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. If you listen to this show at all, especially in the last week or so, you'll know that I'm sometimes kind of hard on Anthropics Claude.
I think in the past 15 to 18 months, it's gone from probably a top tier large language model to almost a forgotten AI tool that's now almost exclusively for developers. In the future, it seems of what it's going to be used for the most is for software engineers, coders, people like that, not the everyday business leader. But,
That's kind of changed a little bit in the past just 24 hours. And there is one new feature in Anthropix, Claude, base pay plan, the $20 a month plan that is already available
going to, I think, change how I work and it will change how you work as well. If you choose to use it and spend the next 30 or so minutes with me. All right. I'm excited to talk about the one new cloud feature that changes knowledge work and how to use it today on Everyday AI.
What's going on, y'all? My name is Jordan Wilson, and I'm the host of Everyday AI. This thing, it's for you. If you're new here, we do this every single day, a live stream, podcast, and free daily newsletter, helping us all not just learn what's happening in AI, but how we can leverage it to get ahead to grow our companies in our careers.
Is that you? That is literally what you're trying to do, right? Yeah, me too. Well, you're in the right place. It starts here with the unedited, unscripted live stream and podcast, but you need to actually make it all happen in our newsletter. That's where you need to go. So go to youreverydayai.com.
Sign up for the free daily newsletter. Each and every day we recap our episode as well as keeping you up to date with literally everything else that you need to be to be the smartest person in AI at your company. All right. So most days we start with the AI news. Not doing that today. That's going to be in the newsletter. So make sure you go check that out. All right. Also, live stream audience, I'm thinking about trying something new. All right. It's Wednesday, Friday.
I'm thinking about this new segment called AI Work on Wednesdays or something like that. All right. I've heard from a lot of you all over the years that you love live demos. You like to kind of get your hands dirty, you know, and see how to actually use ChatGPT, Gemini, Claude, Copilot, you know, maybe some of the visual tools or creative tools.
I don't always like doing that because actually, I think our biggest audience is on the podcast. Well, by far, thanks to your all support, we're usually a top 10 tech podcast on Spotify out of like 90,000. And sometimes I think these things work a little bit better for the live stream and on the video. So podcast audience, you know,
These might be ones that, hey, on Wednesdays, make sure you go to our website. On our website, you can watch the video. So yes, there's a live stream here. I'm going to be sharing my screen. But hey, live stream audience, let me know. Should we do this AI work on Wednesdays thing? Should this become a regular segment? I'll ask you again at the end. So just say yes or no. Or maybe you want to see how it goes. You can say yes or no at the end. All right. So let's talk about that one feature. The new feature is AI.
the combination of a deep research mode inside Anthropix Cloud and the new and expanding integrations. So it's kind of new, but kind of not because Anthropix actually released this about a month ago. However, just in the last 24 hours, they've upgraded it significantly. But the biggest upgrade is the availability. Previously, you had to be on their super expensive platform
you know, max plan that starts at a hundred up to $200 a month in order to get this research and a lot of the integrations. However, now Anthropic has just rolled this out to their base pro plan. So if you are on the, you know, $20 a month, uh, Claude pro plan, like I am, um, then you can go take advantage of this right away.
Actually, Claude is the only one I'm not on the top tier, right? I'm on the $250 a month Gemini plan. I'm on the $200 a month Chad GPD plan. With Claude, I've never been able to stomach anything more than the $20 a month pro plan, mainly because of the limits.
So I, which is funny because I'm going to do a live demo here and I've intentionally haven't used Claude the last couple of hours because I know if I do more than like three prompts, I'm going to run out of messages. So yeah, especially when you're working with longer context windows and having Claude do complex things, that $20 a month plan, it's,
pretty useless. The paid plan, I hit my rate limits usually in about seven minutes when I'm really pushing it. So I'm not going to push it intentionally too far. That's another thing with doing these. Hey, if I do this, you know, AI work on Wednesdays, if I'm doing anything in Claude, there's a good chance it's going to just stop. Right. And I don't edit this thing. So yeah, we'll see how it goes. But that is the new thing. It is the combination of the deep research. It's
Technically just called research, but a lot of people called the deep research mode on Anthropix Claude.
uh with the integrations so i'm gonna explain a little bit more what those are but uh the first of all it's just the research right so if you've used or if you've uh if you've used any of the deep research services from open ai uh from uh gemini i think those are like one a and one b probably gemini is a little ahead of open ai's deep research but it's really neck and neck and then you know grok has their version perplexity has their version uh microsoft copilot has their version as well so
know everyone's really getting into this deep research game and one of the reasons why well number one it does a ton of research for you which is great but in doing that that also reduces the rate of hallucinations exponentially so yes depending on which uh deep research tool you're using it might take anywhere from you know two to 45 minutes so
It's not something you get instantaneously, but these new deep research tools are highly capable. And then when you can combine them with both things like tool use, being able to, as an example, query the internet when it deems that it needs to, as well as working with direct integrations, which is what this new cloud rollout has. Those are huge, huge possibilities. So like I said,
The big news here, aside from it rolling out to the pro plan, is this combines existing and new app integrations inside Claude with the deep research mode. And in this case, it can go anywhere from five to 45 minutes. And this unlocks contextual business knowledge from apps and the web. And this transforms how teams can discover, analyze, and act accordingly.
on information. And the other cool thing here is Anthropic is also rolling out an MCP mode that can work in their deep research mode. So what that is, Anthropic has created and obviously popularized the model context protocol. Don't worry, we're going to do probably sometime in June, we'll do a show on the model context protocol, but essentially in the same way that, you know, SAS,
products or the internet you know the internet can talk to each other with apis right that's how internet's uh internet websites and uh different sas uh you know softwares talk to each other
AI and large language models can't use APIs. So they needed essentially a language, a universal language to talk to each other. And essentially Anthropic invested in it early, popularized it. Now other companies are coming up with their own as well. You know, you have A2A from Google, but MCP is probably the biggest, the most popular and the most widely supported. Even companies like Google, like OpenAI are supporting
Anthropix MCP. So that's huge when you can also, what that means is you can do this deep research and hopefully I'll be able to explain this to you as we do it live and connect to really any data via going the MCP route. So I'm going to give an example working with some of Claude's direct integrations. All right. Also, if this is helpful at the end,
go ahead and repost this. And I have put together a document on the seven best use cases for research integrations. All right. So combining these two, I spent some time, I put together some great use cases. So just go repost this. If you're listening on the podcast, I always leave a link to the LinkedIn show. So if you want those, just go ahead and repost this and I'll send them to you. And if I ever forget, just bug me, Mike,
my LinkedIn DMs are nuts. Just, you know, reach out, you know, just poke me, right? It's like Facebook. It's like, Hey, I shared this, send this to me. All right. All right. So let
Let's learn live. What could go wrong? What could go wrong doing something that's kind of advanced, absolutely live, especially when we're working with rate limits, which is never fun. Yeah, the cloud rate limits are absolutely terrible. So live stream audience, let me know when you can subscribe.
Let's see my screen here. Let's get the correct one up. Hopefully y'all can see it here. All right. Yes, and also I'm not my normal setup, right? So I'm on just my laptop normally. I have my big screen, so I'm gonna be fumbling around. I'm actually in Atlanta. So I have a really cool partnership going on that I'll share with you guys a little bit more later this week.
All right, so let's see. All right, let's just do this. All right, so now, podcast audience, I have opened up Claude, claude.ai. So I'm working with Claude on the front end. Yes, everyone always reminds me, hey, Jordan, you always poo-poo on Claude.
you know, for not being great on the front end and you can use it on the back end. Yes, you can. It's terribly expensive compared to Gemini 2.5 Pro or Gemini 2.5 Flash. You know, so when it comes to capabilities and also keep in mind Gemini 2.5 Pro, if you read our newsletter,
Logan Kilpatrick confirmed they're coming out with a new version. So yeah, even if you're big into software development or anything like that, hey, Claude gives me these couple of points of advantage that's going to be gone pretty soon the next time Google updates Gemini 2.5. All right. Anyways, I am now sharing my screen. I have a super long prompt. So I
already have the results of these prompts that I've ran, but I'm actually making it a little more difficult. So I will share and try to walk through for our podcast audience exactly what I was prompting, the inputs and the outputs. But I'm going to start this one live. Then I'm going to jump back into my slides that I usually have most days here on the podcast. And then we're going to check in at the end.
All right, but here's what I'm doing. And if you are on the pro plan, you're going to see this new button here that says research and it's in beta. So anytime you want that, you're going to, if you want to do the research and again, depending on the complexity of the query and depending on the integration type that you select,
This could take up to 45 minutes. So in my test, it took about 12 minutes. That's why I couldn't do like five of them live. All right. So you're going to want to click on research.
Also, you can choose the different models. I'm going to go Sonnet 4 for this, not Opus 4. I think Opus will take too long. And I also think the marginal gain that you get from using Cloud Opus 4, which is the large version versus the medium version of Cloud 4, I don't think it's worth it in terms of like waiting...
for the time. However, I am going to enable the extended thinking mode inside Cloud Sonnet 4 to force Cloud to kind of think step-by-step, think logically, plan ahead, et cetera. All right. So right now, here's the other thing that will be new. All right. So if you click the toggle, the search and tools option here inside Cloud, and then if you scroll to the bottom, there's a new section here that says add integrations.
Okay, that's going to bring up the integrations kind of section here within Claude. And you'll see there's some direct ones here, such as Google Drive, Gmail, Google Calendar that have been there for a while, GitHub.
And then you can also click this Git integration. And then they have a guide. There's the pre-built integrations, which include things like Asana, Intercom, PayPal, Square, Zapier. And Zapier is the huge one. But also you can create your own just via MCPs. So more on that later in this show. So let's go back.
So now that you know what I'm doing, I have the research button toggled. I have extended thinking on and I do have drive search enabled. And that part's important here. All right. Yes, I have a very long prompt here. I'm going to paste in and then I'm going to tell you essentially what this prompt is doing.
All right. It's actually super small. So let me go ahead and open up a blank Google Doc and hopefully y'all can see it this way.
All right. So here's essentially what I'm trying to do. And this is a real use case. All right. And I always encourage people like don't rely on other people's use cases, other people's benchmarks. You need to be developing your own internally. So this is kind of an advanced one that I've kind of developed to test both agentic information.
agentic tool use in a hybrid model, as well as deep research that is integrated with documents. So I know that sounds niche, but I would assume pretty soon this is going to be available to do in all major providers, Chad, GBT, Gemini, Claude, Copilot, et cetera. So you need to be building out these use cases, right? Don't
Don't just blindly use a tool. You need to be testing them, right? That's what I'm doing. That's what I'm doing here live. All right. So what I'm essentially doing
is I'm telling Claude to look in a specific folder in my Google Drive. All right. And then I'm explaining what's in this document. All right. So, uh, essentially I created this folder in Google Drive not too long ago, uh, probably only like six or seven months. So, uh, after each newsletter, um, I go through and I write the newsletter, right? Uh,
I write it. I'm a human. I use AI to help me, obviously. So I upload the transcript, you know, uh, the AI system, we use cast magic, other things, you know, pulls out some insights, you know, pulls out some ideas, but you know, I'm the one that actually goes through and type it, uh, types it. I type it. Uh, I love doing it. Uh,
I'm a writer. I was a journalist before. It's one way that for me, because I use so much AI, I love still just writing manually as much as I can. But I obviously have AI get me the first 20% pull out ideas because I do a lot of podcasts, right? And a lot of times I forget things. Okay. So what I'm saying is go look at all of these documents. So
Each day I write this leverage portion for the newsletter. So if you read our newsletter and if you don't read it, why aren't you reading it? You should. Like I said, it's humans, right? But at the very bottom, there's this section we call leverage. So there's a little intro and then there's three main points in a section called try this. So it's actionable tips based on, you know, the guest or if it's my, you know, if I'm the just like today's show, right? I'm going to write it up after this.
So I'm telling it there's all of these documents and here's how they're named. They're in this folder. So I'm explaining what these are. So I'm essentially saying, go look at all these documents. There's a couple of hundred. All right. And then I'm telling it, hey, you should also match up this leverage section with the blog post that it corresponds with on our website. Here's the trick. And here's where I'm testing this model.
Are you still running in circles trying to figure out how to actually grow your business with AI? Maybe your company has been tinkering with large language models for a year or more, but can't really get traction to find ROI on Gen AI. Hey, this is Jordan Wilson, host of this very podcast.
Companies like Adobe, Microsoft, and NVIDIA have partnered with us because they trust our expertise in educating the masses around generative AI to get ahead. And some of the most innovative companies in the country hire us to help with their AI strategy and to train hundreds of their employees on how to use Gen AI. So whether you're looking for chat GPT training for thousands,
or just need help building your front-end AI strategy, you can partner with us too, just like some of the biggest companies in the world do. Go to youreverydayai.com slash partner to get in contact with our team, or you can just click on the partner section of our website. We'll help you stop running in those AI circles and help get your team ahead and build a straight path to ROI on Gen AI.
it's there. There's no link in there. I probably should have done the link right all along. But, you know, depending on when I write it, you know, maybe the web post isn't up yet. Right. So each and every episode that we do here on everyday AI, there's a webpage for it. We put the transcript on there. We put some, some key points. We put the YouTube videos. You can go watch. We, we embed the actual podcast episodes. You can, you know, multimedia experience. So I'm essentially saying, yo, here, here,
is the, uh, the leverage section, but you're going to have to go find the corresponding blog post. All right. So, uh,
go find it. And then this task is actually pretty hard. So here's the three things I'm asking it to do based on all of these files inside of my Google Drive and then, you know, testing its tool usability to go in, find information on the web because Claude, as of a couple of months ago, finally got access to browse the web. So I'm saying,
Step one, I want to come up with 10 mashup ideas. All right. So what that means in a simple example, and this is what I kind of told Claude, I'm like, yo, I've talked to dozens of people about AI agents.
I should be able to create three niche and very different mashup episodes on agents, right? You know, and as an example, I can pull a couple of quotes from different people and combine them. So instead of me talking to one person generally about agents, you know, I can have, you know, four or five
people kind of contributing in one episode from past episodes, right? So it's almost like a highlight reel, but creating multiple of them because I've probably had 10 hours of episodes on agents. So it's like, I should be able to create a good show, right? But also it requires more research in between. You know, it would require me to, you know, have intros and segues between all these pieces, right? So essentially, first I'm telling it to go build 10 episodes
mash up episode ideas and outlines. All right. And I'm giving it, you know, very specific things to do. Then I'm telling it to do, do the actual episode layout. So what that means is I'm saying like, yo, go write the episode.
Right. So not only, you know, step one is like write an outline. Right. Give me ideas. Here's the people that could be on it. But step two, I'm saying go look at step one and then actually put it together. Find the actual quotes, put the quotes together. And then also for me, for Jordan, you need to do additional research.
make sure it's up to date, you know, do an intro segue. So, you know, step one, it's helping it for step two. And then for a step three, I'm saying find 10 new episodes site idea. So I'm saying essentially based on all of this, right? So you're going, you're looking at our last six months of episodes. You're doing additional research on your own. Go find things we aren't
covering and we haven't covered and then suggest 10 episode ideas. But not just that, you know, do research on those ideas, outline them like we outlined them in step one, and then also find five potential guests who could be great guests for those shows. All right. So that was a long explanation.
All right, so now let's jump in and let's check in on Claude. All right, so you'll see here, you can always watch. So essentially it's going through all of these sources live. All right, I'm going to zoom in a little bit. Hopefully our live stream audience can see this.
All right, so I can click on this as I wait. So, so far it's been doing this deep research across these different sources for six and a half minutes and it's already gone to 230 plus sources. So it's looked at
81 different sources inside my Google Drive, including one search inside of my Google Drive and then 147 other searches. And I can always go in here and I can kind of look and see exactly the documents that it's looking at.
Looking at, it looks like there was a couple errors. So I don't know what those errors are. It could have been a permissions. So I probably should have changed the permissions at the folder level, which I thought I did, but maybe I didn't. All right. So it's going through all of these different files that I told it to. And then it's also doing additional research. Also, like I told it to.
on a lot of third-party websites. So I'm scrolling here. There should be a lot from our website, youreverydayai.com. I don't see that yet because, okay, here we go. Finally got to it, but not a ton here because I told it to specifically, okay, here we go.
All right, so it looks like it did go to at least maybe 40 or 50 pages, maybe 50 or 60 pages on our website. And you'll see here for our live stream audience at least, it's going back and forth, right? This is the power of large language models that have agentic tool use. That means it's starting as an example, it started in the Google Drive folder, right? And then it did some research on third-party sites, okay? Multiple rounds, right?
Then, let's see, let's scroll down here. Then it went to your everyday AI, okay? Then it went back into Google Drive, probably to confirm some things, right? So this is the difference. And to tell you the truth, this is kind of the difference between like a CPU and a GPU. And this is why I've been losing my noodle for the past couple of months
When it comes to agentic tool use, I don't think people understand how big that is, right? If you're a little bit of a dork, think of the difference between a CPU and a GPU, right? Which is essentially a CPU is a little slower. It's how computers run, right? And it runs one process at a time. It finishes it and runs another process. A GPU can run parallel processes all at once. So it is infinitely more powerful. And I think the same can be said about this big step up
from going from a, you know, kind of an AI chatbot to an AI chatbot that has agentic tool use, right? It is absolutely wild, but also an agentic chatbot that has, or sorry,
a chatbot that has agentic tool use and is powered by a reasoning model, right? That's the other thing. This isn't powered by a quote unquote old school transformer model that is essentially a next token prediction, right? Even though it is, but this is one that thinks and plans ahead like a human would.
All right. So let's go ahead. We're going to check on this at the end, but let me go ahead and finish my little presentation and give you the rest of the details. All right. Stick around. I'm going to go through these quickly. All right.
So before we check in, let's now talk about some of the details. I covered a little bit of this, but how can you access this right now? Well, you have to be on a paid Anthropic plan. Luckily, you don't have to be on the max plan, $100, $200 plan a month anymore, or on a team plan, which unfortunately for Claude, I think requires five users. The other thing, web search, right? You already saw it like a couple of months ago. It didn't have web search. It does now.
There are going to be some upcoming integrations, including Stripe, GitLab, and Box to expand some possibilities. Also, keep in mind, ChadGBT is going to be rolling this out in the same way very soon. They've already started to roll it out. More on the team plan, unfortunately, but we have seen some rumors and rants. If you listen to our Monday show, this is going to be rolling out to other plants as well. So,
Why does this matter? Well, think of, think of what I just described. Remember, I just said, oh, it went and looked at all these documents. Then it started doing third-party research. Then it went to your everyday AI.com. Then it went back to Google docs and it went through, you know, 83 documents. This is a project I need done, right? I could do it. I would probably be the best person to do it. Cause I I'm like, I did all these shows, right?
I cannot tell you how long this would take me to accurately do. Is Quad going to do it at an A grade? Probably not, right? But for me to even do a C job on this would take hundreds of hours.
This is a Herculean task. I mean, we're technically talking about thousands of pages, right? Because those web pages on our site are thousands of words, right? Because it includes the entire transcript and I'm telling it, we'll see how it does. I'm telling Claude to go in and literally pull the best quotes verbatim. That is an extremely time-consuming task. So again, think of your use case. I'm showing you mine. This is, you know,
AI work on Wednesdays or whatever we'll call this, right? But think of how long it takes to find that relevant information across different systems. It is super slow and inefficient. And to be able to have to go and gather that data manually, really, number one, it's exhausting for humans or teams of humans to do that, to manually go out and fetch and retrieve all this information, to summarize it, to paraphrase it, to pull key insights, to just do that, it empties everything.
a tank, right? But then to have to actually create something new out of all that information, it is nearly impossible. This task is,
I've been wanting to do it for years, like since I started, right? I've been doing everyday AI for two and a half years. I had this idea a year and a half ago. I'm like, oh, I should start doing this. It's been on my to-do list for 18 months. And I didn't want to do it because either it was going to take me an ungodly amount of hours or I would have to probably pay a person or a company tens of thousands of dollars to do it. Well, we'll see. Can Claude do it?
Right. And in general, this is the future of work teams that are using this for faster and more comprehensive understanding of their current knowledge and their strategic decision making.
So right now, here's what it does. It can research the web, your Google workspace, including calendar, Gmail, and Drive, which is huge. And any of those connected app data. So I'm going to run through those again at the end. And then Claude says it can take up to 45 minutes. But like I said, depending on what you're having it do and how good your prompting is, never do it once, always iterate, right? But this could end up saving you dozens or hundreds of hours, right, per day.
Each time you use it. All right. So let's talk a little bit about some of the apps that you can integrate with. So I did go over these a little bit previously, but you can do it with Jira, Confluence, Zapier, Asana, Intercom, right? And there's more coming soon. I'm especially excited about the MCP and the Zapier integration because that opens it up to essentially the entire internet.
And the biggest thing here is it gains live context. You know, that's the downside with working with things inside Claude's projects or Chad GPT's projects or Chad GPT's custom GPTs or Gemini's gems, which don't work the best right now, even with Google's own products. The problem is, is so many times you're working with stuff.
static data and you're not always and you're just working not in this deep research mode that has this agentic deep tool use all right so an example is well you could use zapier to auto post auto pull sales data or prep briefs from whatever you know crm or erp or whatever alphabet soup uh you know piece of software you use that's important for your company and here's the thing i'm excited about
pre-built integrations using remote mcp so this was just updated uh today so paid users can now connect to tools uh via the uh integration urls with the remote mcp feature uh so i did kind of show you that on my screen earlier uh but it's very simple uh you don't even have to be super technical uh right the first time i did this i'm not a super technical person right so if you aren't either don't worry
If I'm being honest, what I did is I pulled in all the information from Claude's website. I pasted it into Claude and I said, walk me through this. Right. And it was pretty, pretty simple. Right. But now you have the custom and third party integrations that are also supported via MCP. So let's talk about a couple of use cases. All right. But like I said,
Go repost this. I'll share seven that I think are great, but a couple. On the legal side, you can automate due diligence, case research, and document assembly. Finance, you can accelerate private equity deal screening via automated multi-source data. That one's huge. Support, you can identify feedback patterns and autofile bugs, right? Using, you know, in Intercom as an example or using something in Zapier or a third-party MCP. I think the use cases are plenty.
So how should you use this? Don't worry. We're going to check in on our on our prompts if it's done right after this. But how should you use this and what are the next steps? Well, first, you need to identify the tasks right now that need deep research. Right. And these sometimes are things that you're already doing. So first, I use I identify the ones that are already sucking up your time and your team probably hates it.
All right. And then also find the ones that you wish you had time to do, but can't start with the ones that you're already doing though. Right. You need to win back your time with generative AI, then connect the critical apps, right? Whatever that is, uh, to Claude, I would, you know, obviously start with, uh, Gmail, uh, Google drive and Google calendar. If you are a Google workspace organization, uh,
Zapier for sure. Cause then you can connect to more than 6,000 or might have to be up to like 7,000 apps now. And then pilot, whatever it is, right? Don't just go in there and start playing. You need to talk to your, talk to your team, talk to your people, share this episode with them first, and then sit down and say, what are we going to pilot?
I think one of the biggest problems, and I'm not going to go off on a five-minute riff here, I promise. One of the biggest problems with AI is people just run and go with no destination. Okay? Before you start, before you even go and check, oh, do I have access to this? Well, you should. But before you do that, first...
Listen to this episode again. Okay. Share this episode with your team. Then sit down and say, what are we going to pilot? What is our end goal? All right. Get a finish line before you start running because then you'll know when to stop running and then you'll be able to measure it and then you'll be able to scale it. All right. So let's check in, shall we? Hopefully this worked. Let's see. All right.
It's done. Yay. Okay. So in total, it went to, it didn't go to any more sources from when I checked in at that, whatever it was, six and a half minutes. So it ended up doing 229 total sources, 11 minutes and 29 seconds. And y'all,
Please, please, please always go and click. So it's this little box. I wish all the companies do a bad job at this. Like the UI is not good. Most people don't even know that you can go and click that.
Right. It's this little box. Right. With a little arrow. It's so thin. Go click that because you essentially can read. It's not the actual chain of thought, unfortunately, but at least shows the process that in this case, Claude went through. Because, again, I always encourage people look at the results, look at the steps, update your prompt, do it again. All right. So let's look at the results, though.
And let's see how well it actually did. It did these things. All right. And live stream audience, as I'm going, let me know. Number one, do these episodes sound good? Number two, give it, give it a grade. Give it a grade. All right. All right. Here we go. So first it gave me a summary. All right. Let's see. Oh, is this the right one? Yes, it did. Okay. So.
What's interesting is it said, I've analyzed Jordan's Everyday AI newsletter archive, also has cataloged 500 podcast episodes, which I didn't even ask it to do, right? I just asked it to find the corresponding podcast episodes for
for the leverage newsletter documents. So there's probably about 80 of those. So it went ahead and it did a little more work, which is great. And it also researched 2025 trends. So it found, so it bullet pointed some of the, some of its findings, dropped some guest database. Okay, so here we go. So step one, here's the 10 mashup episode ideas and outlines.
Okay, so the first one, it says AI agents corporate identity crisis. And it says I should combine a recent episode I did with Sarah Bird from Microsoft and Ron Green from Kung Fu AI. And it gave me three episode title options. Pretty good.
And then it gave me an angle. It gave me a micro script. And this micro script is garbage. Sorry. People love to think like Claude is some great content writer. It's extremely average. And then here we go. Give me the guest list. All right. So for the outline, it seems like it did a pretty good job. Let's keep going. So number two, it says the great AI coding takeover. It didn't combine anything.
It just combined topics that I've covered. That's okay. That was allowed. And then on guest list, it's just kind of combining some of my recent episodes. So not bad. Number three, AI research revolution from garbage to gold.
Let's see. Same thing. Give me the title options. Guest list. Okay, here we go. Give me just one. Give me one guest. And then it gave me two of my own episodes. So if I were to run this again, I would give more strict rules. I would say either...
You know, it's a kind of a Jordan only episode or it needs to have at least three guests. Right. So, you know, unfortunately, these aren't the best, but that was probably due to my prompt. Not any anything from Claude, but hopefully we get a couple more here. So number four, the job security deathmatch.
Again, just one guest. So yeah, unfortunately, this run, which is interesting because I ran this one earlier and I think it actually did a better job, but that's the generative part of generative AI. All right, so, so far, eh.
You know, I wish I would have run this with a little more specificity in the guest list with some requirements. And then we would have seen some hopefully better results. However, you know, overall, I think it did do a decent enough job. I would probably give it like a C plus or B minus on this. It did everything. Actually, I'll say I'll say a B because probably the things that I'm not liking were things that I could have done better. All right. Step two. Here's the difficult thing.
Can it build a legit script and weave in actual quotes? And I did say you need to include at least 12 total quotes. I should have designated that these need to be actual non-fabricated quotes. All right, so let's see if it did. So yeah, I believe it made it up. So what it did is it took...
my writing, uh, from the, uh, from the newsletter and it attributed that to a quote to the guest. So again, not good, uh, because yeah, not good. All right. So, um, again, probably my fault though, right? I would run this again and probably, probably do a little bit better. All right. So unfortunately this first, uh, or this most recent go round did not do great.
Let me do let me just go ahead and share this this previous one. So I did run the exact same one right before. I thought it did a little bit better. Let me just go ahead and scroll through to our kind of step two and see if it actually did better. Or maybe I was just a little more impressed the first time around. Yeah.
Okay. So yes, on this one, it actually did it correctly. So in this one, you know, it says here's an episode for AI agents everywhere, the $100 billion workforce revolution. So it wrote a script for me. So what if I told you, all right, here's the intro. Let me know. Does this sound good?
Like some, I would say, what if I told you about the end of 2025, your company could have digital workers with their own corporate IDs. Yeah. Avatars closing deals at 25% of the cost of human sales teams in autonomous agents, inspecting hotel rooms better than six month veterans. That's not science fiction. That's today's business reality. I hate that intro. Um, sounds like AI doesn't sound like me, but maybe it's better than me. I don't know. Uh, okay.
Then it went in correctly. It started to write transitions and then it started to pull actual quotes from the episode pages that I originally wanted. So let me just see if this is correct. So this is Natalie, Natalie from episode 371. All right. So let's see if Claude made anything up. So it said she said, right. It's like,
So this is baby formula for me as a mom. Okay. So I'm going to just take that because it's a very specific thing that we would not normally talk about in a podcast. And I want to make sure that Claude is correctly pulling this information. So the good thing is, is it gave me the citation so I can go in here. Here's our episode page transcript. There's a lot of information on here. All right.
Bam. Look at that. All right. So the good thing is on the previous one that I ran, it actually did it correctly. And it's pulling actual quotes from multiple guests and it's piecing them together. So y'all generative AI is generative. And it looks like
Uh, it did a pretty good job. And then, uh, uh, it put in another quote here from a different one with Al Lagunas, um, a different quote there. So very good. So the first run on this did better. All right. Uh, but that's the, that's the downside of generative AI. Yeah. So, um,
And then let's go ahead and look at the third one, uh, as we wrap up today's show. So this is the 10 new episodes, uh, according to maybe gaps I've missed. So, uh, number one, it's the $300 billion AI infrastructure race, uh, who's building the future, uh,
Two, AI agents in healthcare, the $2 trillion opportunity. Three, the AI creativity paradox when machines make art. So yeah, pretty good stuff. I'm not going to go through and read all 10 of them, but it did go through. It gave me ideas. It gave me a little bit of background. It gave me a very rough episode outline. Not very good, but it gave me a rough one. And then it also gave me ideas for actual guests to use as well. Yeah.
uh, to, to reach out to. So I would have to see if these are actual people. So let's just see for number two, it says AI agents in healthcare, the $2 trillion opportunity. And it said, talk to Dr. Regina, uh, bars, bars, LA, uh, MIT AI healthcare. And it gave me her, uh, Twitter. I'm guessing this might be a hallucination. Uh, so let's see if it actually is or not.
All right. So I just Googled that. Does it look, so it looks like this is an actual person, but I don't think that's her actual, you know, Twitter, Twitter handle. Yeah, it doesn't look like it. So gave me actual information for a real person, but it made up their Twitter handle. So again, yeah,
It's not 100% accurate. I would obviously go through and improve some things before running this again. But overall, live stream audience, what do you think?
What do you think? All right. Don't just think of my haphazardly thrown together, right? This just came out less than 24 hours ago. I wanted to bring this to you right away, right? If I was doing an actual use case of this, I would go through at least 10 times and refine that prompt, make that process better, go through, see where it went wrong and make sure I get it right before I even ran it as a use case or a pilot. But
I'm pretty impressed, right? And the one that got right, not the one I did live, but the previous one, it actually did one of the hardest things and it did it correctly, which was going through more than 80 newsletter articles
pieces that I've written. It went and it found the corresponding on its own. I didn't tell it where it was. It went and found it on its own. It read through the transcript. It pulled out actual quotes from multiple different guests and it pieced them all together. That is dang impressive, y'all. That is really impressive.
right because then in theory i could probably figure out how to use something like you know descript they have their new ai uh editor i could probably somehow figure out to just have descript go put it all together right so again
This is not perfect, but probably with another two to three hours of refinement, I could probably make this work really, really well. And I wouldn't have it do 10. I would have it do three at a time. And I think just that would save me dozens of hours. This is huge. This is huge. Is it perfect? No. Is it changing how we should think about work? Abso-friggin-lutely, right? Y'all, I told you.
This has literally been in my click up like project list episode list for 18 months. And to tell you the truth,
It was getting overwhelming for me to even think about. But I know, man, I think about it. I'm like, yo, I'm sitting on so much great content. I've been so lucky to interview some of the smartest people in the world. And sometimes every once in a while, I'll replay an episode. But I keep thinking there's got to be a way to just bring in bits and pieces from different people and weave it all together.
And I'm like, I could go and do that, right? I'm a journalist. That's what I used to do. I would interview 20 people, you know, pull together their quotes, you know, get the glue together and hopefully put together a nice story that taught people something. It's time consuming. I know how time consuming this is. To do this at scale, to do one of those episodes would probably take me,
at least five hours because I got to go through and either listen to the episode, read the transcript or throw the transcript manually, right? Or let's just say you're not using AI at all. That to put one episode together would take me 10 hours. Using AI would probably take me three hours. Using a tool like this, a deep research tool with agentic tool use that connects and integrates live with your data is
Like I said, I refined it a little bit and then I have a good version within 20 minutes. Right. So the time savings on that is huge. So think I gave you my example on on AI work Wednesdays. What's your example? What are you going to do? I'd love to hear from you. Let me know. Also, hey, this new segment, just tried it out. Yes or no. Let me know. Yes or no. Does it stink?
Should we do it again? Yes. Should we do it again? No. On the podcast, let me know as well, because go subscribe to our newsletter and just, you can even just say yes or no. Just reply to that newsletter today. Say yes or no. I hope this was helpful. And if you do want seven of those best use cases, already put them together. Already put them together. You're going to want these y'all. I put so much work in on this show and our team does as well. Behind the scenes, making sure
that you stay ahead. Because literally, there's dozens of developments every day. You can spend hours trying to keep up. Or you can just let us do our job. But you got to pay the toll, right? This is free. I want to make sure everyone always has access to the best information. But
Hey, if you want to say thank you, go click that little repost button on LinkedIn. If you're listening on the podcast, like I said, check your show notes or check our newsletter. We always have the link to that LinkedIn post. Go repost that. And then probably within a couple of days, like I said, I'm traveling. So within a couple of days, I'm going to share that use case document with you. And if you don't have it in a couple of days, just feel free to bug me.
All right, I hope this was helpful. The one new Claude feature that changes knowledge, work, and how to use it. I gave it to you. We laid the blueprint out. This was fun, y'all. I liked it. I hope you did too. Thank you for tuning in. Please, if you haven't already, go to youreverydayai.com. Sign up for that free daily newsletter. And thanks for tuning in. I'll see you back tomorrow and every day for more Everyday AI. Thanks, y'all.
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 youreverydayai.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.