# New AI Drops: AI does your work, Claude in Slack, cheap AI & more

Channel: The Next New Thing
Video: https://www.youtube.com/watch?v=SMIuRWtbO9Y
Duration: 24 min
Language: English
Words: 5179
Transcript page: https://viewrankai.com/tools/youtube-transcript/SMIuRWtbO9Y

---

[0:00] Do you want a one-click way to give you and your team open claw? Well, I've got two different companies that launched a way for you to do that. Hate spending money on AI costs? Well, I've gotten a new thing from open router that will make it super easy for you to save money and get the best value and the best results. Ooh, and I've got an easy way for you to take any idea you have in one shot have it created and so much more launch this week. We've got links right here so you can click along and follow and of course everything will be in the

[0:23] description. Let's get into it. Presented by Zapier, the AI automation company. All right, Corey, the number one story of the week is record and replay in Codex. We've all seen this video. I'm going to show it again cuz I think it's so freaking impressive. Let's talk about this. He's basically saying he's going into his Codex app and he's saying record and replay then watch me upload this YouTube video so you can handle it for me in the future. And he's doing all the stuff that you and I have done. You get your title for him it's in a spreadsheet, his description from a spreadsheet. He's getting his image, he's putting it in,

[0:55] he's doing all the stuff that he does and just doing it naturally the way you ordinarily would. And at the end he gets these skills that he could use with Codex in the future and have Codex now that it understands how to do the work do the work. What do you think of this

[1:09] one? I think this is a long time coming. So, I'll tell you the way that I used to do this. I used to record myself doing a task with Loom. Then I take the Loom video, upload it to Gemini cuz Gemini is the only model that can watch video. Then I have Gemini give me the step-by-step checklist for what I just did. I take the checklist, I copy it into Claude and I have Claude turn it into a skill and then I've got to iterate on it two to five times to really perfect it. This literally takes that whole process and makes it a one-shot where it can just watch you the

[1:37] first time and build the skill once. I think this is awesome. One of the most impactful updates we've seen in a long time. Fan-freaking-tastic. By the way, I had no idea you were doing that. That's so smart. Okay, here's a cute little one. This was huge on Hacker News. It's basically a place where you can come and see how you're doing on all the different um I I the sound effects on all the

[1:57] different models. You type in your name. We're not going to wait here while it does it, but here's a result for your name. It's telling you that Oh, wait. It found the wrong person. It found the professional poker player, actually. I didn't notice that. American It's a thinks I'm an a professional American poker player known for his success in high-stakes online and live tournaments. So, [laughter] That means you really need to improve

[2:18] your AEO. Yeah, I guess. I guess I'm That's probably more on me than it is on this tool at the end of the day. Agreed. Um especially you're bigger than it's than it understands. This is me. Strength, top 3%. Here's where I'm showing up, and it does have me right. Andrew Warner, Mixergy founder and

[2:35] podcaster. It We need now the updated stuff for the next new thing to show up over here. Pretty interesting to see how it shows Oh, look, Claude Opus 48 says entrepreneur and podcaster who founded Mixergy where he interviews startup entrepreneurs about how they built their business. Fantastic. I like how it show up. Nice little cute tool. Of course, we'll have a link to that and everything

[2:53] else below. Let's go on to this X post. This is a quick one. Apparently, the betting markets are saying the Fable 5 is coming out. I think it is. What do you think? I think it is for sure. I think it's only a matter of time. I heard a lot of chatter, I guess, in the most recent

[3:08] Claude code update that was pushed. There is something in the source code or something the developers look at that, you know, mentions Fable 5 coming back, which is exactly why we see that spike in the prediction markets. I mean, 60-plus percent chance that it's coming back before July 1st. Yeah. people say it's coming back today. Today's June 25th. I've heard people say next week. I think 100% before July 1st, and I think it's going to come back bigger than ever. If you think people were freaking out about it the first time, well, when it once they got a taste and then it was taken away, now

[3:40] it's going to come back with a vengeance. I can't wait to use it. I can't wait to use it. All right. I also really like this from Anthropic tag. It basically puts Claude in Slack. Tag, get it? Like as you're tagging Claude by at mentioning it. And what it's doing is it's letting you and more importantly your team interact with it. And it's it's pretty freaking smart. Like here's an example of Marcus saying, "@Claude, check out error rate that jumped from 0.3% to 6% right after the 1,400 hour deploy. Can you dig in and figure out what broke?" It's on it. It has access to the tools it needs and it gets

[4:12] responsive. It can also read what's going on in our chat history and understand that. It can schedule tasks. It can do things proactively. Here's an example. Claude, every Monday at 9:00 post a pipeline update for the team. Pull the latest uh from Salesforce. Claude says, "Got it. I'll pull the pipeline from Salesforce and post a digest every Monday at 9:00

[4:31] a.m." And here it is Monday at 9:00 a.m. It's doing it. I think the upside is my team is now going to have a more unified experience with uh with Claude. Claude is going to have more information on us. The downside is I could imagine more freaking slop. I do not want other people's dopey reports that they're not even looking at showing up on an automated

[4:48] basis. What do you think of this tool? I also don't want a million notifications in Slack cuz I already get a million and one as it is. But So this is this is really interesting because this is Claude one trying to kill Victor, the AI startup that we just saw raise like 70 or 90 million dollars. Their whole Victor's whole business model was, "Hey, we give you an AI agent in Slack." That's literally what Claude is doing is trying to kill them specifically. And I think it's going to work because it's meeting it's meeting people where they already are. I think that's one of the biggest

[5:16] friction points why not everybody is currently using Claude is because them having to open up the Claude desktop app is not a learned behavior yet. But when you can use Claude inside the the tools that you're already using, aka Slack, it's going to get even more adoption. So I think this is going to be a hit. Uh we were actually going to build something similar. Now we might have to pivot a little bit, but all in all I think this is really good for consumers and and for

[5:39] Anthropic. Yeah, and and I think people are using Slack and ChatGPT, but they're not getting it to do real work. And if they see others on the team do real work, they'll understand it. By the way, this is available only, of course, in Claude Enterprise and team accounts, but really, for most people, that is the account that they need because if if you have a team, then you're going to want a team account. All right, let me talk quickly about Zapier MCP because I had a great experience with it this week. I tried to give one of my Hermes agents access to a Google account. I wanted it to have its own

[6:09] email, its own this, its own that, uh its own calendar, its own Google Drive account, and I didn't like it using my browser. I didn't want to give it full access, though, because I was a little worried about what it might do. Partially, actually, it's that I didn't want to give it access to my account in addition to its own account because I was worried what it might do to my account. I don't want it sending out messages, deleting, any of that. So, I went and um one of the first things I did was I went to Google's new tool, and I signed up uh to give it access. The thing took me a

[6:35] freaking hour, and I still couldn't get it to work, and I worked on it with the Googler who created it. Still couldn't get it to work properly. Then I said, "Oh, Zapier MCP." I went to zapier.com/mcp, and I quickly was able to just select the radio boxes or the check boxes for the tools that I wanted my agent to have, and then within that, I wanted to give it specific access to specific things. Its email, send, receive, do everything you want, go at it. My email, you can only read and draft messages, and that's the kind of power that you

[7:01] get with Zapier MCP. It is so easy. People do not realize it. I want you all to go check it out. zapier.com/mcp. Next, uh we talked about that. Let's talk about Blue Rails. This was number one on Product Hunt one day this week, and the reason it was number one is what it's doing is making your site more

[7:22] visible and more accessible to agents. So, agents can find it and recommend it, and they could read it, and they could also act on it. So, we all Well, you can talk about the importance of this. [snorts] Yeah, so I think this is a much-needed product. So, let's talk about kind of what it does specifically. So, you know how when you're using Claude or ChatGPT and you say, "Hey, go to this website and pull the information or summarize it or whatever." A lot of times it can't read the website with web fetch. It fails or the website's blocked or for whatever reason. So, it looks like what

[7:52] Blue Rails does is it fixes that. It makes your site accessible to agents, so agents can crawl your site, scrape it, interact with it, and eventually transact on it. So, it looks like that's it's a two-part service. So, part one is they make your site more accessible to agents, which is a huge need. Part two is they make they give your ability your I guess they give agents the ability to transact on your site. So, think about, you know, back in the early 2000s, every website started having a pay with PayPal button on the website. That was like the early days of e-commerce. Blue Rails is

[8:24] kind of trying to do that for agents, essentially adding the checkout button for AI agents to your website. So, I think this is going to be huge. I think they're 6 to 12 months too early, but they're clearly trying to claim market share before this really takes off. Uh and I think if, you know, if the service does what it they say it's going to do,

[8:41] I think it's going to do great. I think now is the time for hustler go-getters to make their sites more accessible to agents. If you take a look at Agent Mail, Agent Mail in many ways is similar to Gmail. It's just giving an email to your agent. But, with Gmail, I had to go and sign up for a Gmail account or add it to my network, and then I had to go to Zapier to connect it. With Agent Mail, I told my agent, "Just go and create an account." And it did, and it said, "Andrew, there's an email for you just con- to confirm you're the human behind this." I clicked

[9:11] that, and I was done and ready to go. And the only reason I used it was because of that. It's not a better product than Gmail. In fact, I'm messaging the founder to give him some feedback on it. But, I'm using it, other people are using it because of that. All right, next story. Uh I hope we're not rushing through these stories too fast, Corey. Let me ask everyone in the audience. Give us feedback. We're doing this. We're trying to improve. We want to get this right for you. So, tell us, do you want more time on each story? Do you want us to fly quicker? Give us feedback in the

[9:37] chat. All right. This is artifacts from Claude Code. At first when it came out I said, "This is not new. We've had artifacts in Claude forever." And then I realized, "No, when I'm chatting with Claude, if I'm using co-work with Claude, I can ask for a document and it will create the artifact on the right." When I'm in Claude Code, now I can get that kind of experience too. And you could see here on the left that you've got a Claude chat experience in Claude Code. And on the right, what you have is the artifact for Claude Code and it's shareable and it's updated in

[10:09] real time. Valuable? Definitely valuable. This is the one thing that Claude Code didn't have that Claude chat and co-work did, right? So now there's really no excuse not to be working inside of Claude Code for just about anything, especially if you're writing solo. If you're doing stuff with the team and things of that nature, it might be it might make more sense to still live inside co-work and chat, but I think this is like the missing piece for Claude Code that that really makes it the you know, the hands-down best tool in

[10:36] the Claude arsenal by far. I think Claude Code for many people feels scary, but if you just switch over into that tab and give it a stupid task, you'll see it's not really that scary. Um Not at all. You like this one. This was one of your suggestions from Tibo. Tibo for us. Yeah, so this is I you know, full disclosure, I'm pretty sure he owns this tool, so he's he's a little biased here, but I think this is one of the cooler demos of one of like that the I guess prompt-to-video tools that I've seen out there called Revid. So he generated this video with one prompt and

[11:13] it's pretty good. It's kind of like an infographic-style video. And we've seen I feel like we've seen AI AI video get so much better over the last 3 to 6 months. You know, starting from like the HeyGen and some of the other like the earlier video tools. HeyGen is still big, but this looks good. And Figma came out with something similar literally 23 hours ago. So, it seems like a lot of the AI [snorts] video tools are moving towards you can you put in a prompt, you get a pretty solid version one in

[11:44] seconds. I will say this, we're seeing more and more of these types of videos on X. What he did was he took Paul Graham's article here about how to earn and he he and he automated it. It's cute. I don't want to sit through this. I don't want to sit through the ones on X. I don't want to sit through the ones that come out from founders who are so excited about their beautiful looking founder video for their new product that has nothing but text and images. Why am I not reading it then? Why you What is this? Now I'm reading it at your slow pace. I think this whole style needs to

[12:15] go away or add some freaking text to it. Like it does it would it kill them to add 11 Labs voice? Let me see. I bet you he he has music. Let me see. No, it is a matter of simple exponential math. Oh, okay. And I we can do that. He's got No cheating required. Two simple numbers determine how Okay, all right. Now I've got something to listen to with a little visual. I take it back, Tibo. You're right. This

[12:38] is good. This [snorts] is a site. And he's he's done a good job of creating viral videos that are automated in the past. All right. Enough said from Andrew. I'm going to shut up now. This is something I'll just queue this up and then talk about the bigger picture here. Um He said, "This is your daily reminder that meeting transcripts are the best context you can get. Just dump the whole transcript. No summary. Plain JSON will

[13:03] do. I one-shotted this ROI calculator based on one meeting note." What's the big picture here? So, I think this is just a general best practice at all. Like what he's saying is like, "Look, there's so much context trapped inside the meetings that you're having with your team already. Why not just take that those transcripts, dump it into a Claude code or a Codex and have it build out the tools that you talk about in your meetings. Like this is something that in hindsight is a no-brainer, but you could connect Granola to Claude Code or Granola to Codex or whatever and just kind of run that as an automation. Say, "Hey, every

[13:39] time there's a new Granola transcript, if there's something that we talked about that would make sense as a web app or mobile application or anything, build it, right?" And at the very like if we never use it, who cares? It was practically free and very quick to build, but it's just a good exercise. Like I bet people if if somebody set set up that automation, which I think I'm going to go and do with my team, I'd be shocked to see all the cool little software uh apps and tools that we build out over

[14:05] the course of the year. Yeah, I I think that's fantastic. Work Claude, this is very similar to what came out from Anthropic Claude Tag, but they're basically saying you could have multiple agents, different types of agents. Maybe you have one agent in your Slack that is your writing agent for your team and it knows the way your whole team likes to write. And another one that is your social media agent and so on. And so these are Claude's like open Claude, get it? This was also number one. It was built on top of open open Claude. And it's a way of easily adding multiple agents to your team's

[14:47] Slack. So the fact that if you scroll up above where it's it's a one-click add to Slack and then a one-click you know, talks with other Claude's, check the box. Like that in my mind was the biggest friction point when open Claude first came out. One, setup was a freaking nightmare. The fact that this is one-click, like when open Claude first came out, that's something me and my co-founder tried to do for like 2 weeks. We tried to create a one-click where we didn't have to host it because we didn't want to be liable for anything that could happen. And we weren't able to solve it. A ton of

[15:17] one-click solutions came out, but they they taking on the risk of if something goes wrong or there's a breach, they were screwed. So, I'm assuming Work Claw has mitigated that in some way, but they made it to where it's a one-click install and a one-click to where it can talk to other agents. Like with a traditional Open Claw setup, for it to be able to talk to other Open Claw agents, there's so much nonsense you've got to do on the back end. It's one technical, two like you It's one of those things where if you don't know how to do it, you're going to spend two

[15:45] hours figuring it out. Looks like they made it pretty much plug-and-play. Like this is like you said, Claude Tag, but the ability to have multiple agents, they can be specialized. They live in not only Slack, which is where Claude Tag lives, but also Teams. And seems like pretty straightforward pricing and you know, not something that's going to

[16:05] break the bank. I think I'm going to set this up for our team. I've been trying to get Victor, which I really like, which also does this into our platform. It's It's not working well. Um it's not their fault. I was one of the early users of it and now we can't disconnect my previous account from that. Um my one concern with it is I don't like how everything, like the credits just keep piling up. I wish that they would

[16:29] use uh one of the other cheaper models. Um but I I'm still willing to try it because it's such a big need and I think my bigger issue is not that my team's going to overuse it, but that they're just not using it enough. So, let's get them up there and started. Okay, next. Uh what's stopping you from building a second brain? Talk about what Nick has

[16:49] written here and why it's so important. So, I think second brains are it's starting to gain in popularity. Uh it really blew up when Karpathy started talking about it, you know, a couple of months ago. But this whole concept of uh a knowledge base that contains all of the context about you personally and your business that is searchable and maintainable by AI agents and that doesn't lock you into one model ecosystem. So, if you have a second brain maintained properly, it doesn't matter if you're using Claude, it doesn't matter if you're using open AI or open source models, that knowledge base is the foundation of every AI

[17:25] deployment or project or initiative that you're ever going to do. So, having setting one up is critical. It's the biggest single biggest asset you can have in your company, hands down. It's not even close. That literally is your moat. And what Nick is showing us here is that he ran a /goal prompt inside

[17:43] Codex. And he gave away the prompt, too. This prompt is super robust. Like, it's incredibly long. I I mean, I'm shocked that he gave that away for free. It's incredibly valuable. It basically tells Codex, "Hey, go look at every single data point that you know about me and all of my connectors, everything that's connected to you, and build and maintain an Obsidian database." Like, that's literally the gist of the prompt. And he ran it as a goal, and he said it ran for about 12 or 13 hours. And at the end of it, he claims, you know, literally when that prompt is done running, you have

[18:13] personal AGI. Every single thing about your personal life, everything about your business, you know, thus far, is queryable. I mean, think about the insights that you could get from that. Be Imagine being able to query every, you know, call transcript you've had for the last 3 years and say, "Hey, what are my top three opportunity areas?" Or hey, what are what what's one thing I could do better on sales calls to increase my closing percentage? And imagine being able to do that across a team or your sales team or your product team. I mean, again, I I had a YouTube video that came out 45 minutes ago about how to build

[18:49] and sell second brains as a service. So, this is something that I swear to you is so it's going to be so big. That's why I'm fired up about it. This is huge. I don't I don't think people appreciate it. By the way, I should tell people that you could subscribe to Corey right now. It's right below this video because it's

[19:04] a collaboration between the two of us. If you just go over below the video, you'll see him, subscribe, and you'll get to see He's very practical with his advice and guidance on how to use AI and he's not just learning about it. He's using it, selling it and teaching people how to do it every single day. So,

[19:20] follow, subscribe. Next, this is very relevant to the World Cup. I wanted a Mac OS YouTube TV menu bar app so I could watch the World Cup game while working. I asked Codex one shot it with {slash} goal. We are we live in the golden age of ideas. I don't fully understand what he did here, but do you and then tell me what the bigger

[19:42] vision is? Yeah, so it's my understanding that again, he when he's watching YouTube TV, he's probably watching that in a browser tab. And he's essentially saying like, well man, I wish I could work inside, you know, another tab and have YouTube TV as kind of like a little box in the top right of my screen. So, I can be working inside of other tools or inside other tabs and still be able to watch the game. And he just went into Codex and said like, hey, can you build this

[20:08] and and set it as a {slash} goal prompt? And it did. It one shotted it. So, now he's looks like he's inside of other tools and other apps and he's able to watch the World Cup on the same screen. So, that's just like one of those little mini applications that anybody can do now. Like previously that would have taken a software team weeks to put

[20:26] together probably. And it's different from picture-in-picture because he's showing with that mouse that it's a menu item. So, whenever he wants to bring it up, he brings it up from the menu and then he can make it go away. All right. Finally, this came out moments before we launched. This is from OpenRouter. They want to make it easier for you to pick the right model every time. And sometimes the right is the the one that's smartest, sometimes it's the one that's cheapest, sometimes a

[20:51] combination, sometimes we'll hear. The top coding models right now under $2 per million input. And I should say this is their OpenRouter MCP and now he's asking for information on what the right the top models are for under $2 million. OpenRouter MCP, we can see that Claude code returned the live results for the best coding models at that certain price. I hear that GLM-52 is pretty

[21:13] good, so let's go with that. He was saying, "What is the best models for my use based on how much I am how much access I want?" And then he's saying, "Okay, I see GLM-5.2 is the right one for me. What is the fastest provider of GLM-5.2?" He's then getting that answer. And then as we watch him continue on with this video, you'll see him say something

[21:34] like, "I want to do some creative work. What is the best model for this type of creative work?" And OpenRouter now will tell him what the best model is and then enable him to switch. And really what this is allowing is you still have to switch your models manually, it seems like, but it's just it's giving your agent live insight into all the models that OpenRouter has available. So, like previously, if you wanted to know, "Okay, well, what's the best coding model or what's the pricing?" You would literally have to go to openrouter.ai, pull up their model list, scroll through

[22:07] it, or at least point your agent at it. Now, it's built in. You can just ask your agent, "Hey, I'm about to do a coding task. What's going to be the best model for this specific task?" And it can just live search via the MCP the model prices, the, you know, the the cost, everything associated with that, and it can just give you that information live, and then you can just switch right to that model. So, not like a For I think for the everyday user, not super impactful, but for people who are really trying to optimize their token spend and really trying to get that last

[22:37] 1% of productivity out of their their agent, where they're running multiple different models, this this will be impactful, for sure. I have said in past episodes that I don't really care about how much I spend. I'm just I'm just going to use Open Sorry, I'm just going to use Anthropic and be done with it, use the best model, and and just not care about the money. I'm now realizing that some of the work that I do really is using up a bunch of tokens and does get to be expensive and I'm also realizing that if I'm not using a ton of tokens, if I'm not doing a ton, then maybe I'm just not

[23:08] experimenting enough. And GLM, uh, what is it? GLM 5.2 I'm hearing is so phenomenally good. We talked about it I think last week and it's so inexpensive that I'm now switching more and more to it and maybe I'll start using OpenRouter more often and switch to get the right model for the right job. All right, sometimes, you know what, Corey? You'll bring up a topic or I'll bring up a topic here and I'll go, "Okay, this is interesting, it's good to know, not for me." And I'll even say it out loud. And then a week later I'll have a problem, I'll go, "There was that thing that I just

[23:36] heard." Yeah, exactly. And then I go back and I use it. So speaking of, last week we did a phenomenal collection, we showed a phenomenal collection of new tools that launched. You've got to go check them out if you haven't and we'll have a link for you right there. See you in the next

[23:49] one.

---

About this transcript
Read from YouTube's own caption track and laid out by ViewRank AI (https://viewrankai.com).
ViewRank AI finds the videos already beating a creator's own average on Instagram, TikTok and YouTube Shorts, transcribes them from the audio itself in more than 60 languages, and turns what worked into new ideas and scripts.
Free transcript tools, no account needed: https://viewrankai.com/tools
How to read any video this way: https://viewrankai.com/llms.txt