How to Automate More of Your Firm with Agent Mode in ChatGPT

The other thing that I always do with these AI sessions is update you before you get dive into the topic on what's new in the world of AI, particularly in the world of ChatGPT. I focus on ChatGPT because that's what one most people use and there's actually not much new for a change, not startlingly new anyway. If you go to OpenAI's.

Page where they have their release notes, the only things that have really changed is yesterday they did announce an improvement to search. So if you prefer using ChatGPT for search rather than let's say Google, which I do by the way. I love the search, it's my default now. I go there, I go to ChatGPT to search for stuff. One of the things that they've got is fewer hallucinations. This one's interesting that it can automatically detect whether you're looking to buy something or not and therefore the search will pick up sites like e commerce sites or not depending on your intention. And then better formatting, I haven't tested it came out yesterday. I've not had an excuse doing a search but that's new. And then they've also just rearranged in their settings where personalization is. It used to be a separate. It used to be duplicated in two places. They've just tidied that up. There's some stuff around codecs but that's not so relevant to accounting firm owners. So. So yeah, not a lot, which is good because we can kind of make sure that we are catching up with some of the other amazing things they've done during the course the last six, eight weeks with agent mode, with GPT5 and so on. So let's. Let me, let me just very quickly just check there's no questions. But do remember if you've got any questions, do ask. Put the two capital Q's in front in the format that Sarah showed looks like this. And I will answer questions towards the end in about 40 minutes from now. But let's dive into the main content for today. So we're looking at Agent mode and how we can use that to automate a lot of what we do. And this is fascinating. It came out middle of July 2025, so it's not been out that long, but it's got the potential to be absolutely extraordinary. And so I'm going to look at what we'll look at today is what is Agent mode? And let's get practical, let's talk about what we can use it for in our businesses. What's it best for how can we use it? So let's start with what it is. And really it's a combination of two separate tools that came out I think originally back in January of this year of 2025. One of those is one of my favorite features. It's Deep Research. So Deep Research came out shortly after Deep Seq, because Deep Seq back in December got all the Press. But then OpenAI had deep research up its sleeve, released that and it was even better still. I love Deep Research. If you haven't used it, you definitely want to check it out. And even if you're on a free plan now, I think you can do do a few Deep researchers inside inside ChatGPT. I love it for any type of research. Like buying stuff is great. I use it all the time. If I'm looking to buy any products, I just use a prompt that I've created. So it's a combination of Deep Research but also Operator. Now Operator you probably have never heard of because they brought out Operator. Again I think it was January 2025 but but I could be wrong on that. It was about that time and initially and it only came out on the plus sorry, the pro plan, the 200 bucks a month plan, I do have that so that I know what's coming to tell you about. And when they brought out Operator, it was a bit disappointing. Essentially Operator would go and do things. And so.

This agent mode is effectively it's a combination of those two things. It does deep research and then it goes and does things. And I think it's better than the operator. Now it is getting better. I've noticed since it was released about eight weeks ago, it's already improving. It's improving all the time. So what does it allow us to do? Well, essentially.

Agent mode goes and does things. If we think about Deep Research, if you've used Deep Research, what that does is it goes off and searches the Internet and does much more deeper research than just turning on the search facility. If you just turn on search, which is great, but search tends to only look at typically five different sites, different sources and usually the ones that are higher up on Google rankings. Deep Research will go in much, much deeper and typically 20 odd different sources they'll look for and sometimes obscure things. Deep Research does a really good job and what it comes back with, the reports are always amazing, but the reports it comes back with always text based. It basically reads stuff from websites and gives you the output. But what Operator does. Sorry, not Operator, it's now Agent mode. What Agent mode does is it actually opens up on your screen as you'll soon see a little virtual desktop. And you'll watch it clicking on buttons, it will click through to things, and that opens up all sorts of possible possibilities. With autonomous workflows, it can go and do research, analyzing, clicking on stuff, and you can leave it doing that while you're getting on with your work. You could even set it to do these things on a schedule. So it could do it, let's say overnight, so in the morning it's done the piece of work from you. Essentially, it's shifting the emphasis away from simply just chatting to doing stuff, to interacting with apps, to interacting with websites. So it's pretty extraordinary what it can do. There are so many, many possibilities. Now, there are a few things that some teething problems, there are some issues. It sometimes crashes, sometimes it misclicks on things, just like a human sometimes does. And so sometimes we might have to step in. But one of the things we can do with it is we can step in, we can, as it's going through its process, we can pause it, we can correct it, we can add additional comments and say, no, don't do that, do this instead. We can correct it. And we can also take over control.

And sometimes we will need to do that if it's clicking into a site where it needs a password.

So I think the best way to explain it is perhaps to give you some use cases and then we'll dive into ChatGPT and we will look at it. Now, I know some people are listening on the podcast, and if that's the case, then you might at some point want to hop across the YouTube channel where you can see the demonstration later on. But before I do the demonstration, let's talk about some things that we can do with this agent mode on so we can use it. For example, if you're doing accounting bookkeeping tasks, we can do it. We can get it to go into our accounting software, whether using Xero, whether using qbo. And we could do things like we could get it to go into QBO and do a supplier reconciliation. One of the members of my AI community said that he does that and it works really, really well for doing supplier reconciliations. He also told me that he also.

Used it to go into QBO and journal weekly payroll costs, which I thought was interesting. We could get it to go into accounting software and fix broken bank feeds. So I did this one earlier. There's a prompt that I've put in the AI academy now as part of My training course that does exactly that. And I test out on my QBO and it works really, really well. So we can do all sorts of accounting tasks by going into accounting software and doing stuff.

But we can also use it for many, many other things. We could use it for competitor research and.

Give you. Yeah, we'll do a work example on that. We could go and research and search into our competitors, but we can go deeper than that. One of the things we can do is not just go to their websites, but because it can click on things and do things, it can start to go into things like forums and review sites, things like Reddit and look for things like customer feedback, which is really interesting research that would take us hours to do if we did it ourselves manually, is it can do it in a fraction of the time. So that's a really powerful use case if you want to do any form of competitor research. And the prompt I'll show you today, you'll probably want to try out in your firm. But also if you are, if you're an accounting firm doing business advisory work and want to advise clients, then one of the things you could do to add value to businesses you work with is you could do some competitor research on their behalf in a matter of minutes and then share the report with them and they'll think you spent half a day, a full day doing it. It's a great way to add value. What else can we do? It's really good for organizing unstructured data and making it structured. So I did something recently where I do a lot of work in a. I have a home gym here. And one of the things I do is have one of these fancy scales that if you stand on it with bare feet, you also hold something and it measures your body mass composition and it creates a report which is a PNG image with all sorts of data. It's really complicated. And what you can do, what I do is I usually put it into ChatGPT and it then uses its, its image functionality to, to read that image, extract the data and give me information about it. But what I did with Agent Mode is I got it, I fed it a whole bunch of these and I said, I want you to analyze all of them and get all of the data and then build me an Excel spreadsheet or CSV file showing me over the weeks and months of the data, all the key metrics, and it was really powerful. I was amazed how well it did that. So how could you benefit from that? If you're an accounting firm doing Bookkeeping and so on. Well, one of the things you could do with it is if you get unstructured data from clients, information that you want to structure and put into a CSV file. For example, perhaps it's information for their tax return or perhaps it's information, let's say.

It'S a restaurant and they've got daily taking sheets and they're all handwritten. You could take those handwritten images, stick them into ChatGPT and get Agent mode to analyze it all, structure it and put it into a CSV file ready to upload into your accounting system. So that's a really powerful use case, creating structured data from unstructured data. Also, if you want to hook it up to let's say Google, you could get it to, you could set, let's say a weekly agent task where it goes into your Google Calendar perhaps on Sunday night and it looks at all of the meetings that you've got scheduled for the week ahead. And then what it could do is it could go and identify who is the meeting with and, and then go and do some research to find out what's the latest news with that particular person you're meeting and then give you a report so that during the course of the week, as you're meeting various people, you've got a report that tells you about some of the latest stuff that they've been doing, working on, or if they're in the news, something to give you have a conversation around. And that could be done all inside an agentic prompt. You could do it for lead generation. Building up an email list. I might show you that today if we have time. There are so many possible use cases, so many things that we can do. It's really, really powerful. So I think probably what we should do is let's have a look at one of those. So let me go across to ChatGPT now. And where do you find agent mode, first of all? Well, where you'll find it is the plus icon. It's one of the tools. So you just need to hit the plus button and select agent mode and then you'll see a little, it'll turn blue, you'll see that you've got agent mode turned on with an X so you can turn it back off again if you made a mistake. So that's how you access the tools if you don't see them on the list. Sometimes I find the list doesn't always appear. So the other way of doing it is to hold the slash key and then it lists them. So we need to make sure we turn on agent mode first to start it. And because it's fairly new, it's only been out for a couple of months. It does give you all sorts of suggestions of things that you could do. So you might want to play with some of these for a bit of fun if you are new to it and want to see what it can do. One of the things that's interesting with it is that when it's done its deep research and done whatever it needs to do, it can then create outputs in terms of, for example, a spreadsheet. It can build a spreadsheet, a CSV file. It does a very good job of that. And if you're an accounting firm owner, obviously we like spreadsheets. There's a lot of practical use cases with spreadsheets. It can also build PowerPoint presentations, but I find that a little bit limited because the design is pretty awful. There are other AI tools for building presentations if that's what you want, but give it a go. It gets the text and stuff, great. But I just find the presentations themselves are a little bit ugly and you need to do a lot of work to then tidy them up. But, but spreadsheets are really good if you want to create some structured data. Okay, so that's how we turn on agent mode. Let's find a. Let's look for something we can do. And I'm going to go across to the prompt library. So those of you who are in the AI academy or VPA will know that I have this prompt library software that you have access to, where I've created dozens and dozens of prompts that are ready to use copy and paste across a whole bunch of different categories. Prompts for financial analysis, business strategy, marketing, content creation, and even some fun personal use cases. And I have a lot of personal. I use it for personal use quite a bit. So we have all these prompts and I'm going to go across to a new category that I've set up in here that's actually not there. I'm going to go into type. I'm going to go to an agentic prompt. I've written a whole bunch of different agentic prompts for doing various things and we're going to look at first, let's look at competitor customer sentiment. So let me just grab this prompt. The great thing about the prompt library, if you use it is you just find the prompt, click copy as marked down to clipboard, go to ChatGPT and simply paste it in and there's your prompt ready to use and run, or at least almost. We have to give it some information. What's this prompt going to do? So let's imagine that you are an accounting firm, you're the owner of an accounting firm and let's assume that you are in Bedford in the uk. I've just picked a random place, I grew up in Bedford. So let's imagine that you are an accounting firm in Bedford and you're thinking about your strategy, you're thinking about your marketing strategy and you want to look at your competitors. What are your competitors doing well? What are they doing not so well? And also what do your competitors clients think of them? What sort of feedback is floating around the Internet on things that people like and don't like about your competing firms?

And this is a really powerful prompt. It's also one that I think demonstrates some of the great things we can do with agent mode, which is why I chose this particular one. And if you're a forward thinking accounting firm owner, you could then use this, as I said, to do this as part of advisory work and do this for clients. So what we need to do is I'll take you through the prompt in a little while. It's quite a long one. I'll explain why in a second. Those of you that follow me regularly will know that when we're creating prompts, prompt engineering is really important and the more specificity we have in our prompts, the more predictable and the better the output that we will get. And I did a live stream, I think it was last week on prompting now with GPT5. So we've now got GPT5, which we now have learned is we need to build in some new techniques when we craft our prompts, particularly for agentic prompts.

And also when I create these prompts in the prompt library, many people, if you use them regularly, you'll know I often have variables at the end, I don't have variables at the end of this one. For context, I've put it in the middle, so just let me find that bit. So you need to give it some details of you, your firm, what is your business, where's it located, what sort of clients do you have, what are the main offers, what's the website, what are known competitors? And the reason why you'd want it as a template like this is yes, you could run it for your business first, that would be a great thing to do. But. But then what you could do is anytime you want to do this for a client, add value, all you have to do is Fill in these details.

And if you want a shortcut to filling in these details, I'm going to put some details in here. Let me just grab this information, what I did to get this information in the first place. Let me just put it in here. So I want business type control V what I did. Here's my accounting firm in Bedford. So we've got business type chartered accountants located in Bedford, uk. Here's the ideal clients, here's the main offerings, here's the website. I've chosen Cowley Homes. It's a firm I know of in the uk. If you're watching this from Cowley Homes, you might find this interesting. And I've got the competitors of which I know of at least one of these firms very well. There's one very good firm here.

So you just need to put that information in. A little tip, if you're using a prompt like this, I did a little shortcut, I did a separate prompt and I just put in the website of Cowley Homes. So you could do this with your website. Rather than typing, filling in all this in, you could just put your website in and then copy and paste what was here originally and say, give me the answers to these questions and it will then do it in seconds. It gives me this in seconds to copy and paste. Okay, let's look at the prompt then in a bit more detail. Let's go through it from start to finish. I won't go through it line by line because there's some advanced stuff in here. I want to give you the. I just want to give you an idea and pick out and teach you a few prompt engineering techniques. So we have various sections because using markdown, this is the hashtags, it creates a structure. And one of the things we know is that when we have more structure in our prompts, it really helps the AI system follow them. We also want to give it a role to play. So I'm asking ChatGPT to play the role of a market research analyst. And what we know is that this gives better results. If we tell it to play a particular role, we'll get the sort of response we'd expect from a market research analyst. And then we've got the objective or the goal. And one of the things that we know about Agent mode.

Please don't feel overwhelmed by the size of this prompt because one of the things we know from what the research team at OpenAI have told us, is that when you are crafting a prompt for agentic tasks, arguably the most important element is to specify the goal or the objective? What is the goal that you're trying to achieve? Because the better you explain the goal because of the way it then goes through and thinks things through, it will figure out the best way to achieve the goal. And that does mean that if, if we have a really clear objective or goal specified at the start of our prompt, we can get away with a less specific prompt because it will figure that out. So always make sure that you have your objective or goal in your prompt. Now, we then have some other things here which are more advanced and I have a separate course all about agent mode and prompt engineering. But this section here where it says begin with a concise checklist of three to seven bullets of what you will do keep items conceptual, not implementation level is because again, the research team at OpenAI have said to us that what they found in all their research is if you get the agent to start off the process by telling you what it's going to do in a high level a checklist, then it's more likely to accurately then do those things and give you a better output. So that's why I often have something like that in my agentic prompts to ask it to start with. With coming up with a checklist of what it will do, then we have the contacts. This is the business details that I copied and paste. This is me. If I was the owner of this firm, I would put in Chartered Accountants Bedfordshire and so on. So we have to give it some information and basically it's done all this from based on the company's website.

Then I've got the objective again, I've repeated the objective here. And then we've got specifically what we want it to do. We want it to do competitor research. So start off by using its browsing facility to look for the top 10, five to 10 competitors relevant to the business.

And then what I want it to do is research each of those competing accounting firms and tell me, well, what are the key features and benefits? What do they do? What if available, is the way they price? What are their unique selling points? What do they say are their differentiators? Have they got any proof like testimonials, ratings, case studies? And give me the URL so I can go and check them out if I want to. That's the first thing that I want it to do and I want it to give it to me in a. So we've got deliverables. I want that information in a CSV file with these fields in the CSV file, the Excel spreadsheet If you like. And here's an example of what a record might look like. All I'm doing here is being very specific in terms of what do I want it, what's the output, what do I want it to give me? I want it to give me a CSV file with these fields. You might want a report, you might want a PowerPoint presentation being created. That's entirely up to you. I feel like for this that a CSV file is most useful. And then in part two, I wanted to then go and collect verbatim quotes from clients. And this really illustrates, I want to show this one because this is part of the power of Agent mode, because I'm specifying that I want it to look for client quotes relevant to those firms using public platforms like Reddit, Quora, YouTube comments, Google reviews, Yelp, Captura, Trustpilot, etc. Because what it does, it will then go and click through and this is the agentic side, it will go through and click buttons and things to go and find the comments. This is one of the really powerful things about agent mode, where it goes and clicks on buttons inside websites. And then what I wanted to do is when it comes across a particular quote from a client of one of those competing accounting firms, I want it to tag it. Is it a problem? Is it a complaint? Is it a praise? And my deliverable again is I want a CSV file and I've given an example of what I want. And then also as a deliverable, give me a pie chart of the top complaint themes. What are the biggest complaints that people are saying about my competing accounting firms? And then I want as well, I want a summary. I want you to give me a summary of what to copy. So what are the good things my competitors are doing in terms of their offers, the features, the messaging? So what are the good things I could learn from and copy? And what are the bad things? What are all the complaints, the pain points, the problems that I can then make sure that I address those in my business and perhaps build some USPs around those. So I stand out as being the go to firm.

And then I've just clarified again the output format here, the spreadsheets and the pie chart that is the prompt. So I know it's a long prompt. I just like to create ones that are high specific. But essentially what we're doing is we're analyzing my customers, my competitors, we're looking at what they're good at, they're bad at. We're going across all their websites reviewing stuff and then we're going to dive in and look for client quotes. Right, I've got that done, so let me just hit the button. And we're not going to see the end output here because agent mode can take a little bit of time. Sometimes it can do it in two or three minutes and sometimes it can take. My record is 48 minutes. So you can see it's now saying, setting up my desktop. It's now creating a little virtual desktop. What it's also doing first, because I told it to, is it's now telling me what it's going to do. This is the task it's going to do. And you can see now that it's opened up the desktop, it's searching the web, we can see what it's clicking on and it's giving commentary as it goes. I find this fascinating. It's going through all these different accounting firms in Bedford. It's researching, it's clicking on buttons, it's looking for stuff. And whilst it's doing that, whilst it's going through and doing that process, we have three little buttons at the top here. If we click on those three buttons.

We can see the activity in terms of text, what it's doing. We can stop it if we want. If we need to stop the process, we can do that, hit a stop button and there's also takeover the browser. What that means is that we can step in at any point in time, we can click on that button and we can start clicking on some buttons wherever they are in the site. So we have an element of control here. What we can also do is, in the chat below, it says follow up. And what that means is that whilst it's doing its thing, we can correct it. For example, we can give it feedback if we think in. If we're watching it and it's doing something wrong, we can actually type in and say, hang on a second, that's not right. I want you to be looking for, for this, this and this, and it will then correct its course. So that can be really useful. I do find it a bit addictive looking at, Watching these things. I'm fascinated at the thought process it's going through the buttons it's clicking on. But as I say, sometimes, sometimes it's quick, it can be done in two minutes, sometimes it might take much, much longer, often five to 10 minutes. So these are the sorts of things that you want to let it just get on with.

Get on with the task while you go and do some different work. And you can. You can Run, have multiple agents running at the same time. And I, because I use ChatGPT all the time, I am often got ChatGPT open in multiple tabs, multiple browsers, multiple computers. I'm using it constantly for everything I do. I will have it open on the Mac desktop app. I'll have it in Chrome, I'll have it open in Safari. I usually have a minimum of three versions open and doing stuff at the same time. So you could certainly be running multiple agents if you want to do lots of different tasks all at the same time. Okay, I'm going to let it do its thing and see it may or may not give us an output I don't know before the end of the livestream. I hope it will. But let me show you another one I did earlier. I'm still watching it because I find it fascinating what it's doing. It just is absolutely extraordinary. Let me talk about another idea that you could another use case. You could use it for list building. Let's imagine you are that firm in Bedford. That's your firm. And let's imagine you decide that you want to get some more clients in a particular niche. It may be that you are looking for.

Plumbing firms and you want to build up a database. You want to build up the contact details of other plumbing firms in your area, in the Bedford area, if that's where you were based.

So if I go back to my prompt library, I have one that I've created called List Building Agent. And this particular agent will go through and build up an email list for you and output it in into. And output it into a CSV file. I'm not going to copy and paste that because I've got this one running already. I'm going to open up one I did earlier. So let me just go and open up a sidebar and I'm looking for List Building agent. It should be here. Let me minimize that. Okay, so I ran this particular prompt. This is a skilled business research assistant and it's going to go and research. If I go to the end for a second. I've put in the variables. I want 20 email addresses, 20 contact details. I am in Bedford in the UK and I want plumbers. I'm looking for plumbers. So you could just. That's the variable with this prompt. If you're looking to build an email list, just specify how many contact details, where are you based and are you looking for plumbers, restaurants, engineering firms, whatever it might well be and it will go and do the task. And essentially what the task Is here is to go and do research.

And so it said here, here's the proposed plan for compiling a list of 20 plumbers in Bedford. So it's telling me what it's going to do now based on the prompt, because it's good practice to get it to tell us in advance, what is it going to do? This is a bit like, by the way, a bit like with humans. If you've got. Let's imagine you've got a new hire and you want them to go and do a bank reconciliation and perhaps they've not got a huge amount of experience. If they just go away and do the bank reconciliation in silence, they'll probably come back and do it wrong. But if you get them to tell you what is the goal, what's the objective of the work they're going to do and then to think about and tell you the steps they're going to go through, if they tell you those steps and they make sense, they. They're more likely to do the job right in the first place. So that's why this is such a great tip when we are prompting for agentictask, get it to tell you in advance what it's going to do. And so that's what it did. It took 26 seconds. It thought about what's the workflow? And here it is when I ran this prompt and then it thought through a little bit more. This was actually quite. Yeah, this took 36 minutes, this particular one. It was going off. It searched 64 searches across 429 sources. I can't show the virtual desktop because once it's finished, that's it done. This is just the end result. But what it did for me is it created a CSV file that I could download. Let me just do that. So I'm just downloading the CSV file and that should be done. So if I just go to my downloads file, which I think my folder I've got here somewhere.

Bedford Plumbers, Leeds. Let me just open that up and let me just show you that. So if I switch across to screen, this is what it created for me. It created A list of 20 plumbing firms in the Bedford area. It's given me their website, it's given the name of the owner, it's given the email. Now two of them, it couldn't get an email address, but the rest it did. I've got a phone number and I've got some source notes and I could have asked for other bits of information. There's all sorts of things I could have asked for with this. And then if you really want to take this to another level, you could use a prompt to go and do research to build up an email list of your target market, let's say plumbers. What you could then do is you could run an agentic task that then writes the emails for you. And you could even in the first prompt, you could ask it to identify when it does the research, something about that particular firm, if it's in the news or something. Because then what you could do is write a series of emails. But rather than one generic email to go to all 20 of these new leads, you could get ChatGPT to write an individualized email where it might personalize it with something it's found from its research. They may have published a new press release recently and you want to reference that in the email to grab their attention. And then what you could also do, if you want to, you could connect it up to your Google account and it could even draft the emails and send them out automatically. All of this stuff can be done with agent mode. It's insane how powerful it the potential. It has the limits. Largely our imagination. I think of how to use it a little bit. Like perhaps the first time you use ChatGPT, I remember back in 2023, I saw this box saying this empty box. And I'm thinking, what on earth should I get it to do? And the more I use it, the more I realize I can use it for everything. Same with agent mode. You're going to soon find the more you use it, you'll be using it for so long. Many things automating so many things in your accounting firm. Okay, so that's, that's lead generation. Let me go back to. I don't know how it's getting on with that other task. Nope, not new chat. Let me go back to.

Here we go.

I think it's done. Yep, we've done this task now. So it didn't take too long. It's probably about what, 10 minutes. And what it's now done is it or seven minutes. It worked for seven minutes. It did nine searches, 81 sources and the first thing it's given me, two Excel spread CSV files. One is the competitor research and the other one's the customer feedback. It's also given me a pie chart of the most common complaints. So what is the most common complaints our competitors get? It's split between fairly evenly high fees or price poor, negative advice, lack of information or transparency. We've then got a summary report here as well. That's.

Useful with tactics and things that we could do to capitalize and do what they're doing well and also avoid the things that.

Avoid the various complaints. And then we also have.

Sorry, I didn't show you that screen. I just noticed that. Here we go. Sorry, screens here. So here it is. Yeah. So we've got our links to our two CSV files, pie charts there. And then we've got the report. And then if I do open up the CSV files it's given us on our customer research. Let's just open those up and if I pull those in.

This is the competitors it's come across. That's research. It's got information about them, the address, the URL, what they do. It's got one of the things they do well, what they say they do well, like zero cloud specialists, no hidden costs and so on. There's so much information here about these competitors that we could then use when we're building our marketing strategy. That's the first spreadsheet. The second one then is gone. And research what customers have said about them.

So.

Tax assist Letchworth saved us overpaying our capital gains tax requirements. That's under the problem tag, the complaint. Avoid this company. They do not respond to complaints and only send you back to the branch you had a problem with. And we can find out more information here, which one that's, which firm that is, and so on. So I think that's really fascinating, the quality and the depth of the research that it did. In how many minutes was that? I can't remember. That was done.

In seven minutes. In seven minutes we got all that really powerful information. Okay, what should we talk about next? Well, let's just look at some of the limitations of it. And.

Also if you, if you want to go deeper on this, if you're starting to see the power of agent mode and you want to start using it in your firm. I have just completed a course. I finished it today. I've got one more video to upload and the course is in the AI Academy. So if you're not in the AI Academy, I've got lots of courses here. Oh, Rebecca's just joined. Thank you, Rebecca. I've got lots of courses here, one of which is the brand new agent mode essentials. And in here I go through a lot more detail agent mode, including in much more detail use cases. And I've written all the prompts. You've got the prompts you can use, you can use whenever you want.

And then I've just got some advanced tips and one more video to upload. So if you want to find out more, that course is in the AI Academy along with many, many other courses on how to get more out of ChatGPT. So let me wrap up with.

A couple of other things. So some of the limitations, some of the limitations of Agent mode and bear in mind it's only been out for a couple of months. You will sometimes find browser errors like 404 error messages. Sometimes it clicks on the wrong thing and that happens occasionally. I've noticed in the last six weeks I've been using it, it seems to happen less and less. I think it's getting better, I think it's improving. Another issue is session timeouts. Sometimes it will just timeout for whatever reason if it's taking too long, sometimes there's a speed issue. It takes time because it's going highly detailed research. It's not like the prompts that we're used to with GPT 4.0 where we get answers inside three seconds. Yeah, it's doing some detailed stuff, some highly detailed research also. We just have to be aware of security issues. So one of the prompts I created inside the AI Academy was how we can use it to go into our accounting software. Whether that's Xerox qbo, it can go in and it can go and identify every single client file for each client, the bank feeds. And it can identify do the bank feeds correctly connected or not. And then it produces a detailed CSV file with listing all the clients, all the bank accounts and the state of the bank feeds. I tested that earlier. Works like a dream. But we do have to be aware of security issues because you do need to give it access, you will need to give it access to passwords. And so when it's things like that, what you want to do is take back control of the little virtual browser. And that's easy to do. When I did it earlier, it simply got to a stage where it got to the password bit and it said to me, I need to enter it. And so it took control, but I took control back. I just filled in the password details and then I passed control back to ChatGPT. It finished off the task and did the analysis. So pretty, pretty powerful stuff. Okay, there's so much more I can talk about, but we'll perhaps get to some more advanced stuff in another session. What I do want to talk about though, before we get to questions. I'll take questions in a little while. What I want to wrap up with is, is if you are not yet in the AI Academy, then we are currently running now a seven day free trial. We don't normally do this, we do it very, very occasionally and we're doing it for the duration of this live stream and a little bit beyond. So if you want to join the, if you want to, if you want to try out the AI Academy, my team is going to give you a link to a page that tells you what's in the AI Academy. It's a community of currently over 300, nearly 400 business owners, accounting firm owners, bookkeepers, all sharing with each other how AI is working, how they're using it, and also asking me questions. And then there are dozens of courses. And if you go to the page that Sarah is going to give you the link, you can go in here if you want and you'll find out all about all of the courses. If you want the detail, it's all in here. And if you decide this is right for you, then just hit the register now. And when you do, you'll be inside the AI Academy where you'll get access to the courses. You'll also get access to the community itself. So you can ask me questions. I post in here regularly as well. In the calendar you will see the live events taking place. So right now we are live, as you can see. So if you're in the community, you'll know when I'm live because you'll see a button that will take you straight to the live stream. And then the thing that people say they love the most about the Air Academy is you also get some free software with all of these prompts that I've written specifically for business owners, particularly accounting firm owners, bookkeepers. There are dozens and dozens of prompts to help you with business advisory, financial data analysis, marketing, copywriting, client communication, text analysis, and so much more, as well as some fun personal use cases. So if you want to take advantage of the free trial, the way it works is really simple. Click on the link that Sarah will share with you, the link that I just showed you, click on the button that says register now and that'll take you through to the about page. And for you it'll show free trial. If you do it now, what you need to be aware of is the free trial is going to expire half an hour after the live stream. So just to be really clear, because sometimes people watch the replay, the free trial is available between now and 12:30 Eastern Standard Time or 5:30 in the UK, then the free trial ends. So if you want to take advantage. If you want to play with it for seven days, watch the courses, be part of the community, and then make a decision whether you want to continue. Now would be a great time to do that. And the only other thing I want to that you need to know is if you want to be part of the free trial. You can also register your copy of the prompt library software and we'll turn that on and authorize you straight away. The only thing is that you will only get a few sample prompts for the first seven days. And then at the end of seven days, if you decide to continue, then we will unlock for you the full list of AI Academy prompts. Okay, so why not do that now? If you're not part of the AI Academy, it's free. Give it a trial, spend seven days, explore it, watch the courses, play with the prompt library, have a look at the prompts, ask questions, and then anytime up until day seven, you can just cancel really easily if it's not for you.

Okay, give it a trial, spend seven days, explore it, watch the courses, play with the prompt library, have a look at the prompts, ask questions, and then anytime up until day seven, you can just cancel really easily if it's not for you.

Okay, I think we're ready for questions. Let's see what questions we've got. I think we've got a few questions and yeah, I can see three questions. Three questions at least so far. By the way, if you found this useful, valuable, if you could do kind of the usual, I'd really appreciate it if you just hit the like button. If you've learned something about agent mode, just let me know by hitting the like button. And also, if you're on YouTube right now, if you haven't subscribed, please subscribe. I'd really appreciate that.

Okay, questions. Rosalind was the first question I can see. If you remember the vpa, is there special pricing for members to join the AI Academy? There's not because it's a separate business. It's entirely separate, Rosalind. But if you're in the Value Pricing Academy, what you will see is that your investment in the AI Academy is really tiny at the moment because we only launched it this year. It's currently just $59 a month to be a member. We've kept it really low. The price will be going up. In fact, the price is going up in a few weeks time. But if you join now, you lock in the price forever. The price will never go up. It's one of the things that we do inside the academy, just like with the VPA as well. Roslyn so if you decide that this is something you want to go deeper in with AI, then this is the place to be because all the AI stuff is in one place. Laurie said how do you protect yourself from having Agent Mode and deep research? Click on malicious links or websites? That's a really good question.

And.

I don't know. One of the things that one of the points I made in the video training, which I did today is that we just have to be careful with security issues. And the thing is this is so new, it's only been out for eight weeks and one of the things that I do is I monitor this space every single day. It's kind of standard best practice just to be bit mindful of security. As long as you're aware of security risks, then just be careful. But I haven't yet, I haven't yet come across any reported issues around security. If I do, if I come across anywhere on the Internet somebody reporting security issues, I'll be posting it straight in the community and say be careful, don't do this. But as far as I can see, so far from everybody I follow who's been using Agent Mode for eight weeks now, there haven't been any reported security issues.

The one thing that I would say though is there's kind of two levels of if you are using it to do research and it's just doing stuff that's read only, it's going in and reading things like.

What's the word I'm looking for? It's looking things like looking for research, sorry, for feedback things like Reddit sites, it's looking at customer issues and so on. That's kind of. That's read only, that's relatively safe. I think the thing that you'll need to think about is do you want to give it access to things like your QBO or Xero file? I think that's a decision you'll need to make because once it's in there in theory, it could start messing around and changing things. I know lots of people who are letting it loose on QBO for example, and I've not had anybody say there's any problems whatsoever, but we just don't know. This is so new. This whole AI space is moving so fast. This agentic stuff is very new so far. OpenAI have a pretty good track record when it comes to security, so I don't have too much of an issue with that. But Laura, you are absolutely right to ask the question. It's a very sensible question.

Okay, I don't know what your proper name is, but.

Says may ChatGPT disclose your passwords entered manually during the agent session to third parties? Will it remember in future? I'm concerned about potential data breach. My understanding is, and again, this is because it's so new, I don't know definitively, but my understanding is that when you take over control of the little virtual desktop to put in passwords, it doesn't know what those passwords are. So that I believe is absolutely fine. I don't think there's an issue with that. The one thing that I'm a little bit nervous about is more to do with the connectors where you connect up to things like your Google account and it's permanently connected, although that's been out for again, a few months and a couple of months now, and I've not heard anybody reporting bad stuff yet. If I hear of that, I will let you know. But again, so far I think we just have to be aware of these things and if other people are doing these things and reporting no problems at all, I think I would feel fairly, fairly safe. But again, just like Laurie's question, I think it's a very sensible question to ask. Hi there, John from Ireland.

And John, thank you so much. You're in the community and you told me a bit about your part of the world and how beautiful it is. Just so happens that one of my friends, one of our friends, Sarah and I's friends here in Ericeda is flying there right now and I sent him your list of places to go and visit in Ireland, John. So thank you so much for that. He's really looking forward to seeing some of those places. He says, John says, Mark, can you give me your thoughts on ChatGPT versus perplexity? I find that Perplexity gives better answers for accountants, that is finance answers. I can sort of do that, but not really because that's a better question to ask Emily when she does her sessions because Emily uses a lot of the other has researched and looked at the other ones. I know she uses Claude a lot. For example, I've never even looked at Perplexity. And the reason for that is because I follow a lot of people, I have friends in this space and whilst it does some amazing things, I've taken the view that I would rather use ChatGPT to its fullest and know everything about it. So I can, I can help you with ChatGPT and all of its features and functionality and because as well, every time something else comes along that does something better. It always seems that in a short While later then ChatGPT do something and go back to the top of the list. I know that, for example, Claude. Some people use Claude because they think it gives more human like responses. The responses are better worded. Personally I have not found that, and that may be because I've trained ChatGPT in my writing style and I don't see that as being an advantage. Claudette also has at the moment the advantage of being better for coding. But OpenAI have done some amazing things just recently in that space as well. Perplexity is fantastic. I've heard amazing things about perplexity. So my advice, John, is if you're happy with Perplexity for searching stuff, carry on using it.

If you feel it's a material difference, then absolutely do that. But unfortunately I can't give you a view on that because I've never used Perplexity. Anyway, you went on to say Perplexity integrates real time web search as a core feature, so for accounting queries or tax changes, it can pull from recent sources. Okay, so thank you for that feedback as well. Yeah, I've not used Perplexity, so I can't give you a sensible answer.

Sorry. Okay. I don't know what your real name is, CodeQueen, but I'll just go with that if that's all right. Is there a difference between really easily if it's not for you?

Okay, I think we're ready for questions. Let's see what questions we've got. I think we've got a few questions and yeah, I can see three questions. Three questions at least so far. By the way, if you found this useful, valuable, if you could do kind of the usual, I'd really appreciate it if you just hit the like button. If you've learned something about Agent Mode, just let me know by hitting the like button. And also if you're on YouTube right now, if you haven't subscribed, please subscribe. I'd really appreciate that.

Okay, questions. Rosalind was the first question I can see. If you remember the vpa, is there special pricing for members to join the AI Academy? There's not because it's a separate business. It's entirely separate, Rosalind. But if you're in the Value Pricing Academy, what you will see is that your investment in the AI Academy is really tiny at the moment. Because we only launched it this year. It's currently just $59 a month to be a member, we've kept it really low the price will be going up. In fact, the price is going up in a few weeks time. But if you join now, you lock in the price forever. The price will never go up. It's one of the things that we do inside the academy, just like with the VPA as well. Rosalind so if you decide that this is something you want to go deeper in with AI, then this is the place to be, because all the AI stuff is in one place, Laurie said How do you protect yourself from having Agent Mode and deep research click on malicious links or websites? That's a really good question.

And.

I don't know. One of the things that one of the points I made in the video training, which I did today, is that we just have to be careful with security issues. And the thing is, this is so new, it's only been out for eight weeks. And one of the things that I do is I monitor this space every single day. It's kind of standard best practice just to be bit mindful of security. As long as you're aware of security risks, then just be careful. But I haven't yet, I haven't yet come across any reported issues around security. If I do, if I come across anywhere on the Internet somebody reporting security issues, I'll be posting it straight in the community and say be careful, don't do this. But as far as I can see, so far from everybody I follow who's been using Agent Mode for eight weeks now, there haven't been any reported security issues. And the one thing that I would say though is there's kind of two levels of if you are using it to do research and it's just doing stuff that's read only, it's going in and reading things like.

What's the word I'm looking for? It's looking for research, sorry, for feedback, things like Reddit sites, it's looking at customer issues and so on. That's read only. That's relatively safe. I think the thing that you'll need to think about is do you want to give it access to things like your QBO or Xero file? I think that's a decision you'll need to make because once it's in there, in theory, it could start messing around and changing things. I know lots of people who are letting it loose on Agent Mode, ChatGPT and Agent Mode Copilot functionality. Again, I don't use Copilot and that's largely because certainly until recently it didn't get good reviews at all. And I think the only reason you would use Copilot is if you want the benefit of Copilot is you are in the Microsoft space and there's more integrations and so on and I get that and I think that for some people Copilot is great. I also think though that for most things ChatGPT is better, it's just that it doesn't integrate as seamlessly with all the Microsoft stuff. So when it comes to Agent Mode with Copilot, I can't comment on that because I don't know much about it. I've not used it as far as I'm aware though that the Agent mode in Chat GPT is the most sophisticated, the best one. So if anybody has used a different version of Agent Mode inside a different tool like Copilot, please give me feedback. I would love to know and if you tell me it is much better then I will do some deeper research on that. But I haven't found anybody to say anything that tells me that that should be the default that we use.

Right. I think we have three minutes left. If there are any final questions, please do ask.

If you.

If one of the things that we touched on earlier when I showed you the the prompts are quite long that I showed you earlier. That's because one of the things that we've learned from OpenAI and the research team is that we do have to change the way we prompt a little bit when we are using both GPT and Agent mode. In fact, I mentioned this on my last livestream. They did introduce, they did bring out back on August 7th a GPT5 prompting guide which is really useful where they talk about why prompting is so important with GPT5 and with, with with Agent mode. The only problem with this is it's really highly technical and hard to read. So I qbo, for example, and I've not had anybody say there's any problems whatsoever, but we just don't know. You can have a read of this if you want to go really deep on this. But what I did is I spent, I spent several days last week. This is so new. This whole AI space is moving so fast. This agentic stuff is very new so far. OpenAI have a pretty good track record when it comes to security, so I don't have too much of an issue with that. But Laura, you are absolutely right to ask the question. It's a very sensible question.

Okay. I don't know what you're probably going through this line by line and building out for you how to do prompting. So I know quite a few people have now joined I've heard the I've heard a few people joining the academy. What I will do is just point out that if you want to learn more about prompting then in the prompt engineering course I filmed last week a whole new module here called prompting with GPT. My name is but Stupochenko tax says may chatgpt disclose your passwords entered manually during the agent session to third parties? Will it remember in future? I'm concerned about potential data breach. My understanding is and again this is because it's so new, I don't know definitively but my understanding is is that when you take over control of the little virtual desktop to put in passwords, it doesn't know what those passwords are. So that I believe five and in there I teach you some of the tricks, the techniques to get better results out of GPT5 and some of those things are techniques that you'll also want to use when leave is absolutely fine. I don't think there's an issue with that. The one thing that I'm a little bit new using the agent mode. So when we looked at some of these prompts you may have seen that these prompts here, these use cases, they are quite long. But that's because nervous about is more to do with the connectors where you connect up to things like your Google account and it's permanently connected. Although that's been out for again a few months and a couple of months asking for it for example to plan in advance. You tend to get better and more consistent now and I've not heard anybody reporting bad stuff yet. If I hear of that I will let you know. But again so far I think we just have to be aware of these things and if other people are doing these things, consistent results. Okay, I've got time for one more question that's just popped in from Michael. Hi. Reporting no problems at all. I think I would feel fairly, fairly safe. But again, just like Laurie's question, I think it's a very sensible question to ask. Hi there John from Ireland.

And John, thank you so much. You're in the community and you told me a bit about your part of the world and how beautiful it is. Just so happens that one of my friends, one of our friends, so nice friend here in Erica is flying there right now and I sent him your list of places to go and visit in Ireland. John. So thank you so much for that. He's really looking for Michael. Good to see you again. Can we use ChatGPT to summarize and help us change 4.0 prompts to 5.0 prompts. Yes, you can, Michael.

Absolutely. And I'm doing that at the moment because those of you that use the prompt library, I am improving a few as I go along. And what I recommend you do, Michael, because I think you're in the AI academy, but forgive me if I got it wrong, but if you go to seeing some of those places, he says, John says Mark, can you give me your thoughts on ChatGPT versus perplexity? I find that perplexity gives better answers for accountants, that is finance answers. I can sort of do that, but not really because that's a better question to ask Emily when she does her sessions. Because Emily uses a lot of the other research and looked at the other ones I know she uses Claude a lot. For example, I've never even looked at perplexity. And the reason for that is because into the AI academy and you go to the classroom and you go to this prompting with GPT5. Check out the I follow a lot of people. I have friends in this space and whilst it does some amazing things, I've taken the view that I would rather use ChatGPT to its fullest and know everything about it course the training video on meta prompting. So meta prompting is how we get ChatGPT or the AI system to write the prompts for us. And what I've done in the prompt library, if we go to prompt engineering, which is a category you will find in here, I've created some meta prompts so I can, I can help you with ChatGPT and all of its features and functionality and because as well every time something else comes along that does something better, it always seems that in a short While later then ChatGPT do something and go back to the top of the list. I know that for example Claude. Some people use Claude because they think it gives more human like responses. The responses are better worded. Personally I have not found that and that may be because I've trained ChatGPT in my writing style styles for you. So this GPT5 agentic metaprompter is. Is basically a prompt. Well, and I don't see that as being an advantage. Claud also has at the moment the advantage of being better for coding. But OpenAI have done some amazing things just recently in that space as well. Perplexity is fantastic. I've heard amazing things about perplexity. So my advice, you give it your GPT4O prompt and then you run this and it'll turn it into insert the prompt at the end and it'll turn it into a GPT5 prompt. Michael It's a really powerful and simple little prompt we can run just to improve our existing prompts and you are absolutely welcome. Okay, it is five Johnny's look, if you're happy with Perplexity for searching stuff, carry on using it.

If you feel it's a material difference, then absolutely do that. But unfortunately I can't give you a view on that because I've never used Perplexity. Anyway, you went on to say Perplexity integrates real time web search as a core feature, so for accounting queries or tax changes it can pull from recent sources. Okay, so thank you for that feedback as well. Yeah, I have no clock here in Portugal and in the UK and I did say that we wrap up these live sessions. Also Sarah and I go on our first vacation for a few months. We're driving through the night. I've not used Perplexity so I can't give you a sensible answer.

Into Spain in a few hours time. So make sure if you're not in the AI academy and you want to benefit from that free trial, I will be switching it off.

Okay. I don't know what your real name is CodeQueen, but I'll just go with that if that's all right. Is there a difference between half an hour and so now is your chance to do exactly that. My agent mode ChatGPT and agent mode Copilot functionality Again I don't use Copilot and that's largely because certainly until recently it didn't get good reviews at all and I think the only reason you would use Copilot is if you want the benefit of Copilot is you are in the Microsoft space and there's more integrations and so on and I get that and I think that for some people Copilot next live stream I think is in three weeks time in October and my next live stream I believe. Great. I also think though that for most things Chat GPT is better, it's just that it doesn't integrate as seamlessly with all the Microsoft stuff. So when I'm going to be talking about value pricing and how we can use it comes to Agent Mode with Copilot I can't comment on that because I don't know much about it. I've not used it as far as I'm aware though that the agent mode in ChatGPT is AI to help us build our value pricing systems, help us create our value pricing strategies, strategy. The words that we say as you know I love value pricing. So we're going to do a session on the most sophisticated, the best one. So if anybody has used a different version of Agent Mode inside a different tool like Copilot, please give me feedback. I would love to know and if you tell me it is much better then I will do some deeper research on that. But I haven't found anybody to say anything that's tells me that that should be the default that we use a live stream on how we can use AI to help us with value pricing. Okay, as usual I will play out with some music right? I think we have three minutes left. If there are any final questions, please do ask.

If you.

If one of the things that's we touched on earlier when I showed you the music, it will give me chance for three minutes to read your comments. So please let me know in the chat what's the best thing that you've learned over the course of the last hour? It's great feedback for me to know what's been valuable. What have you learned in the last hour? What's been useful? I will see you when I'm back from vacation. Bye for now. The prompts are quite long that I showed you earlier. That's because one of the things that we've learned from OpenAI and the research team is that we do have to change the way we prompt a little bit when we are using both GPT.