Making Accounting Agents Actually Useful

What is up, you guys? It's Yehuda back with another episode of Tax Tech Simplified: how to run the firm you want using tech that actually helps. I'm here to talk about a really interesting subject, something that has really come on the scene only recently, which is accounting agents. I'll say this is more tax adjacent because it's not tax work in of itself, but as we all know, a good tax return, especially business tax returns, starts with the financials.

And those financials come from the regular financial data we put into it on a regular basis. So I'm gonna be giving a bit of an overview of the current space as of August 2026. Obviously, this is something that is constantly changing every second because a new startup is coming up with another solution to something that we didn't even know that we needed. But here we are.

So let's jump into it. There really has been this explosion of accounting and tax tech the past year. I'd say we've probably added three to four new, completely new categories of tax tech just on the tax side. So that'd be like workflow solutions to create the work papers, which is one, to input the return, which is two.

And now we have this new thing that again is tax adjacent, which is accounting agents to do all that accounting work for you so that way you can do it quicker, do it faster, hopefully more accurate. The three, I'll say, AI tools that we've really been climbing up to. It started off with tax research, which was probably the easiest thing to create because all you're doing is saying, pointing an LLM at a specific group of data and saying, okay, only give me answers from that. Then after that came workflow automation, which is still a very iffy subject.

For anybody, anybody listening to this, actually I'm gonna be hosting an a round table for to discuss AI tax prep. If you've used any of the programs or you're interested in using any of the programs that have come out in the past year or two. I would love for you to attend. Go to Y10Taxtech.com slash roundtable to fill out your information and get an invite to that round table.

It's gonna be October 20th, so you have some time to sign up, but we'd love to have you there to discuss and to hear from anyone here who's already implemented these solutions. It really created it made us rethink how we have our how we how we input our how we create our workflow. Because now if something else is being done for us at different times, we need to think through that. So go to whitentaxtech.com slash round table to sign up for the October 20th round table, 12 p.m. to one p.m.

Looking forward to seeing you there. So we've had AI tax research, workflow automation, and now accounting agents. Accounting agents is probably the most interesting one because again, it's tax adjacent, but also there are so many different ways to go about it. Tax returns are prepared pretty much, I would say, not uniformly across the board, but the inputs and outputs are very similar.

People have very when it comes to accounting, people have very different versions of what a set of financials looks like, right? A set of financials all the way on the top of a public company, gap compliant, that's gonna be, you know, one version. Then, you know, a solo operator who's just running their business and just needs. Financiers so that they that way they can track their income and expenses, track their profit, and then also maybe file a tax return, maybe get a loan from a bank.

That's going be something completely different. So the this is, I'd say this proliferation of these accounting agents, which we'll define in a second, it's almost as if this, this is, it's almost like a big question mark of what it can do because there are so many different ways to do it. The nice thing about AI tools in general is that there is no specific regimented, rigid schema that you need to stick to. You can mix and match whatever flavors you want to create the ice cream Sunday of accounting that meets your needs or that client's specific need.

So the three questions I want to answer today are: what are accounting agents? What do they do and how can they help? So what are they, what do they do, and how can they help? So if you'll if to answer the question of what are they, it's not that complicated in theory, but it can get more and more complicated.

So it's just an AI tool that's built within an ecosystem, to connect to all your accounting tools. So the most simple example of this it is it looks exactly like ChatGPT looks. It just looks like a chat window that you can ask questions about your financials and it's connected to whatever source. So the most basic example is like intuit in ch intuit intelligence inside of QuickBooks Online.

It's a little window you pop open, you ask questions to, and then you can get answers about your financials. Something I did learn with more experimenting this past week is. It can actually post journal entries as well, which is really cool. It's really nice to be able to use natural language to like and even better if you're using voice to text something like Whisperflow, you use natural language to describe the journal entry you need.

And then it can just create that for you, which is so nice. I in my testing I did find specific to Intuit Intelligence is it seems to be not super user friendly if you start typing. And don't send the prompt and then you close out and come back to it. It like doesn't save it the same way Chat GPT does when you're in the middle of drafting a message.

So you have to like get used to using it, but it can actually post journal entries. So what they are just an AI agent that can connect to all the tools you're already using. So again, Intuit Intelligence is a great example of that. Then I'd say, you know, what they do, we already described a little bit.

So the work that they're performing. Is really anything you were doing before. The new, I'll say the new level that we've gotten to is now that spreadsheet that you've been manually tracking accruals with to create manual journal entries because maybe your software doesn't have accrual entry. Maybe you're not paying for that level of subscription in QuickBooks online, or maybe it's just more complicated than you than something you can say, split this evenly across 12 months.

Well, now You can have a product, another example of a product like this is Ramp Stack, which is if you haven't checked out Ramp Stack, I highly recommend, highly recommend you check it out. Ramp Stack is Ramp's AI accounting agent tool that can really do anything. And I'll talk about that in a second because anything sometimes isn't the greatest. So Ramp itself is a spend platform, and then they've been building connections and intelligence for many years at this point, many years, like what?

Five years, but for sure the whole AI age. And now they have this AI product that really it's its own standalone product. I highly recommend you go check it out. You can actually I'll try to link it in the show notes.

You can as a as an accountant, you can sign up to be a ramp advisor, get your own ramp account. And then ramp stack is actually free through August thirty first. So I'm so today I'm gonna be releasing this episode towards the end of August. So you still have time to sign up for Ramp Stack, connect an account.

Connect your personal account, connect your business account, connect a client account, and just run some tests through to see what it can do. They have not released their pricing structure yet. It's gonna be really interesting to see how they, they, they, they structure that because is it gonna be per client? Is it gonna be per usage?

Can you have shared usage across clients? At the end of the day, if I'm gonna have to pay a lot of money to do journal entries that I could do myself, will I want to pay for that? If Intuit intel intelligence can do it for me, because I'm already using QuickBooks Online, why would I use Rampstack? These are all questions that still need to be answered, but I think that's really only gonna, you know, be able to be answered once you actually test it out.

The testing that I've done so far is I'll give you a couple examples. Number one was recreating financials for a tax client. Their bookkeeping for the year was I'm not gonna even say horrible because it just wasn't it wasn't even real. It didn't really exist.

It consisted of bookkeeping that was just all over the place, was set up not by an accountant, and it was incorrect. So I had to go back from scratch and rebuild everything. And it did a great job at giving me a lot of stuff. And AI in general is really good at that.

It's really good at giving a lot of output. And the thing I learned from that is the more defined my input was, my prompt, and I wouldn't even say this is like prompt engineering, like, imagine you are an accountant. No, it already has that hat on of I'm an accountant, here's the work I'm gonna do. But rather the big question to be answered was the in my input was what is the exact output that I want to see that's gonna be useful?

Because the thing with AI is it can do a lot, but the problem is then you don't have ownership over what it does. So if you have if you are the one who describes exactly what the output should look like. So in my case, it was give me a table with these specific columns. And one of the columns should be which statement you pulled this transaction from.

And for me, that was super helpful because then I was able to actually see everything on one page and know where every line came from, as opposed to it just being a humongous spreadsheet and not sure where everything came from. I also included I had it categorize the transactions, but then also give me A column with confidence level and its rationalization as to why it either was not or was confident about how it categorized that transaction. That was one set of testing. Another set of testing is I've been doing some regular work for my clients that I do regular accounting work, and I have some complex journal entries where I have to take a lot of data.

One example was like output from Shopify, Shopify payouts. And I had this custom journal entry that I've just been copying every month and inputting the data myself. I was able to give k Rampstack four months worth of data in a CSV file without having to create the pivot table to split it up by month, without having to sum the totals, without having to do any of that. And it perfectly created the journal entries and then it will show you the journal entries and you can just click a button and say post to QuickBooks and then you can even it'll bring you right there.

And I double checked, it did it perfectly. Four months of journal entries that would have taken me a certainly a lot longer. And that's a great example of something that is just recurring that now I could do without having to think about it. The next question's gonna be about connections.

We're gonna talk about that in a second, but connections are really where these tools are gonna shine. And unfortunately, the lack of connections is gonna really show where the AI is lacking. So bank statements is the big one right now. You cannot get bank statements through most AI tools.

QuickBooks Online is decent depending on the bank at being able to pull the bank statement. That performance has every other month I see different levels of performance. Sometimes it works, sometimes it doesn't, and you never know what it's gonna be. So that's certainly interesting.

I think I hope that banks realize that this is something that's really important because. In 2026 and moving forward, the question becomes less how good a tool is, and more how much access does AI have to our tool, which makes something more and more valuable. So, so, so that's the basic like outline of what these tool tools do and how they work. A big question that I want to answer is why not just use Claude?

Why not just use Chat GPT? Well, I can tell you why not to use Chat GPT because you should be using Claude. At least in my opinion, right now in August of 2026. We'll see next month, depending on what happens, especially since a bunch of changes apparently are gonna happen with Claude, with AI usage, with code coding usage, with all the usages changing and gonna be costing more.

But we'll see, we'll see about that. Why not just use claw? So I would give a couple different reasons. Number one, simply speaking, security.

Because most companies right now. Across the US. Go anyone who's listening to this podcast, you're probably in a very small percentage pool of people that are actually on the forefront of AI adoption. Go speak to most people in corporate America.

They're if they're using Chat GPT, it's on their phone. And their security team, their security engineer hasn't even allowed it. They don't even have necessarily these tools. I someone I know I spoke to a couple months ago and said, my company is right only now rolling out Chat GPT or Claude and he wanted to know which one which one he should use.

So believe it or not, security is really important to be inside these tools because most companies have not implemented these tools because there's such a security risk. When having it inside of a standalone tool that does all that security engineering for you, again in theory, you still gotta check it out, there's a lot there's a lot more there's a much larger chance That you're actually gonna be able to use these tools, which means it's much more useful because if you can't use them, that's not useful at all. The second thing I would say is the context, right? And when I say context, I mean sort of like what I was talking about earlier is the hat that the AI is taking on.

You don't have to be there to prompt it and say, hey, you are an accountant or you're a forensic accountant, and here's how you should work. It already has all of that pre-built. So in Ramp Stack, It has a library full of skills, ready to go. It can manipulate spreadsheets, no problem.

It understands what it means to do a reconciliation. That already exists when it within the library. And that's not something you need to train or even or even update. Mind you, you can train Claude to do this.

And there are skills, public skills out there that you could download and utilize. Doesn't necessarily mean it's going to be as good. And who knows? Again, I'm a big fan.

I recommend testing out what you can because there's a lot out there. The third thing I would say is power. I think from a token spend perspective, my hypothesis is that these single-use tools, accounting agents that are built specific for accountants and accounting work, they're gonna be much more spend friendly, meaning they're not gonna like overuse tools when this when it's a simple when it's a simple task or overuse tokens. I'd like to think that.

Again, we still have to see it'll still need to be proven. The fourth thing is the environment that it's built inside of. So Intuit Intelligence is right inside of QuickBooks Online. It has direct access.

If you use Claude, its access to QuickBooks is usually dependent on the QuickBooks MCP, which right now, as of late, as of the last time I tested it, is not that great. It cannot pull all the data exactly how it should be able to pull. It for some reason can only was p was only pulling accrual data. It just was not connected as well.

So in something like Intuit Intelligence will have direct access as opposed to the QuickBooks MCP. Similarly, I found Ramp Stack to be able to get It was it was having no problem accessing QuickBooks Online. One funny thing, I did post about this on LinkedIn was that I was actually reconciling my own ramp card, like my own ramp credit card using ramp stack. And I was having trouble.

Usually that's a simple process. Cause anyone who's ever reconciled an account inside of QuickBooks Online knows that it usually figures out the numbers by itself and you just click done. But in this case, there were a bunch of transactions where It recorded the transaction on the transaction date, not on the post date. And the post date was really later.

There was like a hotel charge that didn't really land and clear until weeks later after the close of the statement. So it ended up, it ended up s ended up sort of missing a bunch of transactions. And then when I was reconciling, I was sort of was a little bit lost. So I had Rampstack go ahead and try to do a reconciliation.

And at first it was having trouble because believe it or not, It was only able to pull summary data from Ramp, which I thought was very, very funny because this is Ramp's product talking to itself almost, I would like to think. But obviously that wasn't the case. So I uploaded the statement from Ramp to Ramp Stack, and then it was able to reconcile it. Fantastic.

It said, Here are the ten transactions that need to be pushed off to next cycle. It found them exactly perfectly, really quickly, and then I was able to finish the reconciliation nice, nice and simple. And then the last thing I'd say why you wanna use you would wanna use a pre-built accounting agent is that it's pre-built. It's pre-built on day one, you're utilizing it, you're using it to its fullest ex extent.

You're not building it up over time. And here's the thing. Maybe I'll say maybe you'll say, well, I can train Claude over time and then I can train it to be better. You can do the same thing in Ramp Stack.

You can create skills the same way you create them in Claude, you can create them in Ramp Stack. So that's why I would say, you know, that's my case for using these specific solutions. Now that being said, sort of to lead into the next segment, how they how at the end of the day they're gonna be helping us. There are a couple limitations, but these aren't necessarily limit necessarily limitations.

For example, The way we utilize accounting agents is really only limited by our imagination, right? The way that we think about how to use it is by having a problem, knowing AI can solve that problem, and then implementing that solution within that AI tool. AI is a master translator. So you can really get it to create any sort of output based on any sort of input.

So Whatever recurring tasks you have, it can do, which means you don't need a perfect CSV file uploaded in order to perform different tasks. It knows what's going on, it has the background and it can do all of that. So what I would want to leave everybody with talking about this, and I honestly I want to hear your take. I'm really curious if you've tested out accounting agents.

Certainly, if you're watching this on YouTube, leave a comment below. If not, just Tag me on LinkedIn and let me know how things are going. But I want to hear how you're using this and what you think about how accounting agents are sort of gonna transform the industry. But I'm gonna leave you with a couple recommendations in how to utilize these things in a way that's gonna be actually useful.

Cause remember, AI can do a lot, but not everything is useful. So number one is experiment. You gotta experiment, experiment, experiment. You have to try different things.

You have to try it again. Sometimes the first time it doesn't work. And you have to figure out what went wrong. Don't be afraid to push back on the AI.

I was reconciling an account and it I knew that it pulled the wrong end of month it was for a stripe account and it pulled the wrong end of month amount and I figured out why because it was pulling based on like either UTC or Eastern time and the real end of bound month balance was a different amount and because it pulled the wrong amount. It compensated and said, we forgot a transaction on this date when really it didn't, and it was creating doubling up an entry. And had I just agreed with it and not double checked its work, it would have made my books, you know, be off by I mean, this case it was, I think it was 92 cents. But even still, us accountants know that's not okay.

So you have to experiment. Number two is ownership. The power is in the prompt. Like I said earlier.

Your vision is going to be what makes these tools useful because you need to have ownership over the output. Number three is that the connections right now are still manual. A way you can get around this is you can assign something like a bank statement collection to lower level staff, people who have less experience with accounting, and just say, hey, you know, you have a calendar every day. Check the calendar which bank statements need to be downloaded and put them in a Google Drive.

And then let's, for example, ramp stack can connect to Google Drive. And so we'll it will always have access to the latest bank statements when you need to do the month end close. And then above all, I would say keep on building pre-built plus self-built is what's gonna fill the fill the gaps. So you can use ramp stack, and then if you see that there's a gap that you can't fill with ramp stack, then go to Claude and fill it yourself.

You can you can you can you can Fill every gap right now. Now, whether you should or shouldn't is questionable, questionable and you need to use your judgment, which is why, you know, I hope, you know, you listen to this and try to get more ideas out of what you're trying to build. But keep on building. The reality is these tools are new, but the tasks are not new.

We've been doing this for years and years and years. Make sure that the tech you're using is serving you because at the end of the day, the goal is to get the work done, not just to create really cool workflows, even though it is fun creating really cool workflows. Make sure to invest the right amount of time. That means not too little and not too much.

If it's too much, that means that tool's not serving you properly. If it's too little, that means you never actually tried. And then I think with those efforts you can get really, really tremendous, results. So certainly tag me on LinkedIn, tag me on X.

I would love to hear how you're using accounting agents. If you're interested in help with your tech stack and figuring out how AI or your tech stack in general can sort of help you along in your process, you can go to y10taxtech.com. That's y10 t-axte h dot com.

And book a time to meet with me. I would love to talk. I would love to do a deep dive with you on your process to figure out how can we make your firm run better on the tech that you're already using so it can actually help you. And please check out the round table for AI tax prep coming up on October twentieth.

Check it out to register whiteen taxtech dot com slash roundtable. I'll be posting about it on LinkedIn as well. I would love for you to join. And until next time, keep on building and keep on doing amazing stuff with tech.

Have a great day.