Everyday AI Podcast – An AI and ChatGPT Podcast - EP 449: Can Claude’s AI Agent Simplify Your Work? A Live Test Drive

Episode Date: January 29, 2025

Wondering if Claude's latest agentic AI is worth it? Computer Use is an agentic AI system that allows you to operate a virtual computer simply by speaking with Claude. We dive in and explain how it works. Newsletter: Sign up for our free daily newsletterMore on this Episode: Episode PageJoin the discussion: Ask Jordan questions on Claude AIUpcoming Episodes: Check out the upcoming Everyday AI Livestream lineupWebsite: YourEverydayAI.comEmail The Show: [e-mail] with Jordan on LinkedInTopics Covered in This Episode:1. Overview of Anthropic Claude2. How to Use Claude Computer Use3. Critiques of Anthropic's Tools4. Future of AI AgentsTimestamps:00:00 AI agents essential in businesses by 2025.04:48 Google developing AI agent 'Jarvis'; competition intensifies.10:01 Using an API key; GitHub shares code.11:22 Docker is a versatile containerization tool for developers.15:36 Claude Sonnet 3.5 limits commands despite plans.17:08 Replace placeholder with copied API key.23:17 Demonstrating computer vision on a virtual desktop.25:33 Claude retained information without website visit.29:31 Experiencing repeated errors toggling between applications.30:49 Visit everydayai.com, list latest 3 episodes.35:10 Word document created with AI episode summaries.37:12 Direct AI with simple code; needs improvement.Keywords:Jordan Wilson, Claude AI, language model, Everyday AI Podcast, podcast summaries, document formatting, model interaction, AI errors, AI execution challenges, API key, Docker usage, virtual desktop, Word document creation, live stream, Anthropic updates, Claude free plan, API key security, Docker installation, Service tier levels, GitHub repositories, AI in Business, Claude's updates, Google Project Jarvis, OpenAI, Microsoft, Salesforce Agent Force, Amazon Bedrock, Google Cloud's Vertex AI, AI agents, Application Programming Interfaces.Send Everyday AI and Jordan a text message. (We can't reply back unless you leave contact info) Start Here ▶️Not sure where to start when it comes to AI? Start with our Start Here Series. You can listen to the first drop -- Episode 691 -- or get free access to our Inner Cricle community and all episodes: StartHereSeries.com Also, here's a link to the entire series on a Spotify playlist. 

This is the Everyday AI Show, the everyday podcast where we simplify AI and bring its power to your fingertips. Listen daily for practical advice to boost your career, business, and everyday life. Meet Firefly AI Assistant, now live in Adobe Firefly, the all-in-one creative AI studio. Just describe what you want to create and the assistant handles the rest, orchestrating multi-step workflows across Photoshop, Premiere Express, and more in one conversational interface. You direct the outcome. The assistant accelerates execution. Whether you know it or not, you're going to be using AI agents in your business in

2025. Let me repeat that. In 2025, you will be using AI agents at your company, especially if you live in the U.S. Whether you know it or not. All right. So you might as well get ahead, understand what's going on in the space, and stick around for the next 20-ish minutes and you're going to see live. And hey, if you're following

along of the live stream, you can even go ahead and use an AI agent. So we're going to be talking today about how Claude's agent works. It's computer use. And like I said, give me about 20 minutes and we'll go ahead and do it together live. All right, I'm excited for this. If you're new here, thanks for tuning in. My name is Jordan Wilson and welcome to Everyday AI. We're a daily live stream. podcast and free daily newsletter, helping everyday people leverage generative AI and how you can actually use it to grow your company and your career. So maybe you're listening on the podcast.

Thank you for tuning in. Please, if you haven't already, go to your everyday AI.com and sign up for the free daily newsletter for our live stream audience. Thank you for tuning in as well. Technically debuting this live, but it is pre-recorded. So if you want the daily AI news, don't worry, it's going to be in the newsletter today. Also, if you are listening to on the podcast, this is going to be a little bit of a more visual episode. So you might want to listen to this.

If you normally listen in your car at the gym, walking your dog, whatever it is, thank you, number one. But this might be one where, hey, go check out today's newsletter and go watch this if you do want to kind of learn along because I'm going to break down how to use this new computer use tool from Anthropic Claw. I know some people, you know, were asking for this when we did cover this news last week. So, hey, if you ask for it, we're going to do it. All right. So without further ado, let's get straight into it. So last week, Anthropic Claude announced a couple of things. They announced a 3.5 sonnet update.

So it's kind of weird. They already had a 3.5 sonnet. I kind of just call it 3.6 or, you know, maybe we refer to the old sonnet as 3.4. But anyways, they released an update to Claude 3.5 sonnet, which is out. and then they announced 3.5 haiku, which should be out any day. No word yet on what their big model opus, which is still on version 3. If that's going to get any love or any updates, we will see. But probably from Anthropics announcement,

what got everyone talking, including us here on the Everyday AI show, was its new computer use tool. All right. And if you do want more on this, today's is going to be a little bit more of a demo, right? So if you want more on what was announced and what our takes on it, what our takes on it were, make sure to go check out episode 386. It'll be in your show notes as well. All right. So here's what this new computer use is. Well, according to Anthropic, they said it is a new capability in public beta. So it is available now. It's called computer use,

and you are going to use your API key. Don't worry. You don't need to be a developer. even though it says developers can direct claw to use computers the way people do. Anyone can, right? Everyday AI, it's for everyday non-technical people. So I'm going to give you the walkthrough. Don't worry. One thing that Anthropic calls questions, and I couldn't agree more, they're saying it is still experimental at times cumbersome and error prone.

The thing that is most cumbersome is you can barely use it. Yes. If anyone at Anthropic is listening and I always love it, you know, I had some people from Open AI, reach out to me recently, said everyone's, everyone here is listening to the show. Thanks. But hey, Anthropic, you should listen to because, you know, one thing I always do here is I literally have trained thousands of business leaders in the U.S. on different AI tools. And Claude is very difficult.

It's, it's the limits, the usage rates, even in the API are so limited unless you are on a higher tier. So Anthropic really needs to get their program together. If I'm being honest, otherwise they are going to get smoked by everyone else because it is so hard to even use their tool and experiment and to see if it's going to work for your business. All right. Anyways, let's just jump a little bit into the landscape. So not only do we have the computer use tool from Anthropic, but we just got news that Google is reportedly

developing its own version of this, of its computer using agent called Project Jarvis. We'll see if that name actually makes it to production, mainly because. because some other company had tried to use Jarvis in an AI product before. We had been, we've actually been using it now for, I don't know, three and a half, four years since it came out. And they had to change their name to Jasper. So we'll see if Google can actually, you know, come out with this project Jarvis. But they're not the only ones with a computer use tool.

It's been rumored now for probably 10 months that Open AI is working on agents. We saw Microsoft should be out literally any day with its, with its new co-pilot studio, agentic AI, as well as Salesforce with its agent force. So that's why I said at the beginning of the show, you're going to be using AI agents, whether you want to or not. Like I said, it's coming to Microsoft Windows, all right, inside co-pilot studio. It's coming. If you are a company that uses Salesforce, which is the most dominant CRM player around,

you're probably going to be using agent force in 2025, right? So currently it's available. So one thing about Anthropic, even though I don't think it's that good. I don't think computer use is that good. It is extremely limited. It is hard to try out. But hey, they shipped it.

So hats off to Anthropic. They shipped, right? They didn't launch a blog post and wait list, which sometimes Google and OpenAI kind of do Microsoft a little bit as well. So they shipped it maybe too early, but I guess it's better than not shipping. But right now, you can only use it via an API. So if you're using Anthropics API, or you can access it inside of Amazon Bedrock or Google Cloud's

Vertex AI platform. So, yeah, this isn't something where you log on to Cloud AI or download a desktop application from Claude. And it's not how it works. So we're going to do it live. You have to actually download a separate program, you know, if you actually want to take the simplest route. And this is what Anthropic recommends.

All right, a couple of things. Let's first go over some terminology, because when I'm doing this live, I'm probably not going to be giving you definition. So I said, all right, especially if you're listening on the podcast, let me break all of these definitions down that I'm going to be talking about. So first of all, Anthropics, API. All right. So if you don't know in API, it's an application programming interface.

right, but in all of the large language models, you know, you essentially have the option to either use it on the front end, right? So you can go to chat gpt com or clod.aI and chat with the chat bot, right? Or you can use it on the back end. And this is generally what developers will do, all right, when, you know, if you're fine-tuning a model, bringing, you know, your company's data in with rag, if you're building third-party applications, you know, for the most part, you're using either open AIs, API, Klauds, API, Google Gemini's, etc. So all these programs, you can go get an API key and build an application. So that's kind of what we're going to be doing here. So the only thing you need to know about this is you can still use Claude's free plan. So even if you're not on their paid, I forget if it's $20 or $30, I think it's $20 a month.

So even if you're not on their paid plan, you can still use their API. But it is a pay as you go. So you will pay for actual usage. All right. So the other thing when we're talking about API keys, you know, I'm going to show you mine on the screen. And then when I'm done, I'm going to delete it. So you always want to keep your API key kind of secret because if anyone gets it, they can essentially use it in their programs and run up, you know, run up a bill on you

until you notice it. All right. So it's when you are doing this, you do need to have a credit card inside of the Claude API. And you need to preload some money in there. if you want to do this and follow along. Don't worry, I'm probably going to blab on for another three minutes before we do this live. So if you want to, go ahead, do it now. And you can log into the back end of the Claude system, which I'm going to go ahead and

announce that. It's council. Dot anthropic.com. All right. And here's the bad part. It is so limited. All right. So I would say most people, you're going to be on tier one.

So essentially, Claude gives you a tier, depending on how. how much you use their product. And I'd say even people that I know in the AI space, even they're all on tier one. So unless your company has already been using Anthropics API for a very long time, or if you're an individual, you're probably going to have to start on tier one, which is extremely limited. You can barely do anything in this computer use unless you're on a higher tier.

All right. So that's terminology number one is using your Anthropic API key. Terminology number two, GitHub in a GitHub repo. So GitHub is essentially a website where developers, but even everyday people, can go ahead and store and share the code that they write in projects that they work on, then other people can go kind of download them and fork them or edit and add to them. So a GitHub repository or a repo is kind of like a folder with a bunch of code in it. And it holds all of the files and information related to a specific project.

And the reason why this is important, and we're going to be doing this all live as well, Anthropic released computer use in a GitHub repo. So everyone can go on there and they can kind of build off it. But that's how you use it. So like I said, you don't access this via a website, right? You don't go to claw. com. You have to actually grab the information from Anthropics GitHub repo.

All right. So again, think of that as a place where if you don't already know where everyone kind of stores their code, and you can go download it. People can modify it, not there, but they can create versions of it, right? So think of it like templates, right? So people put their code up there. You can go look at the code, you know, improve it.

You know, if you don't know GitHub, it's a great place. And then last but not least, Docker. So Docker is a program. It's one of the ways that Anthropic recommends that you use the computer use program. So Docker is essentially a tool. that helps developers package their applications and everything they need to run them into a small, portable container that can work anywhere.

So kind of the way I like to describe Docker is it's a closed environment where you can essentially run programs. So it's kind of similar to terminal. Right. If you use a terminal on Mac, it kind of has its own terminal. It just helps you run everything in a contained way. And it can work from anywhere that you download it. All right.

So enough chit-chat, y'all. Let's do this live. I said we would try to do this in 20-ish minutes. Let's see if that actually works. All right. So this is going to, this is going to be a fun one here, y'all. So let's go ahead. I'm going to share my screen.

Let's see if we can get this. Let's see if we can get this going. All right. So now if you're listening on the podcast, I'm going to try to do my best to walk you through exactly what I just told you. So first, we are. are now on Anthropics GitHub repo. Okay. So essentially, like I said, there's a lot of different files in here. And there's different ways. There's different ways as well that you can use this.

Okay. So the way that we're going to do it is if you scroll down here, it's going to give you some directions. And it essentially gives you this little piece of code. Okay. So I'm going to copy and paste this code. I'm actually putting up in my browser. So it's all going to be flat. And then I'm going to, oh, let me share that tab. There we go. All right. So now I have it pasted in here. Okay. I have a Word doc. So all I did is I went on Anthropics GitHub repo. I went down here. It gives you the code essentially that you need to run. Again, I'm simplifying this. And so what we're going to do next is there is a placeholder where you're going to need to enter. your own API key.

And then we're going to combine those two things, and then we're going to put them in Docker and run it. Okay? So let's look at Docker here quick. You need to go to Docker.com, and you need to download Docker for your desktop. So I am on, I've already downloaded this, but you're going to go ahead and download this, whether you're on Mac Intel or a Mac Apple. So I'm on Mac Apple.

chip. I think my computer has like an M1 or M2. So you essentially need to download it for your operating system as well as what chip architecture you run on, right? So whether you're on Intel or an M chip for Mac and then on Windows, whether you're on an AMD or an arm chip, and then Linux. All right. So you're going to download Docker and then install it. All right, I already did that step. I think everyone out there, if you're listening to this podcast, you've gone to a website before, you've downloaded a program and installed it. So very simple. All right.

Step one, we copied and pasted that code from Anthropic. All right. Step two is we've downloaded the Docker desktop and installed it and opened it as well. All right. So now, let's see, I'm going to have to share my whole screen here in a second. So I'm also on my computer, FYI, you won't see this live stream audience. but I've opened Docker. And then when we're ready, when we're ready to get Docker going, I will share my whole screen.

So we'll be jumping around a lot. All right. And then our last step is we need to create, well, not our last step, but our last ingredients, so to speak, is we need to get that API key. Okay. So here's the thing when I was talking about limits. So you have different limits. So it says that you have a 50 request. per minute limit if you are on the tier one plan.

I'll let you guys be the judge. I would say that's not true. All right, because it's very hard to run any commands, even though I am on a tier one plan. And it says for this new Claude Sonnet, 3.5, the 1022 version. Yes, I wish they just called it 3.6.

So we didn't have to call it Claude 3.5, 1022. But it does say you get a 50 request per minute. I don't think that's actually the case. Or who knows, maybe it just takes so many. tokens because you're technically using computer vision every step of the way. So that is actually probably how many requests, but to do this simplest thing, you'll see we're going to time out a lot. All right.

So you can check your limits, but you can go into API keys and we're going to create an API key. Okay. So like I said, after you use this API key or you're not going to use mine. All right. So I'm going to go in. I'm going to copy and paste my. API key and I'm going to be deleting it right after this so no one can run up my bill, right?

So first you need to give it a name. So I'm going to call it, I'm just going to call it test dash computer dash use. Okay. Then you need to select a workspace. I'm going to put it in my default workspace and I'm going to click add. All right. And then from there, I have an API key. All right.

So I'm going to go ahead and copy that API key. And then I am going back to this document. Okay. So now here's what I'm going. doing. I know there's probably on my screen a lot of things going on. I always like to do this just to make it a little simpler. All right. So in my original API key, there is essentially a placeholder, okay, where it says API key equals, and then it says, you know, dollar sign, Anthropic underscore API underscore key. Okay. So I'm copying now my API key that I just used,

and I am going to place my cursor and I'm going to start with the dollar sign. So it says API key equals dollar sign. And I'm going to highlight through key, all right, just the Y and key, not an extra space or else you're going to run into some issues. And then all I'm going to do while that's highlighted is paste my key in. All right. And then from there, it should hopefully be pretty simple, right? I did one test on this before.

Sometimes it's a little buggy. So now all I'm going to do is I'm going to copy this. All right. So essentially now I have this command that I'm going to put into Docker that I got from Anthropics Quick Start guide on their GitHub repo. I got my API key. I copy and paste it. And then I pasted my API key into this Docker command that we're going to then go ahead and put into the Docker program. All right. So I hope that makes sense. So now we're going to get a little wild here because now I'm going to share my whole screen. And hopefully this won't be, hopefully this won't be too wild. All right, let's go ahead and share my whole entire screen here.

All right, let me close, let me close some other programs so we're not, so we're not too distracted here. All right, we should be good. I'm going to share my whole screen and let's get into it. All right. So this is the Docker desktop program. So like I said, you, you've download this, you install it, you launch it. All right. Now, here's, it's kind of hidden. Remember how I said this is kind of like the terminal program? So you're going to want to click the terminal at the bottom. And it says a terminal directly within Docker desktop. All right. So from there, you might the first time you run it when you click that little terminal, you will probably get a button the first time. I believe it says like an able terminal or

something like that. So I can't re-replicate that. But there will be one little button. there that essentially says, you know, it able terminal. All right. So now I have essentially what looks like a normal terminal. All right. And I'm going to zoom this up a little bit. So hopefully everyone can see it. All right.

So I have a normal terminal here. Now all I do, I don't have to do anything else. Remember, I copied that combination. So of the, from the GitHub repo and the API key where I inserted mine. And I'm going to paste it. And I'm going to hit enter. And it should take just a minute. just a minute, just a minute to run, right? And so you'll see it says starting. So it's essentially

at the bottom on my screen. I can see it's kind of running through. I might actually run into an error here. Again, it's very, very buggy. So, okay, so it looks like it looks like it worked. So yeah, sometimes if it doesn't work, just try it again. But what you're looking for, you might run this and be like, okay, well, what happened? Okay, well, there's just a little link at the bottom. It essentially says open and then it gives you a local host. All right. And all that is, think of it is this way. It is essentially a local version of a website that is technically running through something else.

All right. So in this case, we are technically running a local website through all this code that we just put into Docker. All right. So now I'm going to click this. It's probably going to open in a separate window and I'm going to have to drag it onto my screen. Don't worry. All right. So let me go ahead and click that.

it did open into a different window. Now I'm dragging it over. All right, there we go. So it is working. All right. Now let me explain what we actually have here. So we are on this local host. And it says Claude use, computer use demo. It says security alert, never provide access to sensitive accounts or data as malicious web content can hijack Claude's behavior. All right. And then you can do chat or you can look at the exchange logs. All we're going to do is we're just going to chat. All right, I'm going to move this a little bit. So we're not blocking the screen. And so hopefully we can see as much as possible.

All right. So essentially, on the left side, we're going to be talking to this clawed computer use. And then on the right side, there is a virtual desktop. All right. This looks straight out of 1990, maybe 1995, if we're being nice. So for our podcast audience, it's a split screen. I'm going to be able to talk to a version of Claude in this computer use demo. And then on the right side, if everything works, it is going to execute things in a virtual environment.

All right. So here's what we're going to do. I'm going to go ahead and paste this in. So here's what I'm saying. I'm saying, please find the largest American companies by market capitalization. Save the top three in a spreadsheet. Include their rank name symbol. I should probably spell symbol correctly, even though.

Claude will probably understand it, right? Because again, I am still talking to Claude, the large language model. That's very smart and can understand human language, right? And that's the key here, y'all. You are still getting the power of Claude, but you are just adding to it the ability to use a computer, right? So it's using a digital computer. All right.

So I'm saying, please find the largest American companies by market capitalization, save the top three in a spreadsheet, include their rank name symbol market cap, and then their CEO. So this part might be a little tricky, so we'll see how it handles it. And then I'm saying, and add that in there as well. I should probably say add that in the spreadsheet as well, all right. So I'm going to say add that in the spreadsheet as well. All right.

So on the virtual desktop, there is absolutely nothing. All right. So I'm going to put my hands in the air once I do this because, you know, I'm sure our live stream audience or maybe if you're watching this later, on YouTube, you might not believe it or understand it. All right. So I'm going to click, go. And then on the left side, how it actually works is it takes a bunch of screenshots.

It uses computer vision and then it maps out what it wants to do on this virtual desktop. All right. So I'm going to go ahead and click enter. We're probably going to get a bunch of errors, but here we go. Adobe just introduced an entirely new way to create, bringing the power and precision of its creative suite into one conversational experience. Meet Firefly AI assistant, now live in the Adobe Firefly app. the all-in-one creative AI studio.

Powered by Adobe's creative agent, Firefly AI assistant lets you start with your vision, just describe what you want, and shape the outcome as it takes form with the assistant. The assistant orchestrates multi-step workflows, drawing on 60 plus pro-grade tools across Adobe Creative Cloud apps, including Photoshop, Illustrator Premiere, Lightroom Express, and more to help bring your ideas to life.

You can also get started with creative skills, a growing library of pre-built workflows for common creative tasks, like batch editing photos, creating mood boards, portrait retouching, and creating social variations. Every step the assistant takes is visible so you can refine, redirect, or take over at any time. You stay in the driver's seat as the creative director. Adobe Firefly AI assistant now in public beta. See it today at firefly.adobie.com.

All right. So for our podcast audience, it's saying, okay, I'm going to, going to click on Firefox and search for this information. So it brought up Firefox and now let's see. So on the left side, I'm seeing each and every time it's screenshoting something. So now it's searching and it says largest companies by market cap 2024. So it brought up a Google search result. So we're going to see if it's going to try to grab this information because it kind of brought in an AI overview. So I don't know if it's going to go in there or go to a website. So it looks like it's not even going to go into a website.

It doesn't look like it. All right. So let's see. It's still running. All right. So interesting. So now I got a new error that I haven't seen before. It said warning.

All right. So now I'm just scrolling up to see, but it's still doing everything by itself. It said warning failed to launch Java LDX. Java may not function correctly. All right. But it looks like if you see at the top here, it's still running. And it's doing some things to counteract this. All right.

So I can still move my mouse, but I'm not taking over anything on the screen. So it's already typing, right? So it says rank company name, symbol, market cap, CEO. Okay. So again, podcast audience, my hands are folded on screen. And now Claude is doing all of these things in the computer use tool. All right. So I'm going to scroll down here because I'm guessing we're going to run into an air pretty quickly.

All right. So again, when it Googled these things, it didn't even click into a website. It grabbed information from the AI overview, right? So, you know, sometimes you put in some information and you don't even have to go to a website. So Claude didn't even need to click on the website, visit it. It essentially retained this information. I'm actually going to scroll up here. I want to see how it's recorded it. Okay. So it looks like right here. It did just scrape all of this information. I'm not sure if that's where it was. Yeah.

So it looks like it did not even go into the information and it grabbed the top three U.S. companies by market cap. All right. And you'll see, I'm pretty sure. Let's see. Okay. I didn't run into an air yet. It's still just a little slow.

I would have assumed that I would have already hit a token limit here because let's see how many screenshots it did. one, two, three, four, five, six, seven, eight, nine, ten, eleven, twelve, thirteen. Yeah, so it already did way more screenshots than it did previously in my other testing. And it looks like it just stopped for whatever reason. Again, Anthropics said, this is very buggy. As you can see, it's buggy. All right.

So now it's going again. So let's see. It said, all right, here's why. Because that original information did not have the CEO. All it had was the company name, the symbol, their rank and their market. kit cap. All right. So this is funny. It kind of failed here. So it said, let me search for Apple's CEO in Firefox. All right. It didn't bring it up on my screen. All right. And then it said, let's see. It looks like you just put in the CEO column on the spreadsheet. And it's using

Libre Office, which is essentially an open source free version of Microsoft Office. So it's, it got some things wrong here. It didn't, didn't do it very well, because now it's saying, again, let me switch to Firefox and search for the current CEOs of these companies. So it didn't, it didn't do it very well here. I'm scrolling down here. It looks like there is an error. All right. So now I finally hit the rate limit. So let's look at the, let's look at the spreadsheet here. So there is something when you run this. There's a toggle screen button in the upper right hand corner.

So it says toggle screen control off. So I can click it and now it is on. So now I can actually go in to the actual spreadsheet, right? So I'm clicking on an empty cell and you'll see I can type in here if I want. I just type the word type. So it looks like this failed because instead of actually finding this information, it just typed in who is Apple CEO 2024. instead of typing that in to Firefox,

it actually just typed that into the spreadsheet. So it ran into an error there and did not complete this. All right. And then you'll see it says essentially retry after a minute and 27 seconds. All right. So in theory, I'm talking right now because I'm biding my time. Sometimes you don't even have to wait that long. I'm going to try it again.

And I'm just going to say I'm going to toggle the screen control off. actually I need to toggle it back on and get off that, get off that, that cell I was on. And I'm going to say, please continue. So I'm not going to guide it. Please continue. All right, we'll give it a second. A lot of times it takes a second for your message to show up. So now it's saying running agent.

All right. So let's see if it can pick up and see where it went wrong. Those little sounds you may not hear. that happens essentially each message. I might just have to mute this so it doesn't thug me. There we go. All right. So not doing the best here.

So again, we're just running into some repeated errors. It looks like for whatever reason, it's actually struggling to go back to Firefox. Right. So it looks like it's struggling to toggle between the two. At least I'm not seeing anything on the left hand side. it's showing me all of these, all of these screenshots that it's taking. And it's really just every time it updates something in the, in the spreadsheet, it's just taking a photo, right?

So didn't do a good job here. I'm going to go ahead and say it failed this task. All right. So I'm going to toggle screen control on. I'm going to go ahead and exit out of this document here. All right. I'm going to go ahead and kind of clear this. All right. And we're going to give it a minute.

We're going to give it one more, one more kind of one more run here. I'm going to clear the cache. So I'm essentially clearing everything, right? I'm going to go ahead also and reload this host. There we go. All right. So I have a blank, a blank screen here. So now all I'm saying is I'm saying, please,

go to the website, Your EverydayAI.com with the HGGPS. You have to include that. And I'm going to say and find the latest episodes. Instead of saying create a spreadsheet because I just did that and it was struggling. So I'm going to say create a word doc and write basic info for the last five episodes in the doc. Include the episode number, title, and a description. And, and a description. And and I'm going to say, I'm going to say to do this for three, because five, we're probably going to write into a bunch of errors. And I'm going to give a command that you would maybe give a large language model. And I'm going to say, please write a witty intro that will catch people's eyes. All right.

So this is the last demo that we're going to do. So now I'm clicking this. Again, we're starting from a blank doc here. If everything works right, computer use is going to click. on Firefox. There we go. Presumably, now it's going to go to your everyday AI.com. In each step, it's sending a screenshot, right?

So it knows what to do, and then it gives it the coordinates. All right, it looks like it's having a problem rendering something on our website. It made an emoji like size a trillion. But that shouldn't keep it from using this. So let's see. So it ran into an issue. It said it couldn't open a directory file. Finally, it got it.

All right. So it didn't also, I didn't see it go to the episodes page. Let's see. It looks like it only went to the homepage. So let's see if it realizes that. So now it says, let me scroll through the episode page to gather the information about the latest episodes. All right. But again, for whatever reason, when I did a,

a demo of this earlier, right? Live demos are the worst. It did find kind of toggling between spreadsheets and Word documents and Firefox. For whatever reason, this time around, it looks like it is struggling, right? So it is still running here at the top. So I just have to give it a second. Hey, live stream audience, I know this one is going on a little bit, but let me know. Are you going to try this?

Are you going to use it? Or have you seen enough? And you're like, ah, this doesn't really look that good. It looks buggy. But I will tell you this. I know that the anthropic team can ship, right? Really quickly. So, all right, now, now let's see what it's doing.

So now it is in, back into the document here. It wrote some quick recaps. I'm going to see it here in a second. It looks like it went to the save dialogue. So it's going to save this. It saved it as everyday AI episodes. Technically, it kept the untitled information in there. Let's see if this is finishing it.

Let's see. Okay. So it didn't tell me that it finished yet, but it did what I asked it to, right? It didn't format it great, but it, you know, it says, let's see. It says your everyday AI serves up the freshest, most digestible AI insights that'll make you sound like a tech guru at your next coffee break, right? So this is all content that Claude wrote. This isn't from our website.

So it went to our website, looked at the homepage, looked at the episode page. So then it wrote a quick recap for the three. So it says episode three, you know, and I did tell it, you know, I think I said, hey, be witty. Let's see, what did I say? I said, what did I say to the large language model? I said, please write a witty intro that will catch people's eyes. All right. So it said episode 388, the duality of AI productivity, the fascinating exploration of how AI

can both enhance and challenge our traditional notions of productivity. Discover the sweet spot between AI assistance and human creativity. All right. So it looks like it did. Did the job there. Fine. Wrote a quick recap. And then it told me here, essentially, yo, I finish.

It said, I have created a word document with a catchy intro and information about the latest three episodes from your Everyday AI. The episode has been saved as, and then it gave me the name, EverydayAI episodes. ODT. In your home directory, the document includes, then it tells me a witty introduction, information about the three latest episodes, and brief descriptions. And then it says, would you like me to make changes to the document or would you like to see it in a different format?

So I'm going to try one more thing. I'm going to say, please format the document and add paragraph. breaks between the descriptions, and I'm going to say add more engaging content. All right, I'm going to leave that one open-ended. I may or may not even let this one finish because I know this video is dragging on a little bit. I haven't even tried this yet to see how well it does at modifying documents that you may have it make.

So again, podcast audience, I am just typing with Claude in real time. Okay, it highlighted all of that content. And it's essentially rewriting it. It looks like it's failing again. So instead of a paragraph break, it just added, it looks like a little symbol that would, in theory, denote where you would want to have a paragraph break. But it did write a lot more content. Looks like a little bit more engaging as well. So nothing here that is going to, you know,

nothing's ready for production. Let me just say that. We're going to wrap this one up. I'm going to keep an eye on it on my screen here as we wrap. But this is nothing that right now that it's going to change the way that we all work. But it's laying the groundwork, right? The fact that this technology is available right now. And I just walked you through it.

Yeah, it took me longer than 20 minutes. You should know that now. Right. So, but you can go through and follow this. It doesn't take long. Like I said, I'm going to leave, I'm going to leave that kind of little piece of code both in the episode description as well as if you are listening here on LinkedIn. That information should be there as well.

But with very little developer savvy, right, little bit of copy and pasting, you can literally direct an AI agent, right? It's not good right now. Don't get me wrong. I'll say it's downright mediocre, but it works, right? It's buggy, but it works. Right now, anyone out there with a computer and a credit card, it doesn't cost a lot, right? I'll have to go check my usage.

Anyone can go do this, and you can have a language model, a very capable one in Claude 3.6. We'll just call it 3.6 Sonnet. Come on, Anthropic. All right. So with 3.5, Sonnet new, fine. You can go have it use a computer. And this is not the endpoint, right? This demo, this virtual machine, this is just to showcase the capabilities.

This is not the end goal, right? This is just wait because any day now, we are going to see developers create fully functioning, robust, polished tools. That's what this is all about. It's giving businesses, developers, third-party software providers access to this technology, right? This clunky demo is just a clunky demo, right? This isn't the end use, but y'all, AI agents are coming. I can't even say they are coming. They are here.

I literally just walked and talked to you through it on a live stream. And you don't need to have a technical background. You don't need to be a coder. all you need to do is to be able to type to a large language model. It's pretty exciting. All right. I hope this was helpful. If so, please go to your everyday AI.com.

Sign up for the free daily newsletter. Also, tell me, what else do you want to see? What else do you want to hear? I did this because a lot of y'all, after we covered this last week, they said, hey, Jordan, I know it might be a little more technical, but do, you know, do a demo, show us how to use computer use. So here you go. You want it.

You get it. What do you want to hear next? Thank you for tuning in. We hope to see you back tomorrow and every day for more everyday AI. Thanks y'all. Meet Firefly AI Assistant. Now live in Adobe Firefly, the Allman One Creative AI Studio. Just describe what you want to create in your own words and the assistant handles the rest,

orchestrating multi-step workflows across Adobe Creative Cloud apps, including Photoshop, Premiere Express, and more in one conversational interface. You direct the outcome while the assistant accelerates execution. Stand control with the ability to step in and refine at any. any time. See it today at firefly.adobie.com. And that's a wrap for today's edition of Everyday AI. Thanks for joining us. If you enjoyed this episode, please subscribe and leave us a rating. It helps keep us going. For a little more AI magic, visit Your EverydayAI.com and sign up to our daily newsletter

so you don't get left behind. Go break some barriers and we'll see you next time.