Can AI Agents Replace Your Bookkeeping Team? What Law Firm Bookkeepers Need to Know

One of our hot topics in our monthly AI meetings that we have at the Law Lab is people have been asking me about AI agents and can they stop paying for a global team to clear out the bank feeds? Can they just use an agent or AI to do this? I get it. Feeds don't sleep, so this can all be done while you're sleeping. Matches can pile up and partners want their books closed and they want their reports by the 5th or the 10th or the 15th of the month, and you want your evenings back. But here's my honest take for bookkeepers who serve law firms. Agents can't take the real work off of the pile. They just don't replace your good judgment, your human eyes. And they don't own the IOLTA 3-way for sure. The trust account is still going to need you. What agents are and what they're not. I don't mean the chatbot that answers, how do I categorize this? That's in QuickBooks right now. Tools that take a goal and use connected apps and work in a queue and pull a feed, suggest matches, draft a question, open a checklist when a deposit looks like a retainer. It's useful. It's easy to oversell it though. If your chart of accounts is a mess, the agent will confidently messy it up to scale. The agents will help you by clearing your bank feeds. They can help you by reconciliation. What I've seen so far is nothing short of remarkable how they can fix up a bank feed, fix a reconciliation that's got a beginning balance that's off. Mind-blowing, right? This is the stuff that's incredible. It can undo a bunch of reconciliations and redo them. That's wonderful. It can match payments. You can give it a bunch of payments and say, which one of these Clio payments go together? And it will figure it out with the total. Total and the dates. It can do predictable operating expenses. It does all that really well. One of the places that might cause it an issue is it might not be able to see duplicates. It's hard for even staff to see duplicates or payments that don't match open bills or deposits that were in the wrong account. It might not see that. That's where your time can shrink. And you need to measure it to see if it's really worth the value. You have to test it out yourself. Look at the cost, track your time training it, and then really let it do its thing and see how it does month over month. So far what I've seen with it, it's really quite incredible. I have seen it help me figure out how to use MyCase and where MyCase is failing as a program and how we can make it work better. It has that ability because it's got this knowledge base of how to use MyCase, or maybe it knows MyCase better than I do because it's a program I don't use all the time. It's great at drafting up an email written in my voice. It can read the emails that I've sent to specific clients and see that this guy, I talk in a different way than I do a very formal way with this other client. It can see that. It can draft the messages, but don't let the messages go out without you getting your eyes on them. Make sure it's saying what you want to convey. So another place that's good for is it can draft follow-ups, like when you need it to ask for specific statements or bank statements. It can create those emails and queue them up every month. You can tell it once a month, I want you to tell Ms. Jones that I need these bank statements because I don't have access to this particular bank account. That's where it's really cool. We've been using Carbon for that, but if you don't have Carbon, this is a great new feature that you can pay for the software, which is not cheap. So if you're just starting out your bookkeeping firm, you're using Asana, using 17hats for your contracts, and you don't have the budget for this, it can be pretty pricey, but if you've got the work, it's going to allow you to open up space. I've used it to help me with creating quotes or even reading what was in a meeting and explaining it, maybe what the client saw in the meeting, maybe where I failed to convey myself very well. It will call you out. Don't forget, AI likes to please you and we don't want to always be pleased. We want to know the truth. So you can ask it to, you don't have to be my friend here. I need you to tell me the honest truth that I convey the services that my company is selling correctly to the client. Is there any place where I dropped the ball? Is there any place that I could do it more efficiently? Is there any place that I could maybe be more clear? And you can use it to maybe clear up your contracts or your engagement letters with your clients as well. So where you really still need humans. And I don't think there's any exception here, is IOLTA and trust. There's clients, there's matters, the purpose. The 3-way bank reconciliation, although it can assist you here, you have to make sure it ties back to the bank. You have to make sure that it ties back to all the individual client ledgers. Agents can prepare the prepackets, but you should never let AI or anything like that, or any staff member, nobody should have the ability to move the money. Except for the attorney, right? So same rule applies here. You want to be really careful here. It might not see if an attorney puts money into the wrong bank account unless you show it. Any wrong recording or wrong coding can really cause an issue because it's not a place you can plug, so to speak. In fact, that was the terminology that my bot was using with me was I want to plug number here, plug. I don't want you plugging numbers anywhere. So you have to convey that as you're training it. You want to make sure that it's able to do the work, but it might not spot a wire fraud or any payment changes. It might not notice that a vendor maybe needs updating. Those are the things that people can see and we do, but you might not be able to see that exactly or clearly with AI. It's a great tool, great at clearing the bank feeds. It's really great at reconciliation. Those are the kind of basic tasks, but you just need to make sure that you're using it as an assistant, right? It is still senior humans are on the job to look at this stuff. You want to design it properly. A poor design is going to give you a lot of trouble. You want to make sure that you've got everything in place to be able to give your client a great product back. It is pretty good at going through the file and you can tell it to look at the file. You can tell Claude, can you look at this file overall? Give it a general ledger, give it maybe some paperwork where it can see the past and the present, and tell me where the client's file could use some cleanup, where maybe things aren't categorized properly. It does a really good job with that. It does a pretty good job with forecasting and looking at maybe where you have a client has a lot of software charges, having it aggregate large pieces of data. It's very good at that. It's interesting because AI is here. It's faster than even I thought it was going to be, and I knew it was going to be fast this time. But AI adoption rate was, at first it was about 9% of the firms. If you were one of those 9% in 2024, I know we were starting to adopt it, starting to use it. It went up to 41% in 2025. That's a huge leap. I think back to when the cloud came and how people were afraid of the cloud and how the adoption was a lot slower paced. This is like Mind-blowing how fast it's coming. And 35% of that 41% that started to use AI is actually using it daily. And it's in a lot of our products too, don't forget. And 77% of firms are planning to use or increase their AI investment because this has a cost value to it. You have to be able to look at the AI that's out there if you're purchasing it on the side, like ChatGPT, Claude, any of those. There's a cost to it and you actually want to buy in. You want to have that protection of the product, but you also want to be able to make sure your team is well instructed on how to use this. There's a learning curve to it. And the only way I think the team's going to really get to learn how to use it or how to implement it is to use it. I think you have to use it and you have to be using it now while everybody's learning and feeling their way through it. But you want to have some good processes in place with your team. On what is allowed and what's not allowed and explain why. We have a great product that we use called Wiser. It's W-I-Z-E-R. If you have a firm and you want to have a firm understand AI and understand where things can be hacked and where things can go wrong, it's little videos every month to keep our team on their toes. A lot of times I look at it as stuff that can even help me in my personal life when I start to see Some of the ways that the bad actors can get in. We have to protect ourselves from that. Thomson Reuters did another survey and asked a lot of firms, asked 1,816 professionals across 62 countries in March and April of 2026, and found that 74% of those firms were using AI tools multiple times a week. So we're using it. And 44% of them were using it Multiple times a day. So maybe you're that person. And I think that if you really start to look at the difference, you're going to be left behind in this industry if you're not using it. It does have a lot of value to it. But remember, as you start to take on these different types of software, whether it's Claude, whether it's Grokbot, whether it's ChatGPT, there's a funny comedian on Instagram. His name is Husk IRL, and he makes fun of where AI kind of fails. It's a cute thing to follow, but also when you start to look at these things, you might need all of them to work for you. So GrokBot is your teammate. ChatGPT can gather across large contexts of data. Claude Cowork can do multi-step knowledge work, build spreadsheets, build beautiful pieces of data, and it can do that with connected apps. So it's a really powerful tool. Again, expensive. Perplexity Computer is also— when the job is running long, it can do a lot of research. And there's just a lot out there. So you want to make sure you're using the right tool for the right product and that when you are using the tool, you're not wasting tokens so that you're using it where you're able to really be able to deepen the use case for it and use the right tool for the right thing. So yeah, it's a really complex world that we're in. And I think that making sure that working with the law firms that you keep it down to that low-risk work to start, give it simple tasks, let it just start to do it. And before you know it, you're going to have a built system that's able to do quite a lot of things. And maybe even some of those complex spreadsheets that we have to create or develop, maybe Claude Cowork can help you with that. But we really have to jump in headfirst as accounting firm leaders. And start to implement this, but implement it with a lot of guidance and care. So I hope that's helpful. This is a topic we touch upon every single month at the Law Lab. So if you're interested, please join us and help us build a great community that supports one another, that helps one another. We learn from each other's wins and mistakes, and it's an amazing community. So if you want to join that, there's a link below in the article. And I hope you all have a wonderful day. Bye now.