486 First Impressions: ChatGPT AGENT For Accounting Firms

Today on the podcast, this is a big moment, gang. ChatGPT agent. It's a upgrade over ChatGPT operator, which was OpenAI's like very first navigate around a computer and do things for you in an automated way.

Agent is kind of the successor to Operator and it does some really cool.

Things that are particularly relevant to accounting work, believe it or not. But with that new tech comes some new interesting security considerations that they outlined. So today on the podcast, we're getting into like what it does the last generation doesn't do.

We're talking about whether this is something.

You can actually like put to work almost as like a quasi AI employee. Like, is that something we can do now? And we're talking about how this actually in some ways changes my thoughts on what the accounting tech ecosystem potentially looks like this year. Like, I hadn't thought of this as being a this year sort of thing, but some of that may be changing. So come on in, let's get into it.

Foreign.

Okay, if you're new here, welcome to Jason Up Firms. We talk about all things accounting firms normally from a much nicer looking studio than we're at today. But hey, I'm out with the people. In fact, today I'm in Chicago at an AI summit, believe it or not, but got a big release at the end of last week. It's called ChatGPT Agent. Not very creative on the naming, but it's now available basically to any paid ChatGPT plan or I should say to plus users and to team users. If you're in an accounting firm, you should absolutely be using the ChatGPT team plan. Comes with some data protections. Last thing we want is like our prompts being trained into the model when we're talking with ChatGPT about anything client related. So I, I send firms to ChatGPT team because I don't want your staff like not having access to ChatGPT and then doing like the end around and using the ChatGPT free plan without permission. Like that's just something we got to avoid. But in this team plan, you now have access to this agent that can autonomously go out and do tasks on its own on the web. And that's exciting, but it's also a little nerve wracking, right? Something that potentially has access to client information like going out and doing things on the web. The idea is that it can do it unsupervised. Is that wise? That's one of the things we're going to talk about. But first let me sort of just break down how it works for you at a very high level. ChatGPT Normally that is like a chat conversation, right? Or a speech conversation. You're talking with it. So you got this back and forth conversation stream maybe, what was it? Maybe six months ago. Now we got a tool from ChatGPT called Operator and actually lives at a completely different website. So if you're just using ChatGPT day to day, you wouldn't even bump into Operator. But honestly you weren't missing much because it wasn't all that useful yet. And historically Operator was only available on the $200 a month plan. But what Operator had was like this built in Internet browser kind of in the chat experience. So I could say is an example I used, I've used in some talks. I could say, you know, go out to the website for this conference, look at the agenda and find any AI talks that I may want to catch. And so it would spin up a little browser and you can see the browser in ChatGPT, it's been a little browser. It would find the website, it would find the agenda page and it would look through kind of all the different talks and pull a list of anything that was AI related. And this is pretty exciting, right? Like this is what folks generally refer to as agentic AI, something that can go out and you set it with a specific task and it will try to complete that task. I don't want to say by any means necessary, but it will just keep working away in the background until it completes it. Now sometimes it will get stuck, sometimes it will ask you for help. And like day to day business applications, Rowdy is most of the things that we use are behind like logins. So another feature of Operator was it could actually remember your logins. So if we're going to have it do stuff in QuickBooks Online or something like that. But biggest issue with Operator was it was slow, it got confused easily and wasn't really good enough to be that useful yet. But you could see where they were headed. Like it was an exciting signal for what was close. And now what we just got the end of last week, Agent, it is a really, really big step forward of basically the same thing. So they've combined kind of the best of what ChatGPT has become now, which is much, much more adept at like pulling information from the web, deep research, that's been a huge development for accounting firms this year. But they've pulled the best of all of that and the ability of ChatGPT to see into like a huge number of websites to answer A question and pullback sources and all that. They've merged that ability with the operator ability, that is the ability to visually see and navigate a website. Because sometimes it can work just fine and do what it needs to do without visually needing to see the website. But then other times, it does need to see it. And so ChatGPT agent now will, on the fly, figure out, do I need to be able to see and interact with the website to be able to do this task or not? Whereas Operator before, all it would do is load up every single webpage. And so if it's like trying to get some background information and trying to figure something out for you, it like, navigates to the website and opens up the whole website and it scrolls through the whole agent. Doesn't work that way. Agent takes the very best of what ChatGPT can do very quickly, just pulling information from websites without, like, visually having to navigate to the website. But then it can also do that loading up its own little computer environment with an Internet browser, but only when that's necessary. And so, to give you some very basic examples of what I already have tested with Agent, I gave it a few tax source documents, and then I gave it a spreadsheet workpaper template. And I said, go through these source docs, and this is like a very common thing, kind of the first step of putting together 1040. Go through these source documents, they're all government tax forms, and enter them into this spreadsheet workpaper template. What I used for the spreadsheet was Excel 1040. It's just this unbelievably huge spreadsheet workbook where they've rebuilt a bunch of the government forms and they have all these sort of input pages for W2s and brokerage statements and all that stuff. But this thing's like 60 some tabs, this spreadsheet. I gave ChatGPT agent the spreadsheet and I gave it three source stocks. Two of them were W2s and the other was a 1099B, like a brokerage statement. And I said, enter this stuff into my work paper template, the spreadsheet. And I was able to look at the spreadsheet, make sense of all the different tabs, find the right place for the W2s correctly Enter all of the data from all of the boxes from the W2s into the W2 tab, and then pull all the dividend information from the brokerage statement, put it on a dividends tab. It did all of that flawlessly. And that's not rocket surgery, but that's a way bigger volume of tasks than ChatGPT's ever been capable of performing in one swing before. Like this was probably 70 or 80 numbers that it entered into a spreadsheet. And one of the sneaky unlocks here with ChatGPT agent is it's vastly better at spreadsheet stuff than anything that we've ever had before. They actually showed this off in the demo. There's like some spreadsheet benchmarks that is like how good different ARI tools are at using spreadsheets. Cause because of course that's a thing, but this one is much, much better than any other tooling we've had before at just hustle in a spreadsheet. Like. Like a big step. Like, if I remember on the benchmarks, there was human level of goodness with a spreadsheet. There was previous state of the art, like the best anything was before. And Agent basically halved the difference. So took a humongous step towards average human ability. And this is maybe one of the things that I think people are getting wrong about Agent already is it's kind of like chatgpt where we jump to like, what's the biggest, most impressive, hardest thing that we know. And we're like, can it do that? No. Well then I'm safe for now when the opportunity here I think is to use it for the menial, easy stuff that we're having to do day to day. So that was one example, plugging tax source docs into an Excel spreadsheet. I think it's also very interesting because we have an entire category of tech tools that do this for us. What does it mean for them? We're going to circle back to that. But then the second thing I had to do was loop through all of the clients that I have access to in QuickBooks Online and see if any of the bank feeds are broken. Super common annoyance. You get in there when the month end close is due on the 12th, you log in on the 10th and you're like, let me just knock this out real quick. And then you find out that the bank fee's been broken since midway through last month. Now, I'm not a real accountant, so I no longer have my QBOA account that had like 300 and something clients under it. I now have a QBO QBOA account that has 11 accounts under it. But all I did was I told ChatGPT agent log into my QBO, or maybe I said QuickBooks online and loop through each of the clients that I have access to and check to see if any of their bank feed connections are broken. That was literally all that I told it, like two sentences. So it navigates out to QuickBooks Online, it asks me to log into QuickBooks Online. We're going to talk a little more about that once it's in there. It takes a, it takes like a bit of clicking around for it to figure out how to switch between clients. But eventually it did the entire thing itself and then it reported back to me in a table, all of the clients and the status of their various bank accounts because most of them have multiple accounts connected first try like it, like it, like, yeah. And so I, I don't know what that would do with a hundred clients. It probably wouldn't do the whole thing. It probably do some of them and some of them and then stop because I imagine on the back end there's only a certain amount of time that this thing is allowed to run so that OpenAI servers don't melt. But that actually works. And that's super, super impressive. But is it ready for primetime? Is it something we should actually be using?

Today's episode is sponsored in part by cloud accountant staffing who helps you find talented offshore and nearshore accountants to come to work for your firm. It is running by former accountants and a cool thing they just launched is what they're calling a candidate portal. That is you can hop in, see the candidates they found that could come and work with your firm. Filter by years of experience, filter by software experience, filter filter by region. Are they in the Philippines, Latin America, South Africa? That matters. Depending on the role. Is it important that the time zones align? Are they gonna be doing anything client facing? But they've now got kind of a self service portal where you can go in, see the folks, book interviews. Like in real time? No, like weird kind of recruiter, gatekeeping. Right. And like I mentioned cloud accountant staffing. The folks who put this company together.

And actually still run it to this.

Day, they're former accounting promoters. So they actually understand what you do and what it takes to like effectively integrate folks into your team.

If you want to learn more about.

Cloud accountant staffing, check out the link down in the show notes. Today's episode is brought to you in part by Mercury, the fintech platform that over 200,000 companies trust for banking and running their finance. Financials.

Finances.

Because traditional banking websites, banking apps, they.

What words are we allowed to use here? They're no good, they're very bad.

You hop into Mercury, you're like, oh, okay, this is like What a modern SaaS platform for managing this stuff in my business ought to look like stuff like being able to invite clients to set up their own accounts, manage multi user stuff, manage access. It's actually pretty fully featured like invoicing, bill pay card management and all the stuff you need for like moving money around, wires, ach, all that stuff. The before and after of working with a client with a modern platform like Mercury versus a traditional bank. I mean if you're doing bookkeeping, it's a tremendous time saver. Vastly simplifies the month end close. But oftentimes we don't feel like we're, we don't feel like we can be opinionated about what bank they use. It's actually a great example of when we talk about in say a three tier proposal letting someone just rock up with whatever bank they want to use that's going to cost them more money for me and at a certain point, honestly that like that may not even be an option. There is an easy solution for crusty old banks and the workflow issues that they create. I encourage you to check out Mercury like you will hop in there and be like okay, you know, I get it.

Okay.

I can tell you that with confidence because it's what I use. Honest to goodness htg you can even get it up to a thousand dollar referral bonus per eligible client. Okay, disclaimer time.

Mercury is a fintech company, not a bank.

Terms apply for bonus eligibility. See show notes for details. Learn more about Mercury.

Yeah, look down there. Check out the show notes. Sam Altman said something in in the demo. Here's an exact quote. He said I would explain this to my own family as cutting edge and experimental. A chance to try the future, but not something I'd yet used for high stakes uses or with a lot of personal information until we have a chance to study and improve it in the wild. They're basically saying this is pretty cool but it's still very, very early use kind of at your own peril. And this isn't because they don't understand, you know, how your prompts are affecting it or something like that. It's because if you give it your bank account number and you tell it to go and do something, they're saying maybe keep an eye on it for now. Maybe don't just cut it loose with that stuff because it does have access to the web and if it gets confused or puts it in the wrong place, there's something you should maybe pay attention to. And this kind of surprises me that they made this available on the team plan and the plus plan already. Previously, operator was only available on the $200 a month Pro plan. But the pro plan doesn't come with data protections like it's more of an experimental plan almost. Whereas the team plan is like the pleated khaki pants sort of business plan.

But everybody on a team plan has.

Got access to this now. It's why I want to get this episode out so that you could communicate to your team ways to use this and ways to not use this right now. And my takeaway right now is a few things. One, right now I'm going to monitor everything that it does. Like, I'm fascinated to test this and see what it can do and what it can't do. But right now I'm going to cut it loose, especially on anything high stakes without actually keeping an eye on it. The reality is like, we have access to enough sensitive information to where, like, we just have to be mindful of ensuring that that thing's not going to crap the bed and not put that information somewhere where it shouldn't be right now. Maybe it's totally good enough right now and that's not even a problem. I don't know. Reality is like, it is too early right now to know for sure. And so that's why I'm saying let's monitor it for now. Second, I still wouldn't trust it with login credentials for that same reason. And there's an easy workaround here because at any point as the agent is working, like, you can see its screen and you can take control of its, like, environment, its browser. So it's literally spinning up its own little like, desktop environment with a browser running in it. And at any point you can take control of the browser and, you know, enter the credentials to log that in and you get a, you get an alert that's like, oh, you know, nothing you type in or you put in here is. Is retained by OpenAI. So at any point you can hop in and kind of get it past that step. So for example, when I did the QuickBooks Online thing, that's how I got it logged in so that I could do the rest of those steps. Now, the Tax Source Doc example, where that went into Excel, is a great example of something that actually didn't even require its own little computer environment. But it's something that didn't work before because ChatGPT with Excel files or anything spreadsheet related was a little bit hit and miss. Like, you open it up and it looks like a spreadsheet that a Toddler put together and still, like, you gotta think about how do I set this thing up for success? Like, if I just gave it a pile of documents and said, summarize this into an Excel workpaper, you're not actually giving it instructions on how you want it to put that together, as opposed to what I did, which is, here's my work paper template. Take all the information from these docs loaded into my workpaper template. I'm just asking it to move some data from a form into this other thing so that the formatting ultimately, like, that's not a big deal to me. I've already locked it down in my template, so I'm not asking it to take that task on. I got derailed before my third thought on how and when you should use this. So first thought is, monitor everything it does for now. Second is, I wouldn't trust it with login credentials yet. Third experiment, like. Like, start kicking the tires on this thing. Because my mind's already jumping to small brain tasks that it would be really nice to have it do for us. Because another really cool thing about Agent is when a task completes you, you can schedule it to recur. What are all the ways you can use this? I just got an email from somebody yesterday that's like, they had bought a firm and they'd inherited kind of some annoying workflows from the past owner. One of them being, like, they work with restaurant clients and they get these, like, daily printouts from restaurant clients, and they have to enter it in a sort of this spreadsheet template. But it looks a little bit different from each client because they're all printing this information from different systems. I mean, there's loads. Like, I had all sorts of examples of this working with dentists, and, like, we had all sorts of dumb little workflow things where it was like, so knowing that it's a problem, but at the end of the day, not the end of the world, because these people are paying us a lot of money to move some numbers around. It's also an example of where has there been a way to automate it previously? Probably, like, you could build a workflow. Like, you can build this sort of thing with Microsoft, not Power, Automate, whatever the cloud version of this is. Like, you can build a little app to extract the same numbers from a form that looks the same every single time and, like, put it into some sort of spreadsheet. There's ways to do this. It would take me, like, six hours of watching YouTube videos to figure it out. But the reality is like most people don't have the awareness or won't make the time to do that. You think about that workflow with ChatGPT agent. It can absolutely do that. Let's say I have my client upload those reports to a shared folder every single day or something like that. And then Agent goes out at 8am every day, pulls those reports from the folder, goes to the spreadsheet and enters like the specific items that you tell it to enter into the spreadsheet. It can absolutely do that right now. And the reaction to ChatGPT agent so far has been. Has been kind of mixed because I would honestly say because it can't do really big complex things, but I would argue it doesn't need to. I think tech folks especially are guilty of this. Software folks, they, they think the problems that we face. And I oftentimes teach software folks that they're like trying to solve 200 IQ problems when my reality is being mired in 60 IQ. Like client headaches day to day, they're not hard problems. Oftentimes they're not technically challenging problems. They, they're just annoying. It's getting information from this system that doesn't have any way of getting information out in an automated way and having to get it over to that system, like that's not my fault. That's a workflow headache. Somebody has to do it. It's annoying and I wish it didn't exist because I'm very overqualified to move numbers from one thing to another thing. But these types of situations, it's what gets me very excited for ChatGPT agent now. Right now this just works in the Internet browser, but there's some other more experimental AI agents that will work on your desktop. And that's kind of the future that we're headed to. Is the same sort of thing that can work in your desktop environment. How cool would that be? The counter argument. And this is, this isn't just with agents, but it's with a lot of things. AI is oftentimes. Well, what about this other way to do it? There is another way to do it. You just got to do X, Y, Z. You got to be able to do like a couple SnapHow, Excel, Macro, SQL Query, and then it just, it does itself. In fact. Oh man, we made the mistake of posting a, a little short video, I may have shared this last week, a short form video about using ChatGPT to like do some cool stuff in Excel. And the comments were just. I mean we got absolutely torched from all the the sweaty Excel folks who are like, you didn't need Chachi to do X, Y and Z. Because everybody always loves to one up each other on like the, what's the Excel thing that they know how to do that the creator of this video must not have known how to do. And so all the comments are calling out like the most advanced Excel stuff that next to nobody knows how to do. Literally someone is like, well, you could have written a line of SQL to do this. It's like, what percentage of people know how to write SQL? And like, for those folks, my, my response is always like, ChatGPT is a tutor for all. Like, you know how to write SQL. This is going to help you write even better SQL. Like working with AI. But I think we're entering this, this era where it's like you've got an AI agent that can do these things and that it'll almost be like the lazy way to get that thing done. Like, sure, there's, there's other like much more sweaty automated ways to probably do the same thing and pour one out for the polar bears. Because those, man, those servers are going to be humming like running everybody's AI agents. I say that 50% joking, 50% not joking.

This episode is brought to you in part by Makers Hub. Hot off the presses. We recently did a video on the main YouTube channel.

If you've seen it yet, it's kind.

Of like a masterclass for how to automate the accounts payable department. Soup to nuts from beginning to end, from how you wrangle all the bills to how you extract the goodies off the bills to how you get them through like an accounting approval that is making sure that the stuff on the invoice is correct and that the correct accounting codes are associated with every single item on a bill to payment approval, which is completely different because usually a different group of people have the keys to the castle to be able to make payment. But because most systems combine those two approvals, you end up with like the person who can make payments approving like the factual stuff on the bill when it's like they may not even actually.

Know whether that was invoice correctly or.

Not clear through to like how payments get released, what accounts they're coming from, when they're going out the door. Super cool stuff. This is like the, the area of accounting that AI has most impacted is accounts payable. I did a load of AP management in my firm. Even if you don't do it, you got clients who do it. I would super recommend actually you share that video we just did with clients that are doing this stuff because it completely changes how you build your in house accounting team. But for you as the firm, if.

You do this work, it's, it's actually a big time opportunity right now.

So get out there and do it. Learn more about Microsoft Check out the.

Link down in the show notes.

This episode is brought to you in.

Part by Client Hub.

Client Hub been one of my recommended practice management systems for a while, but they're doing some really cool stuff. One of the first PMs investing heavily in workflow and if you have followed my Maslow's Hierarchy Pyramid, a food chart of the tech needs of your accounting firm, right now the workflow tools are getting super, super good, but almost feels wasteful to buy it. In addition to a practice management system, that's something Client Hub is working on. So they do all the practice management stuff that you want from the dashboards to track your work tasks that can be assigned to clients. A really good client portal time tracking, CRM, email integration, all that stuff. But they've also got a really good QuickBooks and Xero integration for stuff like resolving uncategorized expenses, even uncategorized deposits, requesting receipts from clients all like in the same client portal they already use.

They're also doing some cool AI stuff, quote unquote magic books review.

Identify expense inconsistencies and other review items across the balance sheet. P and L ClientHub definitely worth checking out if you are a bookkeeping or cast practice. And I love that they're going hard on workflow now.

Learn more about them.

Check out the link down in the show notes.

This is actually always been a thing and it's called, I guess called an abstraction trade off. And so an abstraction is like, you know, as an example there was a time when you had to write code for like every little piece of metal inside a computer. Like you were writing machine code and instructing the computer on how to manage its memory and all these nitty gritty very detailed things. But over the years all that's been abstracted to more like simplified faster ways to develop code where it manages the memory for you and you don't have to think about all that. And you have the really, really hardcore people that understand machine code and they say what Python? That's so lame. That's like kitty code. We can squeeze more out of the hardware and do better resource management and all that with machine code, but it's extremely hard to write and very few people know how to do it and so eventually in all but the most resource intensive situations, people just went to the more simple like abstracted code language. AI agents are going to be a massive example of this. We where there's going to be a bunch of ways to get a job done, but the default for people, it's going to be going to an AI agent and saying hey, every Monday can you just go do X, Y and Z? Because I just, I can't be bothered. And you probably could have written a bit of custom code to do that and make a couple API calls. Maybe you could use something like Zapier to do it. But people won't do that because ChatGPT agent's gonna be so much easier. And so if we bring this conversation back to like the tech ecosystem, a lot of folks at the end of last year were super, super excited and bullish on AI agents and I was like eh, I don't know that 2025 is the year where like everybody's using them and they kind of become this big mainstream thing. I said probably 2026 and that's probably still the case. I'd probably still say like for it to be a thing that almost every business user is using. I probably wouldn't say that's a thing until 26 or 27. But I will say what we've got right now with Agent is more than I thought we would have. And you know, if you look at just some really basic examples in the accounting firm's tech stack, let's say you're using a tool to automate your tax intake and that tool doesn't integrate with your practice management system. And every time a project finishes you have to manually move that, move those files over to some other system. Maybe you have to set up a project or change the status of a project in your practice management system. That's absolutely a thing that chatgpt Agent is good enough to do right now. You set up the recurring job where it's like, hey, every Monday if something comes through here, pull it over into this system, find the client, pull those files over, change the tax term project status to, you know, prepare or something like that. On the bookkeeping side, like we have even more fiddly examples of this because everybody's systems look a little bit different. And so us fetching files from the client and then putting them in a specific specific place, I mean the pre accounting process of what are all the client specific things you got to grab and then put in some sort of file folder, even down to like, I mean fiddly little things with your practice management system where clients will upload files, but then it won't automatically put it into the folder that you want it in. So you've got this like, file upload inbox that somebody has to go through and manage. There's a lot of little things that I feel like Chat GPT Asian can already do for us. They're not big brain problems, they're small brain problems. But how often do we make tech buying decisions based on, well, the native integrations. It can pull this thing to that thing, and I still want that. Like, I don't want to have to have the band Aid solution. But agents like, I mean, the super fascinating thing right now to me is obviously this is the worst that they're ever gonna be. And we're so close to this tipping point that kind of just totally changes how we use tech, right? Like, once that thing is just good enough to do some of these tasks for us reliably, you better believe we're going to jump out of the way. Think of the same things that you do in a month end close in every client's QuickBooks file. Like, what's step one? Like, look at your SOP of what are all the steps that we have to do to get this done? How many of those are simple little standalone things that an AI agent can do for you? Your review checklist, how you sign off on something being correct. I don't want to, like, I don't want to get too excited here because it will let you down. Like, and that's just how AI is. You'll think like, oh, this ought to be a thing that it can do. And just for some reason it can't. And we're still going to be dealing with a lot of that. And like, the more you try to throw at it in one task, like getting it to go and enter 200 things. Not going to do that. Like, it's better suited for small, annoying, you know, daily things. That's probably the use case I'm most excited about right now. But in terms of, like, the overall paradigm of computing and how we use technology, I mean, let's, let's just acknowledge being able to just tell a thing, go and do xyz and it figures out how to do it. You can instruct it on things to fix if it didn't do, if it did it wrong, and then you can set it to recur like, that genie's not going back in the bottle. And I don't know that we're thinking big Enough yet about everything that that impacts. In fact, I was talking recently with a founder of one of the leading practice management systems in our space, and they're talking about like platforms like, like cracking down on bots and like automated tools on the platform. And I'm like, this is right before. This was before ChatGPT agent came out. And I'm like, man, I actually feel like we're very, very close to the day when there's more machines that use these platforms than humans. I'm thinking about that as soon as ChatGPT agent gets just good enough and everybody is using not one agent, but a hundred agents at a time, going off and doing your bidding. Like, I think with ChatGPT agent right now you're limited to like three. Well, with operator, you were limited to three operators that could run simultaneously. I don't know how many agents you can have running simultaneously. I know you can have it do 40 tasks a month for you right now, which is not very many. That'll definitely improve over time, obviously. But this software founder, they're talking about platforms cutting down on bots, accessing those platforms. But if we become reliant upon AI agents, I'm just not gonna use a platform that won't let bots work on it. Right? I mean, that's. There's gonna be a big percentage of our work that starts being done by AI agents. And platforms are gonna have to decide, are we gonna be the agent friendly platform? Are we gonna be the platform that like, doesn't let you do any of that stuff? And it feels like we're really, really close to the tipping point where the tools that we have today are hustled by our AI agents as much as they are by us, our users. Like, you look at where Agent is right now, where could we be by next tax season? I mean, certainly by the following tax season. And so when you have general AI agents that are really, really useful, how does that erode what we ask the rest of our tools to do? Like what things do we then start using AI agents for that we maybe don't use our normal tool set for? I don't know, that's really tricky to get your head around. But I think this is the power in experimenting right now, is beginning to think about these things and ask these questions. The firms that are on the leading edge of this right now, I mean, I think there's actually a bunch of use cases that just became possible. What I'm grappling with a little bit is can I trust it to run on its own. But if the answer to that is yes, we just talked about a firm with their restaurant clients where I think stuff like that is completely automatable. With agents right now. I get very excited not only about what this looks like, but the next, like, the next generation of agents. The one after that, like, then it's like, man, throw out a lot of how we get our stuff done. Start thinking about what are the common tasks that this whole team is doing that can be rolled up into an agent? What are the more valuable things that that team can do? How do we upskill them to be using agents for kind of their very specific workflow headaches? And then how do you put that stuff on autopilot right, so that it just runs itself? I would love to hear the luck or not luck or luck or lack of luck that you're having with ChatGPT agent if you've had a chance to test it yet again. I wouldn't throw big, giant, meaty things at it. My mind goes to what are the little annoyances that I got to deal with that are not complex, but somebody's got to do them? That stuff seems pretty suited for AI. That's what's got me excited. Speaking of AI, so I'm on the road today. Fun AI events as you're hearing this. Let's see, this comes out on Wednesday. Tomorrow I'm in Atlanta, Georgia for one day AI conference, and then the next two weeks, I've got four more live events I'd love to see. Yeah, July 31st in Austin, Texas. These ones aren't AI events specifically, but you also can't really talk tech without talking AI right now. So. July 31st in Austin, next day August 1st in Dallas, August 7th in Minneapolis, and August 8th in Boston. These are free events, people. Come out and hang. Bring your team. I met a ton of you in the last couple of months at live events, and it's been super, super fun. Come hang out, geek out, make some friends. That's the other cool thing with these events, is you meet some local people. Like, almost everyone leaves with a new friend that lives across town that they hadn't met before. So would love to see you there. Thanks for coming and hanging today. I'll see you in the next.