Loading summary
Transcript25 lines
- [00:00]
A
Welcome to the IA on AI Podcast, part of the Audit Podcast network where we bring you weekly updates on AI from the internal auditor's perspective. So if you listen to the Audit podcast yesterday you got a little bit of Brian Kinsey, his opinions there. And so now he's back for our AI segment. So to kick things off, one, if you didn't listen to Brian's episode, go check that out. This way we can skip Brian's intro. We don't have to redo that again. So there we go. Let's start with Claude Cowork. Brian, what is that? What are plugins? Just briefly for the sake of the audience here.
- [00:39]
B
Yep. Let's define some terms. So Claude is anthropics AI LLM. It's the product. And so they offer three things. They offer Claude Chat, which is just normal, you can ask it just like you would any AI tool. Claude code, which is kind of how Claude took off it is the command line interface where you, it's basically a coding agent. And then they, they just released Cowork, which is what we're going to talk about, which is the desk. It's basically Claude code in a desktop version ui, easy to use for non programming people.
- [01:32]
A
And the, the end result of working with Claude Cowork could be, should be,
- [01:36]
B
would be the end result is, is it opens up your, It's. I would think Claude. I would think of Claude Cowork as a, a, an actual personal assistant agent that lives and can operate on your desktop, it can get to your files, it can help you do tasks, et cetera. And we'll get, get to why that matters here in a second. But you could essentially what they did, they being anthropic, is they took Claude code and you could build out agents for yourself in Claude code. And then they, not only did they add cloud code to the desktop app that they released, but they also created this thing called Cowork which just made the things that you would build to do all the stuff for yourself easy and accessible for all of us non, you know, all the non programmers in the world. Right. So what a lot of people are
- [02:30]
A
doing right now is kind of, I mean a lot of different things, but one of the more common ones, someone has like their to do list, whether it's like an asana or todoist or something like that. And maybe it goes through there and prioritizes that for you, assigns tasks to certain people and due dates and things like that. Obviously you would go back and review that and then next step further could be if one of the to do items is book a haircut, I'm saying this as a reminder to myself because I have that tab open for me to do. Literally when we get done with this book a haircut, then you could also basically have it start doing some of these things for you also.
- [03:07]
B
Oh yeah, and you, I mean, and they, and, and what, what is big for. For our audit for your audience is plugins and the, the, the plugins that Anthropic released, which on the screen here I have fiance up, but you'll see over here they have all these different plugins. So what is a plugin? What does that mean? And every, you know, you, if you're using ChatGPT, you know, which is OpenAI or Gemini, which is Google or even Copilot, like everybody's got different names for different things. Full disclosure, I am a Claude user. I have been. I. That was the horse I chose. I thought they were going to be the winners of the finance Accounting AI race if there was one. And so far I feel very confident in my horse choice, but I'm not affiliated with them in any way. I. Whatever. So plugins, when you talk about Claude plugins, it's called Plugin Marketplace. It's essentially just a package of tools that allow the LLM to connect to the world. So you'll hear mcp, which is model context Protocol, which is akin to like an API. So if anybody's familiar with APIs, it's essentially a way to connect different tools. And I'm going to use tools generically, but tools being your, your Jira or your ERP or any SaaS solution that you might have. You know, you'll hear Jensen Wang, which is the CEO of Nvidia, Mr. Mr. AI, if you will. He's talking about like AI is not going to replace everything. There's. It's a tool, should use it to connect other tools so it can do stuff. And so what a plugin is, is in the Claude definition is a package that they have or others you can create your own. But there's a package of all the things that you would need to have Claude talk to the world. And it's in a nice package, it's easy to install. It provides you all of the hooks and APIs and all of the things that you could possibly want to do all these skills and tasks. Okay, so I'll pause there and we can jump into that here a little bit since I'm in the GitHub.
- [05:22]
A
Yeah, let's talk us through what we're Seeing on the screen.
- [05:26]
B
Okay, so this is the anthropic GitHub. So for those that don't know what GitHub is, GitHub is essential. Where it's a, where all your developer information lives. Some of it's open to the public, some of it's closed. Every company has a GitHub for all their developers. Anthropic has made all of their plugins public, so it's essentially open source. You can, you should really use this as a starter kit. All of these over here on the left are all the different plugins that they've created. So there's like a, you know, a plugin for sale, your sales team, marketing, legal, we're going to talk about finance, customer support. And within each of these you can drill down into, you know, there's like an instruction, like a read me like what is this? It explains all the stuff but more importantly there's the underlying code that you can augment. So what I'm going to do is I'm just going to start at the readme which is where you should always start like read the instructions. Sure that you do that with all your kids toys and why this is important. And so this is, this has been out, these plugins for Cowork came out in January so just last month. And these plugins are about a month old, give or take. And why I think it's important these are not like, these are all in beta. So these are not quote unquote production. They even, you know, important. This plugin assists but does not provide, you know, you got all the disclaimers, right. What, what Anthropic really did was they gave everyone that uses Claude a starter kit to customize the their needs to their environment. Okay. There is nothing in any of these plugins that you or I couldn't have created on our own and made public. Like they, if you are a cloud user you could, there's a marketplace of plugins and you can submit, you can create all these things and submit it and have other people download it.
- [07:44]
A
It's a farmer's market. You create your own stuff, you bring it, people can come use, buy.
- [07:49]
B
Absolutely. So there is, there is nothing that they have done in any of these things plugins and we'll drill down into it specifically. But there's no, we could, this could have been you or I, but Anthropic did it. And so this is going to touch on the first thing why I think this is important. And I, I, I was, I've, I was so wrong A year ago when I saw OpenAI so different, you know, Open AI is a different provider, but I saw Open AI Operator, which was a browser agent and that totally shifted my perspective on AI. That was really the, that was like, oh, this is, this is big and this is real. And essentially what that is what they call operator. It's actually agent now is essentially just. It was chad GPT in your browser and it could you instruct it to do these things. You could on your behalf, click, download, go to places, etc.
- [08:45]
A
First use case was go book me a haircut.
- [08:47]
B
That's why I come back to that. Exactly, exactly, exactly. And that was a year ago where we've come in a year. I, I mean I knew it was going to go fast. And I always tell people, whatever, whatever crazy thing you think is going to happen in a, you know, in whatever time period is probably going to happen quicker. The funny part is when I saw this, I glanced because they talked about doing financial analysis, building models, doing, you know, DCF things. And then I started drilling. I go, oh, journal entry prep, close management, audit support. Like what are these skills? And so what I think is this is you. Everybody should look at this as a signal. I think the big takeaway right now is this is, this is, this is big because one, it's a signal. We are, we are not only doing this for ourselves, but we are slowly going to provide this for the masses. So that's, that's number one why I think this is big. So I'm going to pause before I kind of jump into what all this stuff is that we're looking at. If you have any questions or thoughts,
- [10:02]
A
there's something I want to make sure that I want to ask you so that the people that are listening. Would you recommend people doing this on their work laptops?
- [10:15]
B
So I'm looking for a quick no
- [10:17]
A
right there because I don't trust enough people.
- [10:20]
B
So I, the reason I pause is because anthropics very like, they're known for security, right?
- [10:30]
A
So this is done like in a sandbox or virtual environment. That's where it's written to. But nonetheless, I'm going to advise Brian. You can take a different approach if
- [10:40]
B
you want to play around with it.
- [10:41]
A
All you have is a work laptop, spend 200 bucks, go get a laptop and do this on that.
- [10:47]
B
Yeah, I, I think, yeah, I agree. I think, I think you should, you should definitely play around with this
- [10:57]
A
and
- [10:58]
B
you should take all precautions. But, and this is where I, I would give you a counterargument. Maybe. Go ahead. The counterargument would be the. This is not the. What we're looking at, what we're going to talk about is, is it's Definitely not a SAS production, SoC2 covered, et cetera thing. But if you work for a company that uses Claude Enterprise or even ChatGPT Enterprise or whatever, you could do all this on your own. Like, going back to what I talked about earlier, all of these things that you see before you in this GitHub repository are things that you or I or anybody could have designed themselves. And it's really a starter kit. It is not a finish like, we'll get to this in a second down here. It's. It's basically telling you connect to the stuff if you really want to use this, like to its fullest extent. Yeah. And so, yes, I think, I think understanding what you're doing, cowork the product cowork that we're, that we're. This is a plugin for, and this is also a plugin for cloud code. But this is all done in a virtual machine, Linux virtual machine within the desktop application. This works. It started out on a Mac, it now is, is supported on Windows PCs. There is, there is definitely security built in through, through the, the Base Enterprise or even the Pro. You know, they have different subscription levels. Just be careful about how you, you know, what you grant access to, etc. I think that's what you're saying, but I would feel, I would personally feel comfortable downloading Claude code onto my work computer, but limiting its availability to get to different folders, different directories, et cetera. So definitely don't give it access to your entire directory, your network directory. But if you want to give it access to just your desktop and play around on your desktop, I don't think that that would be too out of bounds. But if you want to just get your own laptop and download it and do its away from everything, perfectly fine.
- [13:18]
A
I, I think I'm still gonna suggest that. I'm still suggesting that. Okay.
- [13:21]
B
I think, I think, I think the, the takeaway is, however you feel comfortable playing around with this solution, with what we're talking about, do that. I think, I think doing something is better than sitting and waiting, as some other people want to do. But. Okay. Okay, so real quick, I'm gonna try to go through this real quick and you guys can get to the. Well, hopefully we can share this link or you can just, just type in Claude code or Claude plugin finance and it should take you to the GitHub. What you're looking at is on the screen is slash commands skills. And then these kind of give you the examples and then kind of gives you some in MCP integration documentation skills are, are sort of what Claude's known for commands. These slash commands are basically like shortcuts. So like for us this is internal audit podcast. So I know like it's funny they have socks testing here as if like a public company is going to like utilize, which I think is like the skill is audit support. But they have slash socks testing, which I think is funny. So I'm gonna, so I'm gonna jump into skills because that's where the, the real important stuff is I think and just go to audit support. And so this markdown file, this dot MD file is, you know, like when you talk to your LLM and you type in a prompt. This is essentially what this is. It's a, it's an in, it's instructions to Claude to essentially use this, you know, use this skill. I don't want to be able to describe it, I don't want to describe the thing with the word, but it gives CLAUDE context, it gives CLAUDE instructions. You'll often hear like you can create your own skills based upon these things and other things. So essentially what Anthropic did was they said look, we're going to go ahead and we're just going to build out, we're going to build out these built in skills for Claude for finance and accounting. And you know, if you just read the Description, it's it SOX 404 compliance with control testing methodology, sample selection documentation, standards used when generating testing work papers, selecting samples, classifying control deficiencies or preparing for internal or external audits. And it just, it is a very long, I think very well written instruction document to give context to claude. And what happens is, is you know, when you buy your laptop and you download CLAUDE desktop because this is, you can also use this in CLAUDE code, but that's through like a terminal and they have made it super user friendly to just use the desktop application. And you, and you want to, let's just say you want to do like sample selections or have Claude help you determine like a risk assessment or whatever. They've literally just Anthropic has basically said here's context. This, you know, you're, you're asking a question about audit. It's going to have the skill, these, this information, this context to actually know what you're talking about. And so it's basically taking the vast LLM knowledge base and then shrinking it down and saying use this stuff, use this context. Why? So one, I think this is important because it's a signal. It's basically saying hey everybody, we're, we're, we're not, we're, we're getting more specific in our, in our product offerings. And it's not just anthropic. OpenAI is going to do this. They, they released their Frontier which is their enterprise architecture I literally just read about today. They're, they're partnering with all the big consulting firms to help companies integrate. Right? Google Micro, they're all going to be, now they're just copying each other. So if Anthropic releases this, I have no doubt others will release similar things too. Which brings me to my second reason why this is important. This. The they are clearly making, they're clearly making access to the non programming world. The majority of us who don't sit in no engineering development, they're clearly making things more accessible. And for the accountants, the auditors, finance people of the world who don't feel comfortable using a terminal, they are basically saying hey, we're, we're going to give you guys the starter kit and you can even use our tool to help you set it up. But it's going to be available out of the box, ready to go. And I would not be shocked within a year that all of this stuff is, not only is it production ready, but it will be SOC 1, 2, 3, whatever, secured. And so now you know, the big push the last two years has been a lot of VC money is coming to fintech. There's a, you can probably name probably a handful of companies that are building a gentic AI whatever. Like this is a signal. Be prepared because I can just sign up for Claude Max, which is a highly, highly, highly valuable like the amount of token costs that go into all these things is astronomical, which is a whole separate podcast. But like all this stuff is going to be available to the masses and it's, you know, it's user friendly and I think that's why this is a big deal.
- [19:15]
A
Thank you for listening and be sure to follow the link to green skies analytics.com in the show notes and schedule time to see how Green Skies can make the hype of AI a reality in your internal audit department.