeah there it's happening now in coding yeah you know some like Claude would develop um open AI can do the quality assurance or you can do vice versa so it's happening so yeah even on the accounting perspective still exactly even though they check each other yeah um you know you need to sign off

[28:49] People needs to sign off exciting ha ha ha ha ha it is I think maybe we start to wrap and I just wanted to ask you a couple of things um and I suppose for the listeners of this is some key takeaways that they could pull back from this session is maybe if we

[29:10] Agenic to generative don't be afraid of agenic AI yeah in the right context so using it with guardrails so I have those in play talked about claw code we know Xero's got that covered the stuff out there like jazz which is a gentic but it's not 100% there yet as far as um having I don't think you could use it in New Zealand or Australia just yet with with everything that's going on yeah

[29:44] And then having the ability to you know fear of the future I think it's more of an opportunity then it's going to be doom and gloom for accountants right big opportunity it's a for me it's I mean you know just to wrap up it's very exciting technology is at this point where it's headed um

[30:09] You know the the opportunity to increase revenue and at the same time really lower cost is is here uh as long as we know how to properly use it don't over company starts with the letter a big company uh spent a lot of tokens and they suddenly remove the subscription for some or actually reduce the subscription for some I'm not gonna mention the company

[30:36] Uh it's a big uh you know outs uh consulting company we don't you know you don't want that you don't want that to happen but it may so have the right strategy in place um our episodes about Ledger the future of Ledger yeah is already here um the vision last year where hey you know you may be just talking to an AI agent uh and AI agent working with your Ledger the right way the more secure way is already here how to do it you know what tools do I need what skills do I need

[31:12] We have a lot to talk about and maybe we can do that in next episode yeah I think uh you know for my final rap is you know that I'm excited about accounting with AI I really am I think you know I go back to the days of when Xero came on board in cloud accounting I think yeah we can this will be another episode you sold your CPA firm maybe we should start another one hey but we have because we're doing GERI outsourcing GERI outsourcing yep and I think that's the way you know of the future

[31:48] You know especially with smaller firms is having these tools that we can build ourselves and be more efficient then what it's been done in the past so that's part part of the reason why we're doing that so maybe that's a good good break off there we can have a coffee and we can talk about um that in the next episode so yeah absolutely yeah well thanks alright yeah off to the golf course now and if you wanna talk about AI in your firm

[32:20] It's all in the unaccounted for podcast.com again unaccounted for podcast.com and the links is down there in the show notes oh well that's it then so I'm Wayne I'm Marvin and that's been what the lecture doesn't capture

About Unaccounted For

Unaccounted For is a podcast from The Back Room Offshoring about AI and the future of accounting firms. Each episode, CEO and Founder Wayne Findlay and COO Marvin Galang talk through what is changing, what is hype, and what firms can start using now.

Want to talk about AI in your firm?

Book a call with The Back Room. Bring the question this episode left you with, and we'll talk through what AI and offshore teams could look like in your firm.

Follow the hosts

Wayne Findlay on LinkedIn: linkedin.com/in/waynenz

Marvin Galang on LinkedIn: linkedin.com/in/galangmr

More from The Back Room

All episodes: https://www.unaccountedforpodcast.com

Website: https://thebackroomop.com/

LinkedIn: https://www.linkedin.com/company/thebackroomop/about/