509 Why AI Is Making Spreadsheets Cool Again!

I am more excited about accounting AI than I have been in probably over a year. But specifically accounting AI that can now automate a whole bunch of our workflows in spreadsheets, which is almost certainly uncool to say in 2025. But the reality is AI has gotten so good at hustling spreadsheets that there are, man, there are so many examples of things that I've done in the last decade, like inside of firms where we spent tons of time cranking in spreadsheets and AI can now do this stuff nearly flawlessly. I'm going to give you some actual real examples of this today. A tax example, accounting example, even a review example, because it has me swinging back towards like, man, there's some aspects of my workflow right now that I could simply build out in a spreadsheet and then use AI agents to manage in that spreadsheet. And this has all become possible from like new AI that we've gotten in the last month. We're going to run through that today. This one may make your head spin a bit. It has been kind of tripping me out how exciting it is. But let's get into it. If you're new here, welcome to the Jason on Firms podcast where we talk every week about all things running accounting firms, basically. I know, really cool, right? I think so. But AI spreadsheets, you want to talk about cool, let's talk spreadsheets. So here's the thing. ChatGPT's agent mode, which people are still sleeping on, you know, on the podcast lately I've been all hopped up on this, but I continue to find unbelievably cool things that agent mode in ChatGPT can do in our spreadsheets. So for context, Agent mode, it's available in any paid Chat GPT plan. But if you're using chatgpt inside of your accounting firm, you should only be using the team plan. Comes with really good security. Don't have to worry about prompts being trained into the model and all that. If you haven't taken the dive yet, I hopefully by the end of this episode, I've got you convinced. Because the marriage of ChatGPT's agent mode with spreadsheet stuff is wild. Basically what Agent mode is, is inside of ChatGPT. It has this built in web browser and it can like visually browse web pages and grab information and all of that. But interestingly, Agent mode is also the most powerful spreadsheet AI that we've ever had. Not only it creating spreadsheets from scratch but what it's actually even better at is modifying existing spreadsheets that you provide it with. And frankly that has absolutely nothing to do with browsing the web. The fact that it has a built in web browser, but for whatever reason right now in ChatGPT agent mode and the way that it thinks and is able to handle tasks is very, very good at managing spreadsheets. To be super clear, this is not the same as giving ChatGPT a spreadsheet and say do X, Y, Z. To enable Agent mode, you have to click the little plus and toggle on agent mode. Now for most ChatGPT conversations, I'm not using Agent mode. Agent mode is only really necessary when it needs to be able to see and browse and do stuff on the web, like actually in the web interface, like I want to log it into a system and have it do stuff in that system. That's the idea behind Agent mode. But for whatever reason and this, they brought this up in the announcement and there are like spreadsheet benchmarks, believe it or not, that benchmark, like how good humans are versus how good AIs are at doing spreadsheet stuff and all of that agent modus is off the charts at doing spreadsheet stuff. And this has kind of gone under the radar because agent mode is very good at other things. This is an unbelievable story from Yakov Jacob. I'm sorry if I just, if I blew that pronunciation. Check this out. A story posted on LinkedIn. AI just prepared this tax return for context. He's a tax pro. AI just prepared this tax return. It went to the TurboTax website and logged in. It answered all the personal questions and made assumptions correctly. It input all the income data. It refused to be upsold multiple times. It reviewed the tax return, it finalized the tax return and then Yakov went in and downloaded it from the TurboTax website. So how did it know what to put in there? He provided it with his TurboTax login and his basic details. W2 gave it a prompt, very high level on like what he wanted it to do. But that was it. It ran off, took 14 minutes. And the output, the end result was the exact same as what he got when he ran it through his professional tax software. This is. There's so many levels of irony here. A big one being, boy, I sure wish that like the leading tax softwares in the US right now were cloud based so that browser based AI agents could help us with our tax software and actually do meaningful things in there. The fact that the first Tax software we have that we're capable of automating with AI is TurboTax. What, like Dimension, what timeline are we on right now? Because it feels like a dark one. But he goes now, he goes on to say, people are always telling me that I can't prepare tax returns yet. Maybe in five years it'll be able to. And then we need to figure out how it'll, it'll affect jobs. He says, we aren't talking about five years away, we're talking about two years away. I would argue myself, you know, Intuit Pro Connect is probably the only cloud based tax software in the US that works for some firms. And I would argue we've already reached that threshold where it's like there is an AI good enough to actually do meaningful things when it comes to preparing tax returns. And we're going to talk, we're talk accounting example, tax example, and a review example. So don't worry, I won't, I won't blather on about US tax stuff for too long. But Agent Mode has been making headlines and justifiably so, because it's doing some really cool stuff and we need to be exploring that and thinking about like, boy, how much can I trust this thing to go out and work on its own inside a system where it could potentially also mess stuff up? And this is what keeps bringing me back friends to the humble spreadsheet. Because when I started in this profession, one of the things I did a ton of like the first five years was we had these contracts with the state to do like trust returns for. I think it was folks who would like pass away and they couldn't find any next of kin or family members or anything like that. And so we had to do this, this basic trust accounting until everything was closed out. And to do this, this basic trust accounting we had this Excel spreadsheet. And I think everybody has some version of this that is like a very basic accounting ledger. And the ways, the way that ours worked is on the left hand side you got the date, then you got the bank statement description, you've got the amount and then you've got a running balance. So the amount updates the balance as you're going down. You can like foot the balance to whatever the statement says at the end. But then after that you have a whole bunch of columns per categories. And this is, I mean this could be advertising expense, office expense, more trust, specific stuff. All the columns after that are just basically the accounts. You can code stuff too. But then before all those account columns, you you have a category column, and we would put a letter in that category column. So if advertising was the very first account, we would have an A above that column. And so if I put an A on that row on that transaction, the amount would get pulled over to the advertising column. Welcome to 2025, where we explain Excel spreadsheets and audio podcasts. Don't worry, that was uncomfortable for me too. But I think we've all got some, some version of this where you, like, bodge together a basic accounting ledger in a spreadsheet. And we're putting together a main channel video around GPT5, the newest AI model, and sort of the impact on accounting firms and what sort of stuff we should be looking at. And agent mode came up because I can't stop talking about it. And I rebuilt this little accounting ledger template inside of a spreadsheet. We're kind of testing some different stuff, like trying to figure out what's some cool stuff we could show off in this video. And so I rebuilt this little, this little accounting ledger in a spreadsheet. I gave it to Agent mode, and then I gave it 25 bank transactions. And I said, chuck all these transactions into the spreadsheet and categorize them as best as you can. And I didn't have like a huge degree of faith. Like, I feel like there's a lot of things like this where we're like, oh, yeah, that'll probably work, right? And then it goes and does it and it's like, wow, you found a. A way. I could have never expected to screw it up. To be clear, not unlike working with other humans, right? They will always find ways to surprise you and go in a direction that you hadn't even thought about. But I gave it those 25 transactions and the spreadsheet. And I said, here's a spreadsheet template that I use to do some basic accounting. Chuck all these transactions in there and code them to, to the best of your ability. And I, I didn't. I figured it'd be working for a while on that because it's got to go through 25 transactions in less than two minutes. It came back and said, here you go. And I'm like, yeah, sure you did. I download the. I download the spreadsheet and hit it. It had entered everything in. It had understood the spreadsheet without me explaining anything. I didn't. It didn't break any formulas. It only put the stuff into the cells that it should actually be touching. It didn't break the formula cells. How many colleagues have you had break your glorious formulas before? It even filled the formulas down because there weren't enough of them and they had to have more rows to put the transactions in. My only complaint was some of the categorizations like, and like. There were accounts like dues and subscriptions, office expense, software expense. There was a lot of gray areas there. Even, even advertising, like, you know, know HubSpot. You're going to put that in software, you can put that in advertising. I didn't agree with some of its classifications, so I did the same thing again. I said, come back to me with any transactions you're unsure how to categorize. And it did the next time, about 70% of them. And for the other ones, it came back to me like good questions, like Amazon, PayPal, some of the fringe stuff. And it was, honestly, it was as good as I think a human would do that. But that's only 25 transactions, right? I was shocked that it did that in less than two minutes. And so we kept testing how, how deep does this rabbit hole go? And we stopped at 500 transactions. Guess how long 500 transactions took? Less than five minutes. Now it came back to me with a bigger list of things for me to clarify, but generally for each of each of them were merchants. And there were like, you know, eight transactions for that merchant. So I provide that clarification once and never asks again. And it makes all those coding updates in the spreadsheet just like that. And it nailed it. It did not break the spreadsheet. It went through every single one of those in less than five minutes. It was not that long ago, what was, I mean, 10 or 15 years ago, where I was entering that stuff in off of bank statements. And honestly, there's somebody doing that right now as they're listening to this. But today, not only is Agent mode good enough to, to enter all that stuff, 500 transactions into a spreadsheet and kind of be your assistant and be like, hey, can you clarify X, Y and Z? These are kind of the fringe ones. And it's got a really good sense of what are the fringe ones I should ask about. But you can also give it a pile of bank statements. And we've been able to do this for a while. Like the models have been good enough. You give it a pile of bank statements and it, it extracts the transactions and it all foots and ties out to the bank statement. And there's a, there's an old quote from, from Paul Graham, software guy. You've already seen the future. You just Dismissed it as a toy and so that you could look at this and be like, oh, wow, cool. Like, that's very junior work. Entering transactions into a spreadsheet, encoding them, like, who cares? The reality is we still employ people to do that stuff, right? You still have these dumb little workflow headaches that you wish didn't exist, but they do. And agent mode, man, like, if I go back to thinking, when I was a junior, if I had these tools, would that have saved me time? For sure. I mean, how long would it have taken to key in, you know, 500 transactions off of a bank statement? I probably would have been working that for 90 minutes, two hours. Are there ways to get extracted information from the bank statement? There are services where you pay per page. Is that a problem that should exist in the first place? No, probably not. Why don't we just have the transaction download? Well, somebody didn't ask for it. It is an elegant solution to a problem that shouldn't exist in the first place. But it makes you think, okay, I mean, it's not hard to have like a very basic accounting ledger for a schedule C or a Schedule E. I mean, there's so examples of this, something that's going to have less than a couple hundred transactions a year. And I will just do that. Today's episode is sponsored in part by Maker's Hub. You know, the fastest way into the heart of any company that has inventory or complex purchasing stuff. It is the bill pay process that was our foot in the door to doing all sorts of advisory for these companies and cash planning and stuff that actually paid premium pretty well. And this is now a juicier opportunity than ever before thanks to how AI is changing accounts payable and Makers Hub. Man, they are like right at the forefront of this if you haven't seen them yet. We've done a number of demos on our channel of just how far automated AP has come. In fact, I've talked with firms now in the wild who have successfully automated like big chunks of the accounts payable department and various clients, businesses. But the real value of getting into this work to me isn't like the arbitrage opportunity that might be temporary around, like just showing them a more automated way to do it. Rather, it is having the completeness of all that information, the bill pay info, all the invoices. It unlocks some really cool advisory opportunities. So my firm, I did a lot of bill pay. Not because I liked bill pay, goodness knows I did not. But because of the other types of more valuable work that that actually unlocks the advisory opportunities that we didn't otherwise have access to because we didn't have that really valuable information. If you want to learn more about just how automatable this stuff is now, man, check out the link in the show notes chat with the folks at Maker's Hub. Today's episode is sponsored in part by Stanford Tax. Do you have a plan for next tax season yet, Buddy? It is high time to be using the tech now that you're going to be using next tax season. You don't want to ride into battle in January and February on stuff that you that you're not already comfortable with. And if you're just feeling like overwhelmed at not knowing what tax intake solution to use, buddy, look no further than Stanford Tax. One of the earliest ones in the game, one of still the most widely loved tax workflow solutions. And they're doing some cool new stuff. Stanford Tax now it's not just a way to like push a portal to your client for them to submit information, though you could certainly do that. Growing number of firms are just using it internally because you compile whatever source docs you got in there and it will automatically like check off the items from the Pro Forma checklist to like, see, like, do I have all my info in now or not? Like, it does that automatically based on what it sees on the docs. It builds out this like nice indexed sort of workpaper file, auto renames all the files, organizes them like almost like a, like a PDF, like a bookmark PDF, but in the cloud. And you can now even like do annotations in that. So you literally just take your big old pile of source docs from the client, chuck it in Stanford Tax. It does all the organizing and then you've got a dynamic checklist that will tell you do I still need more information from them or not. A massive upgrade to traditional methods for gathering 1040 info, which is just kind of a mess. I can put Stanford in front of my clients and they can submit the docs, or I can just use it internally. And any old way that docs come in, they just get piled into a single centralized place. Very, very good platform. Check out Stanford Tax. More info down in the show notes. Now some other examples. Another example we've talked about a few times here is taking tax source documents, taking the data off of those source documents and entering it into an Excel workbook template. And the one that we've always used in in testing is the glorious Excel 1040. One of man's greatest modern achievements. It's at excel1040.com Every year a guy builds out 60 tabs of government forms input pages where you can key in a whole load of different tax info from Social Security to brokerage statements to your W2s. And all, all the government forms, not all the government forms, but a huge number of them. And then reproduces over 30 like tax forms, like 1040, page one. It's unbelievably well done and detailed, but this giant Excel workbook, and I've talked to a bunch of people now who are like, dang, we had our own Excel workbook that we used to double check some stuff. We're now just going to use this and modify it for our needs. But you can take this Excel workbook, give it to agent mode and ChatGPT, and then give it a pile of tax documents. And I've tested it on like up to 10 tax documents and it's done it flawlessly. And that was 200 to 300 data points that it pulled off of the tax document and put into the 60 tab spreadsheet exactly where it needed to be. And if we think about review, how could we use this to. There's loads of ways that this can help us, right? But if we think about quality assurance and reviewing our work, a spreadsheet like this that reproduces what the correct output ought to be, you know, the page one of the 1040, when we have this sort of double check built into the spreadsheet, it makes the work of checking the work of the AI much better too. Right? And that's one thing we're, we're kind of stuck on right now, especially around research, is I can chuck a gnarly research problem at AI and it comes back with a convincing answer. But now I'm like, where do I begin to even prove that out? Did it miss any details like how do I actually unpack this? And that's where you got to think about like, if I'm going to have a human on my team do this, how will they prove to me what the right answer is? And that's, that looks pretty much the same as how to get the most useful AI research is not just give me an answer, but give me the fastest and shortest path to being able to prove out the answer. So best practices there are like, you know, pull out exact quotes from the source material and all that, because when I would ask a junior to do that work, they, they'd have a bunch of supporting work papers and highlight the relevant parts and all that. So in the end. That's really the same thing that we want the AI to do. But this, this 1040 workbook, it's a great example of how. It's an Excel template that has the double check built in, so it does all of the work, but then it's very quick for you to confirm against your tax software. Once that's all keyed in, do those two things match? Another example of where I think this can be useful for review. In the US you gotta prepare 1099s every January. 1099s report payments made to contractors. So W2s report payroll to employees. But for contractors, generally payments over 600 bucks in most cases, or total, total payments in a year for services over 600 bucks, you got to report it on a 1099. It's a big old pain in the neck. You got to get like the Social Security number or ein from the recipient. They don't understand how this stuff works. In some scenarios you don't have to issue one, like if they're a corporation, super annoying. But the form itself and the process of completing it is very basic, very administrat. Usually you've got the 1099s and then like this 1096 report on top. That is like the sum of all of the 1099s that make it up. It's like some summary information. And in our firm we had software that would do this 1099 prep for us, but we had an Excel template from the software vendor where we'd actually put all the data and then we would import that into the software and the software created the forms. We print those to PDF, e, file them wherever they need to go. But the actual work of like building those 1099 and completing them, it was happening in an Excel spreadsheet. And that was not detailed work, it was annoying work where it was like, ah, the person said this or that. Does this mean that we ought to be reporting to them personally, or does it mean we should be reporting to the business? There's some nuance like that. But the actual completing of the forms, super annoying, super trivial. And you can see where this is going. That Excel template, buddy, these days you, you chuck that into ChatGPT agent mode along with all the, the supporting documentation, be it emails with information or whatever the sources are for that. Basically whatever your supporting work papers would be normally for doing the work, it pulls all that stuff into the Excel spreadsheet. And then if you've got a, a review process for these 1099s, we would basically, we had Like a peer review process for this. So one person would prep, somebody else would review. Didn't have to be a senior person, really. They're just ticking off that we didn't bungle any of the entries. Well, man, AI can do that too. Whether that's like a separate conversation in ChatGPT or frankly even like a custom GPT that you can share with the rest of your team. Because this, I don't know that this isn't reviewing the work. I don't think would be modifying an Excel file or anything like that. So you give it the completed stuff and you give it the source docs. You say make sure that this is correct. Maybe you give it specific steps for how to, how to review it. But that's something that it is very capable of doing right now as well. I mean, if not better in many ways than a human, because we're so bad at, like, there's this big, you know, nine number string of, you know, an ein or something like that. And we see it here and we see it there. Are they the same for whatever reason? That's a hard thing to do, especially when you're doing a large volume of them. But if your second set of eyes, your double check is to chuck that into a custom GPT where you've, you tested that like you, you make mistakes on purpose and throw it at it and say, does this identify those mistakes? Maybe you give it some more like, literature on like, who should the recipient be? Should it be the business, should it be the person? What are the considerations there? So that when you give it the source documents and there's, you know, notes of those discussions and all that, even if a junior person makes an error there, where there's an issue in like, the facts of how it's being presented, it could even help call that stuff out. This stuff is like ready, like now. Like, it is good enough for that. In fact, this, this 1099 season ought to be a tremendous help. I think the thing that, well, the thing that most firms are still blocked on is security considerations. And man, we got, we got to do whatever we can to work through that. We talk about this a lot on the podcast, but the only thing different with AI apps than any other app is are your prompts being trained into the model. And if they aren't, it's no different than any other app that you use. So ask yourself if we would put these 1099 into my practice management system, what makes that fundamentally more secure than ChatGPT, which comes with SOC 2 type 2 security, multifactor authentication. It is a vastly bigger and more sophisticated company than anybody in the accounting vertical. We got to get real first principles about like, okay, what is the objection here? Because some of it is just AI spookiness. The fact that you'll still see news stories every single day saying don't put your medical information in AI. And that's, that's true. But what they're talking about is free consumer AIs where all the prompts are trained into the model. But we're not using that in our business. We are using business AI that is secure. ChatGPT team being an example of that. Microsoft has their own version of this in their business plans. Google has their own version of this in their business plans. Still, to me, nothing is as complete and robust, especially when it comes to agent mode, as ChatGPT right now. That's the thing that most firms really need to get over. But man, the beauty of the spreadsheet and the fact that there's so many useful things we can build in spreadsheet form to then have AI work on and complete, is that it? It makes, you know, you know, problem with AI right now is you give it a task and the format it comes back with is not something that works nicely in my workflow. You know, I think it's something that we're kind of struggling with. Like, it is a helpful thing, but it is not integrated into my workflow at all. So I then have to take the output and like create a workpaper for it or report or something like that. That's not to say that's not better than nothing, but the even better version is it's like the output work paper is exactly what we use as a work paper. And so if we can have an Excel template that's good enough to serve as the work paper for this thing and you can lock all the cells that you don't want it to mess with. That's one thing that I did in the little sort of quasi accounting ledger thing is I locked down any cells I didn't want it to mess with. The first couple times I didn't have it locked down and it didn't break anything. But I kind of wanted to see, like, will it still work fine if I lock down the stuff that it shouldn't be touching and kind of, you may as well, right, if you don't want it to touch, that stuff still works totally fine. So you can build out this workbook template, lock down the stuff you don't want it to touch, and then really anything, you know, that you can build workflow wise or for supporting documentation into that workbook. AI I think just got tremendously useful at building that for you. And some of the early examples we've, we've heard of success stories of people using Agent mode is like, you know, that client's got that giant profit calculation spreadsheet or some sort of big hairy allocation. Oftentimes there's multiple partners in the business so they want, they want a third party to put together this profit sharing calculation or something like that. And it's just this dumb thing they send over once a month and takes a couple hours to go through. You check it to a junior, you review the output. It's this sort of thing that I've seen people saying they're having success with that Agent mode is every, every bit as good, if not better than a junior at that sort of work. But the higher volume, more turnkey, repeatable version of that is the supporting documentation that lives inside of a workbook. The workbook is locked down so that it can't mess with the stuff you don't want it to mess with. But then you give it all the inputs, the context, the supporting documents, whatever it is that needs to go into that workbook and it does it for you. And that's really cool. But honestly that's probably still just step one. Like after that you can think about like what's a review step look like? Does that require Agent mode or not? Maybe it doesn't because all it's doing is looking at what's there, looking at all the context. You give it a bunch of information of all the ways that this stuff commonly goes sideways and what to look out for and then it does that review for itself. Man, that's really cool. Today's episode is brought to you in part by Carbon, the practice management system. Couple big announcements from Carbon over the summer if you've missed them. One is a totally brand new client facing portal, like a beefed up client experience called Carbon for Clients. You can push tasks to it documents, E signatures, all that stuff. For clients with multiple like entities and they're 1040, it's easy for them to quick switch between those accounts because I mean any business owner client we have, that's a problem, right? They've got their business, they've got the 1040, they've got eight other LLCs. Carbon for clients makes that easy to access. But then another big announcement was a first of its kind, super deep integration with Stanford Tax, another tool we really like on this podcast because you got these really cool tax workflow tools but then none of them talk to your practice management system. Right? Carbon wants to change that. Carbon's like an ecosystem first player. If you want the PM that's going to integrate with best of breed, take a look at carbon. PMs right now are kind of like separating into we're going to build our own versus we're going to integrate. Carbon's going hard down the integration path. The whole Stanford taxing is a good example of that. Got some videos on the YouTube channel about carbon if you haven't seen it in the wild before. I know folks down to like solo firms clear up to like some of these PE roll up platforms are using carbon one of the most mature options in the game in a, in a good way I would say. More information about Carbon down the show notes. Today's episode is brought to you in part by Mercury. Mercury is the FinTech that over 200,000 companies now trust for banking and running their finances. I'm a happy user myself. Long before they sponsor this podcast My money it's been with Mercury because it's just like man, traditional banks, they don't really seem to be going anywhere or getting any better. And you know all too well that struggle in fact and it's dumb to even say out loud, but the greatest upside of your clients using better banking solutions might just be access. Not having to do two factor authentication Olympics, not having to have clients like forward you codes, you know the codes they get over text, the text that say never share this code with anyone. And then the process that we communicate to our clients is hey, when you get this, send this to me. The just the janky things we got to do to get access to things. Because traditional banks, they don't let you handle like micro user stuff. Did I say micro user, multi user stuff. Granular permissions for the different, different people in your team. And even in the scenario where a client does give you your own login, then your entire team is like sharing your login buddy. There's like so many problems with that and so much risk in that where if you're just a little more opinionated and we're like hey, hey, just so you know, all this fuss and headache around like getting access to this thing, there's a better way that we recommend come over here and check out Mercury. Sometimes we're like apologetic about what a pain it is to do this banking stuff. Stop apologizing, say client, go use this better alternative. There's a solution to this and it is your fault that you're not on the right platform. I'd encourage you to give Mercury a look. It is, it's genuinely, really good. More information about Mercury down in the show notes. I don't think of it as a replacement for like, the software platforms that we use as much as a way to make all the other stuff a little more turnkey. So the way that we put together, you know, working papers to support whatever the work product is, that looks very different firm to firm. Firms kind of have their own ways of doing that, oftentimes reconciling very specific things and, you know, say a tax return or putting together a dashboard in Excel for a client. Oh, man. It's what's made Excel simultaneously very cool for a long time while also being very uncool is more. The engineer folks will be like, well, if. If it's something you're doing in Excel over and over again, it would be better built as software. And it's like, okay, cool. But we don't know how to build software. And the beauty of Excel is it's manipulatable. Like you can keep changing things, make small modifications to it over time. It's. It's almost like the original no code app, the spreadsheet, because there's a bit of programmability in there with your formulas and all that. But when you layer in an AI agent that can use that really, really well and go through a meeting transcript and pull a bunch of details out of that to complete it, like all sorts of different types of context where it can see into vastly, vastly more context and information than you and I can much faster and then pull all that together into a structured, a structured format that you provide it with. Man, that's pretty cool, right? I think that is so cool. And my head is still spinning at the different ways we can use that, from, you know, the sales process to onboarding clients to the reporting that goes out to clients. And is why I think I. It's kind of making the spreadsheet cool again because it's like, it's almost like the coolest upgrade that spreadsheets have gotten in years, right? If you add a program that you use, say your tax tax software or QuickBooks or something like that, that had an AI agent that was like, really good at using that program, that would be a tremendously, like, valuable upgrade to that program, right? Imagine an AI agent that could hop around in your tax software or hop around and do basic tasks for you in QuickBooks. We now have this in Excel and It's the best thing that's happened to Excel in a long time, I think. Pretty cool. It's like an uncool thing to say, but it is very cool, very exciting. I think that TurboTax story also just kind of blows my mind that it's that good right now. I mean, I rail on this quite a bit, but I'm obviously a big supporter of software companies inside of the accounting vertical because I think it's so important that we as accounting professionals have tools for us tailored for the jobs that we do. If we're reliant on just using all the software that normies can use and have access to, that's a bummer, man. Like for sure we'll still be better than them at being able to do X, Y and Z. But this is where purpose built tooling for what we do becomes so important because what we're talking about right now is like a feature that just got built into ChatGPT and will soon be available too. Like developers, as if you think about like the way that, you know, the very domain specific platforms can begin thinking about building AI agents within their platforms to do useful things for us, that's super exciting right now. One weird sort of asterisk that has always like kind of tripped me up when, when people talk about AI agents is if you look at the state of our ecosystem as an example and you say like this practice management system company, you should build an agent for your platform. Normally what they're going to say is like, well, we just need to build the platform to do whatever you need, not create like these pseudo users to do stuff in the platform. And you can kind of see that. But then you and I are like, oh no, actually what I really need is, is the agent that will work in my practice management system and my tax intake tool to, you know, sync the documents back and forth and the statuses or between my practice management system and my accounting software or you know, my month end closed system or something like that. The agent you get really excited about is the one that bridges multiple apps. In fact, I had this conversation with a executive at Sage like two years ago who gave a whole presentation on AI agents. And that was when people were kind of first talking about AI agents. And I was like, aren't the agents that people are most excited about the ones that are going to, that are going to work across all of our apps? Like why would we be super excited about the agent that just works inside of your platform versus the agent that can work across all of my platforms. And he's like, yeah, no, you're right. Doesn't mean that the platforms shouldn't themselves have agents. But yes, the most powerful agents are the ones that will work across the 12 apps that you use. But if you look at the ecosystem today, who builds that? What class of software is that? Because OpenAI could definitely build very powerful generalist stuff. But when somebody builds the AI agent that knows accounting stuff very well, knows how all the accounting tools work, you think about something very basic, like right tool, which is, you know, a Chrome extension that improves upon the QuickBooks interface, and it actually makes changes to the user interface and all that. Consider an AI agent that has capabilities like that and knows everything about how that app works along with the other accounting apps that you use. It's almost a new category of technology from a totally different vendor, because you could see how they're could potentially be like an adversarial relationship between the folks building the agents and the folks building the underlying platforms. In fact, there's even a risk to those core platforms that the agent stands between our relationship with that platform. Like, if I can just tell the agent, go do xyz, I will probably spend less time in that platform. If the agent completes this giant Excel workbook for me, I'm probably spending less time in Excel. And so who are the builders of these agents? It's almost an entirely new layer of technology. And so if we zoom all the way out to, you know, what the Bill Gates of the world and all these people have been saying for years, oh, software is just, it's going to be conversational. We're just going to tell it to do stuff. And that's always been hard to get our heads around, right? Because accounting is so much about data and seeing reports and all these things, but that layer of abstraction where the agent is then working across your different apps, you can kind of get your head around that, where it's not 100% of what you do, but oftentimes what you're doing is you. You're telling it to go do X, Y and Z across various different apps. And that is like a new layer of software between you and all the different sort of separated tools that you use. It is, again, an elegant solution to a problem that shouldn't exist in the first place. Because wouldn't it be nice if we had a single piece of software that would do all of these things? Well, but even if, even if you just look at practice management, you got some practice management systems that say they want to be all in One others that say they want to be ecosystem first, but none of them right now yet are doing, you know, tax intake. Well, automated tax delivery a month end close automation. They're not touching any of that stuff, let alone being the accounting ledger, the tax software, these are all entirely different jobs that a single platform is not going to do well in the foreseeable future. So then is the solution a layer of software between us and that that is the AI agent. And this isn't a new concept by any means. This is what people have been talking about for a while where it's like oh, we're just going to talk with AI now and it's going to do everything but agent mode. Doing cool stuff in spreadsheets I think makes this feel more real than it's ever felt before. And to to then swing this clear back to something hopefully hyper practical. Consider you know, next week or this week whenever you're listening to this. Is there an Excel template we have inside of our firm that does any job? I gotta think the, I mean the answer is yes for 99% of accounting firms. That Excel workbook could AI complete it and or could AI review it, start there with it in its current state but then think can we actually expand on this to do more things? Because if it could magically self complete itself then we might want to build that out to be something a little more robust. We might want to build out even another tab that does kind of some self checks so it can review itself as well. Super, super interesting to think about. I've been heads down in this this video about GPT5 almost all week and still the thing I'm most excited about by a mile it is agent mode and what it can do with spreadsheets. Super exciting times. If you're still stuck on the security scaries, go back. I mean even in the last few weeks on the podcast we've done deep dives on on all things AI security. The sooner you can work through that stuff the better because the opportunities right now are super, super exciting. Told you this was going to be a nerdy one. Any cool stuff you've tried, successes or failures, would love to see that stuff in the comments. Thanks for coming and hanging today and I'll catch you in the next one.