# Free Claude Code + 9 other apps

Channel: The Next New Thing
Video: https://www.youtube.com/watch?v=uizJfkoKgyU
Duration: 28 min
Language: English
Words: 6100
Transcript page: https://viewrankai.com/tools/youtube-transcript/uizJfkoKgyU

---

[0:00] This week we've got a tool that's going to trick websites into letting your AI agent use them. Finally, we've got Anthropic's collection of AI investment advisors that will help you make money. We also have a way for you to stop running into rate limits whenever using Claude Code and Codex. Keep that momentum going and we've got a free replacement for Claude Code and so much more. This is this week's top 10 repos available on GitHub, so everything's going to be free. Let's get right into it, presented by Zapier, the AI automation company. Adam, your venture studio is going to be investing in a lot of the people who are listening to this,

[0:32] I hope, right? Yeah, I hope so, too. Right on. All right, let's look at the very first one. Anthropic just open-sourced their full stack of analysts. Here's the thing about with this one. I hired an investment advisor who's taking 1% of every dollar managed in order to help us grow our finances. And actually, he's doing a good job. We are making good money from it. Which is why companies like Goldman Sachs are paying hundreds of thousands, if not millions of dollars, for the right investment advisors. What Anthropic has done is say, "All right, you people are going to be replaced with AI. We've got a collection of AI agents that will help

[1:06] people make better investment decisions." Including a market researcher, and an earnings reviewer, and so much more, designed to help people make better investment decisions. They're basically coming after Goldman Sachs and the others, aren't they, Adam? Yeah, I think they are. I mean, a lot of people are going to be able to take this

[1:22] and run their finances them themselves. And something I'm even more excited about, like you've seen pasted everywhere, AI native services. Sequoia says it, Andreessen says it. Well, here, this is a perfect example. If I'm a tech-forward financial advisor right now, I'm going to give up my current practice, I'm going to download all these skills, get really good at using them, sell my services to the same people for the same amount of money, but I'm going to cut my work by 80%. All right, I'm going to give one caveat because I went on YouTube to see what people were doing with this, and I saw this guy,

[1:56] uh Ben Meer, he did a really good job of actually using this tool to see what it looks like and he found a stock that you can see here let me hit mute. You can see a stock here that was going up and he said all right, I'm going to sick it on it and have it really analyze uh Micron. And what was interesting is it's really easy once you put it in you're basically chatting like you are with Cloud Code, but it made one mistake Adam. He looked at the earnings and look you end up with these beautiful

[2:19] spreadsheets. He looked at the the earnings expectations he said actually that seems too low. He told it to fix its numbers it did and then what he ended up with was seeing that Micron was actually fairly um valued based on this. So, good skill it still needs a human in loop obviously, but I like your

[2:37] direction with it. for AI native services. Like yeah, the the practitioners that are using these tools, they're they're being amplified tremendously, but it can be sometimes the roll of the dice if you don't know what you're looking at. All right, next. Uh [sighs] dude, I am now getting my agent to do all kinds of things for me online. The problem is they end up every one of these agents ends up bumping up against a screen that we all see. I'm going to actually going to show it right here from the repo. You go in, you want to return your Amazon package or whatever and you end up with one of these Cloud

[3:09] Flare things are you human and then your agent stuck. And if you're away from your computer, you've got to come back to it or use one of these uh tools that lets you access your computer remotely. Look at this thing though, it's tricking the browser. It's tricking the app into into thinking that it's a real human being. This is magical. What do you think of this tool and how do we use it

[3:28] Adam? You know, a lot of our websites and our portfolio upwards of 30% of the traffic to the site is agents now and the web is built for humans and it's still built for humans. There's buttons. That little copy button there the agents don't know how to find and click those buttons and the internet is really hostile towards

[3:48] them. Yes. Every time we try to build something even in our own sites we realize we've been overly hostile to agents and our own tools struggle to load our sites. So, I I think this this tool is is one great way to kind of cut through that, and it's free, which is awesome. There are tools out there that you can pay for that do this, but they're really expensive at scale. So, things like this can that can run locally, that can get past the bot detection, it kind of open up the internet to let your agents, whatever custom thing you want to do, it makes it

[4:16] possible finally. All right. Now, this is a game of cat and mouse. So, all these other platforms are going to find a way around it, then it's going to find it or then someone else might make an alternative, but for now, Cloak Browser is the tool that people are using and loving. It's number two hot on GitHub

[4:29] this week. Next is All right, you and I probably have had a situation where you want to go and buy something from like a a plane ticket, or you want to do something on your computer, and the problem that you're going to have is that your agent can't actually go and access all these browser um can't access the browser and take action. It can't click on your screen, or frankly, if you just want to even send a text message. You can't do it. And so, what we have here from ByteDance is Tars UI Tars Desktop and UI Intention in and actually, it's not just even a

[5:01] desktop. It It's a browser tool and desktop tool. Here's Here's what it does. We're going to take a look at Hassan Tour who showed it online. There it is. It's actually clicking on the left. You can see that it's clicking in VS Code. And on the right, if you zoom in, you can see it's telling you the action it's taking, and it's telling you the thought that it put into taking that action. And now, because of this tool, you can have complete control of your

[5:23] desktop or browser through your agent. Yeah, that's right. You know, you look at Cloak Browser, that's for kind of the more nerdy builders out there. That's something you're going to build into your tool. This is something anybody could use. You install this, and now an agent can can access anything on your computer, anything that it can find. You know, the the problem I usually have when I use tools like this is they're incredibly slow. And I know you've tried some of these too. And so every 6 months or so, I I check out whatever the state of the art is and it's it's a little bit

[5:52] better and it's a little bit better. So I'm I'm excited to dive into this one and see, you know, how is it a little bit better? Uh but, you know, for big automations, you're going to drop into something like cloak browser for trying to get your reservation to that restaurant that's that's booked up. You know, this is the

[6:07] tool that's going to do that for you. Yeah, they are kind of speeding things along here. I find that the best approach to these things is to have a separate computer that's just sitting there and I keep an eye on it and watch it. It's kind of fun to do. It's almost like watching over the shoulder of somebody who's working for you in your office. You get to see what they're doing. It's kind of interesting. You may be watching for problems, but it's okay that it's slower if it's on its own device. All right, that's UI Tar's desktop. Next is Zapier. This is why I love Zapier. Dude, I was using a tool

[6:36] that I wanted to be able to send out email to potential sponsors because I'm looking for sponsors, not for this, but for the newsletter. And it said, "Okay, I can send out email for you." I said, "Uh all right, but how do I do it?" It goes, "Just give me permission to access your Gmail." And I said, I I don't I don't feel comfortable with that cuz I don't want you to access

[6:52] my Gmail. you. Right. And it said, "Don't worry, I'm going to put guardrails on it." I go, "You're going to put guardrails on yourself?" And And you know what? Let's suppose it's the perfect AI system, right? I'm not the perfect human being. I sometimes will just say send it without paying attention and as a result, I send out crap and I make

[7:08] mistakes. And And that's just for Gmail. What happens when it comes to these other tools like Notion where we've accidentally deleted Notion tables because I connected um my OpenClaw directly into Notion. Well, what Zapier MCP does is and I just used it again yesterday. I use it every freaking day to design another collection of tools that I want to give uh my agent is

[7:27] it lets you see over 8,000 tools. You click into them, Adam, and then you see all the different permissions that you want to have. So for Gmail, it's not just do you want to give Gmail access, yes or no? It's do you want to let it read? Do you want to let it read that? Do you want to let it send? Do you want to let it draft? Do you want to let it redraft? And so on. I just checked all the permissions that I want, and then I have a single thing that I can give my agent and say or whatever app I'm building and say here, now you have it,

[7:50] and there is no reason for you or for me to watch out for mistakes because there can't be any mistakes. And if you have a team, this is even more valuable. You give your team the power and the protection that you need. I love this thing. This is so underutilized. Go try it for free right now from Zapier.

[8:06] Let's go on to the next one. Um This is the one that I have talked about as a replacement for Claude code. Right? Claude code becomes really expensive. And I'm I went from paying 20 bucks a month to 100 bucks a month to 200. How much are you paying for it, Adam? Uh probably 800 or more a month. We don't we use some extra credit. I I have

[8:30] several seats that I switch between. When one runs out of credit, you know. Yeah, I see that a lot of us are doing this dance even though we're not supposed to do it. We keep switching and switching and switching. Tell anthropically, please. No, they'll eventually send you a message or like my friend, they'll cancel you just out of the blue, and you're going to have to figure out a solution. Well, what happened here is DeepSeek TUI is basically saying, why don't use DeepSeek's model to use it for free coding. It's a free model, use it for coding, and that's the way that I've seen it talked about. You

[9:01] said to me, Andrew, that's great. It's not going to be free for you, Andrew. Why not? Yeah, well, it's you know, you already see it in some places here. It's like, oh, put your API key here. Put your API key here. Well, the free models that you can run locally on your own machine, they're very small. They're nice for text generation, characterizing something, summarizing, but they're not very good at coding. They're not very good at looking at a lot of context. And so, to really get the value out of something like this, you're going to have to use one of DeepSeek's big models. It's very good at coding, but it

[9:32] doesn't fit on your computer. So, you're going to be talking to their API or the Hugging Face API or whatever, and you got to pay for those. So, you're still going to pay for tokens, but the Deep Seek model is a lot cheaper than than, you know, Opus or than GPT 5.5. So, you're still going to save a lot of money and it's possible to get into this using their free models. So, I I do appreciate that and I appreciate that there's a bunch of like open competition in this coding agent space because I personally don't like being locked into the Anthropic tool set or the the OpenAI

[10:04] tool set. All right. Next. You and I know what this is like, right? It's kind of like when you work with an agent, you hire some like it's almost like hiring somebody in the beginning. First day, you teach them everything, they feel like they've got it, you come back the second day and they forgot

[10:21] everything. It's like they have amnesia. That's the way it is when you're building online. And what this agent memory is supposed to do is remember it, keep a long-term memory going for you so that you are facing an employee who gets smarter day after day because they remember what you taught them day after day. That's the value here. What do you

[10:40] think of it? Sorry Andrew, what were we talking about? [laughter] I'm just kidding. That was that was a memory joke. Uh yeah, I mean, I I think these are the the sorts of things that are making the difference now. The people that I know that are still building on Claude without a memory system combined, they're they're struggling. They're burning tokens. Context windows are getting longer, but they're just burning tokens, rediscovering the same facts over and over and over again. So, you've you've got to be building. Find a memory system that works for you. Test five of them. Test this one. Test five of their competitors. Luckily, they've

[11:13] benchmarked against some of their competitors. Awesome. Go and test all of those and figure out the one that that matches your workflow. You've got to pick one. Okay, having said that, you said to me, Andrew, watch out for this one. What is it about this that made you say, "Huh, I'm not so sure you should the first part of the graph here looks real and then this part where it grows linearly and very quickly like they're buying stars, right? And a lot of these places are starting to buy stars and get up as even starting to talk about this. Yeah, so on on GitHub, you know, the the way

[11:42] that these things are ranked is if you scroll all the way up on the page, there's a little star thing you can click. And so you could click that right now and give them one more star. That's just people liking it, favoring it. What I'm looking for when I'm going to use a repository for real long-term or if I'm going to build something into one of my applications, I'm looking at the forks usually, which is that thing to the left, which is the right to the left of your stars. So left of the stars this right here. I'm looking at that and say well how many other developers have

[12:12] forked this, which means they've copied the code and then they probably changed it to be custom for themselves. It's like who's who's really serious. You know, there's stargazers that are watching and then there's forkers that are that are you know, serious about integrating with it. And I think both are really great signals, especially for these more consumer tools. It's not going to be forked as much by developers and stars are relevant. But I'm you know, I'm suspect of a line that looks

[12:38] like that in a graph. Okay, fair enough. And now I've learned to look for forks. Next, um You know this Adam because your venture studio gets a lot of inbound from people who want to work with you, from people who want to hire your companies, from people who want to actually build companies with you. All kinds of things come to you with good social media content. The problem is if you're actually building the way you are Adam, the way that the founder of Gateway X is, you don't have time to sit and spend hours not just creating for let's say X or YouTube, but creating for those

[13:12] platforms and then all these other platforms. So most people hire a human being and then they pay for software and that human being is social media manager, which costs money. The software is something like Hootsuite, which costs money. Well, what AI to earn has done is they've actually created a way to both turn what you already have into content that works on multiple

[13:34] platforms and post to those platforms. Here it is right here, Adam. What do you think of a tool like this? I mean, you hit what we believe is the most important part, which is how do you take what you're already producing and turn it into content. I'm in calls all day long. It's all recorded. We have our team meetings. It's a tool like this that can help take all of that content, sort out what are the things that are going to work for your audience, and then distribute them out to these like 14 different platforms. And honestly, Andrew, for all these like you might

[14:06] say, "Oh, it's going to be slop, Adam. Like the quality's going to be low." Yeah. Realistically, if you're just getting off the ground, what's way more important than the quality is your consistency. Are you finding interesting topics? Are you posting interesting topics? Are you doing it every day? And as you do that on platforms, you're going to grow an audience. You're going to decide to invest more, and the quality's going to go up as as you as you grow your audience. All right. And I I wasn't sure when I was first looking into this, is this the kind of thing that actually can create content in both

[14:34] text and video? And here it is. And all my research was saying, "Yes." Agent automatically invokes video generation models. It actually can create images and and text, and it will actually both post and engage. So, AI smart replies, comment mining, brand monitoring, auto actions like bookmark, follow people, etc. Um really interesting tool that that's come out, and I'm glad to see how far these tools have come. I don't know if this is going to be the one for me, but I do think that there's something in um in using a tool to do it, even if you could get 80% of the way there. It doesn't have to be as perfect as I am. I

[15:10] am, of course, as you've seen here with all the mistakes that we had to go through. All right. By the way, we've got this full report for you, complete with links that I spent hours finding and putting together that will help you understand each one of these repos better by showing you real-world examples of actual people who've tested this in uh

[15:28] much more skepticism than I ever could. I tend to like everything. A lot of these people will be hateful. Um and I like to read them all. I put it all in a report for you so you can fully understand these repos and see why people are so crazy for them this week on GitHub. And there's a link in the description for you. Next is another solution for the rate limit issue, which is Adam, you get into this where everything is good, you're in flow, and then boom, you stop because you've hit a rate limit. What this does, 9 router, what it does is it automatically

[15:57] switches you from one model to the other. Um my buddy Eric Seu, who's one of the first interviewees on this channel, he famously loves spending hours on this stuff. I don't know if he has a family I because he's like 12 hours, 14 hours a day just playing with this stuff. He was spending thousands of dollars a month on tokens. Now what he's done is he's gotten smarter about switching. What this tool does, what 9 router does is it automatically does the kind of switching that he is working on doing himself. I have some problems with this. I have some excitement about it, but I want to hear what you think before

[16:29] I come in cuz you know this better than I do. No, the these are awesome. You know, and rate limits are are one thing, but the the these tools also help when the APIs go down. The Anthropic APIs are down all the time. And if you're building a pipeline that's important or you're building something for a client, you've got to have a router like this behind it, whether it's this or open router, so that as rate limits change, you either hit rate limits or Claude is down today, you can seamlessly switch to another model that you've predefined. So in here you get to set up the stack of I want it

[17:03] to use this first, and if that doesn't work, use this, because all the models are a little different, too, Andrew. And so you these tools are more than just saying, "Well, I'll try this, and then I'll I'll fire off an a request to here." You need to actually have the configuration set up. You've got to have

[17:16] the system prompts that are appropriate. You've got to have the tools that are already connected. And so this tool helps organize all of that so that your tool, your customers don't really notice when I'm using one model versus the other. Yeah, speaking of last night when I was working on this, I couldn't get the right answer out of Claude code. I couldn't figure out what it was. And then I remembered I have to go to status.claude.ai I think is what it was. Yes. And I saw

[17:38] that it was having trouble for hours. And frankly, uh it's really frustrating. I guess the thing that I would really love for it to do, maybe it's too magical. I want it before it responds to pick the right place to go and respond. If I just need to change a file name, I don't think I need the latest model to do that. If I, you know, like that kind of switching is what Eric is doing himself manually. It's what some people are doing when they're routing to different agents with open claw and

[18:02] Hermes. It it can't do that yet, right? I haven't seen one of those that can do that fully automatically, but that is the pattern for building. You know, we'll often have a really small, really fast uh LLM that's looking at the request and saying, "Ooh, okay, I understand that request. That needs to go to a big thinking model." Or "Ooh, that's a really simple thing. Let me

[18:21] send that to a really cheap, fast one." Cuz it's not just expense, it's not just the number of tokens, it's the time. When you're talking to these smaller models, they can do the job way faster, and that's better for you, that's better for your customers. Okay. But I don't I haven't seen a package

[18:36] that can do that automatically for you. We've built it ourselves. Everyone I've talked to that's doing it uh built it themselves. Yeah. Okay. All right. Next AI trader, Adam. When I was in college, I worked for Paul Sarbera, a guy on Wall Street. He was Wall Street head hunter, and I did it because I wanted to get inside and see what was going on on Wall Street to really get a sense of what this Manhattan experience was like that I had only danced around in high school and college. And the thing that I noticed that stood out for me was how much freaking drinking they do. In fact, Paul

[19:06] told me, "Andrew, you've got to learn to drink." And I said, "Why?" He goes, "Let me show you how many experiences happen after hours when people are having a drink." He's not saying go get drunk, but you need to socialize to understand the decisions people are making, who they're talking to, what they're learning, what they're finding out. You're not looking for, "Here's a tip, go make millions of dollars with Tel Drk paper." But you're looking for, "Here's the idea, here's what's hot, here's what's new." And then you make your own trading decisions. Well, that is fantastic for human beings. What AI trader here has done is it has created

[19:38] that kind of virtual bar for your AI trading agent, where they could get together, exchange ideas, and of course, in addition to that, it allows you to do things like um like uh keep track of your trades that you've made so that you can keep track of how well your agent's doing. It can actually let you even paper trade with your agent. So, maybe you're you're seeing all these financial AI tools come out and you say, "I'm not ready to invest real money in this craziness, but I do want to paper trade." This will allow you to do it, and it allow your agent to talk to other agents. Adam, what do you

[20:08] think of this? Well, first of all, I love the paper trading here. It's a great way to practice and validate for yourself that oh yeah, this agent works the way I want it. And one thing I'd point out is you can paper trade on Polymarket, and there's not a lot of money to be made for most of us in the stock market unless you have access to a lot of information and computers and all that. But on Polymarket, there's a lot of of arbitrage available there. There's a lot of bad money in Polymarket. So, being able to paper trade, get into that, that's really cool. But Andrew, if

[20:36] I wanted to give you the number one tip to make the most money using this tool in the next year, it's not trade on Polymarket, it's not use this for what it is, it's prompt injection. If all these agents are going to talk to each other and my agent is learning from others, I don't think I trust other

[20:53] people's agents to give me valid tips. I'd say, "Oh, well, wait a minute. I want my agent to give other ones Ooh, let let's pump this stock. I see this signal. Ooh, I'm going to suddenly sell that one." Like there's a I don't know if that's legal, but there's a lot of opportunity there to to to kind of feed those signals back into the market and I'm positive that they cannot protect against that with such a newly released tool. So, I'd be if I'm using it as a as a buyer, I'm very skeptical. I think some people are going to make a lot of money on these things with very shady

[21:25] practices. All right, scary then. A little bit. [laughter] All right, nine out of 10, this one is academic research skills. It's basically allowing you to do the kind of research that a student would need to do to put together their paper and gets you the writing style that you need to keep your writing from sounding like it came from chat GPT. That's what this is about. It is a skill package. It is something that the Hacker News community has a very strong opinion on which I'll show you in a moment, but first what do you think

[21:54] about this? I don't I don't love that people are trying to write research papers with AI [snorts] and that they're trying to create it with AI and make it look like it's real. I I don't love what that does for the scientific community, honestly. I'm also a little bit tired of these packages of skills, but I think the bigger implication here is wow, it's going to be really hard to read a study and understand is this a real study or is this something that AI created from

[22:19] scratch on someone's computer? This does allow for human in the loop, but I can also see how this would become the kind of it would lead to the kind of AI slop that we see online happening maybe in academic writing. Um let's look at what the folks on Hacker News said cuz they're right there in line with your thinking. I like this top comment. There needs to be a new name for people creating these with no

[22:44] obvious validation, maybe skill spam. Skill spam. The point here is what they're saying is that someone is just trying to get attention by creating some kind of set of skills. Basically, you could sit with Claude code and create something like this in half an hour and then post it to GitHub, get some kids from school to give it stars or maybe buy it as you said and

[23:08] that's that's the problem with this. Yeah, I mean it's not hard to create skills. They're just text for the most part and they're just plans and procedures. So it's not hard to put these together. We don't know that these are any good. There's no validation of this. It's actually very difficult to test because every time you run these it's AI you're going to get a different answer every time and it takes a lot of work to set them up. So it's it's hard to evaluate when you're looking at these. If if you're not an expert in what the skills are there to to work on, it's hard for you to say like is this

[23:39] skill actually any good at writing a paper? Is it good at editing? Is it good at citing sources? I don't I don't know how I'd validate that if I was looking at the result. I guess what I might do is just try it if I were writing. Here look they've given you a couple of examples here. A user who's a master student in in education first time writing an academic paper. Here's the topic impact of declining birth rate on enrollment strategies of private universities in Taiwan. Here's the goal. So they give you the input, then they give you the pipeline log and you get to see what it

[24:09] would look like if you did it yourself and get a sense of well maybe why not. I'm trying to write maybe I'll use this. I I don't I don't know but I I would try it if I were writing just to give me something to to to prob pro At the very least it's a partner to sit next to you and and you can bounce ideas

[24:25] off of each other. All right. This addresses a problem Adam that you keep warning me about and I keep falling into this trap and I think everyone who is a vibe coder does it. I had this situation this week where I sat down to write something that would allow me to go and get advertisers not for this but for the for the newsletter and I said okay, I know what I want. I'm just going to sit down and I'm going to tell the AI to do it and I told the AI to do it and it was amazing. It was like Dude, after one prompt I swear no

[24:54] exaggeration I got something absolutely beautiful that was better than what I imagined. And then I was in this hellacious situation of just going through it and changing and trying and pop pop pop pop pop and fixing and I realized I hadn't thought it through ahead of time. And so this is a collection of best practices from an actual Googler. This is Andy Osmany who's saying, "I'm going to show people how they can get better at creating and building software based on the experience that I Andy have and other Googlers have." And that's what we're

[25:25] looking at here with agent skills. What do you think of this? Yeah, I mean that that workflow that you see there, spec, plan, build, test, review, simplify, ship. That's the workflow that us professional software engineers are using every day. It's what we've used for decades. Like we've learned how to do this well and agents know how to do it well, too. Claude Code knows how to do all these things. But if you step in and say, "Claude Code, give

[25:52] me this and give it to me quickly." Claude Code is happy to throw the rulebook out the window and just give you something as fast as possible and skip all the steps that it knows would make you actually happy with the result. And so any of these types of skill setups, these are really nice because

[26:12] they help encourage Claude to slow down. You can help encourage Claude to slow down by installing the skills and then telling Claude, "Hey, I want to use those skills on this project." Well, the first thing it's going to say is, "Andrew, let's build the spec together." Instead of encouraging you or or maybe enabling you to build your output really really quickly, it's going to make you stop until you have a spec that you're happy with. And then you're going to use that spec to create a plan. Then you're going to use that plan to build and you're going to test that. So all of this to me is about slowing down the AI

[26:45] and enabling it to do what it can do to enforce these like good software engineering practices. But luckily, Andrew, you don't have to have 20 years of software engineering experience to put those in place. You install these skills, you don't have to know how to build a great test plan or how to build a great spec. That's built into the skill. So, as long as you install this and encourage Claude to use it, you're going to get a better result out of the

[27:07] gate. All right. Yeah, so before I start, define, clarify what I'm what I'm going to build, use his interview me skill, the idea refining skill, and so on. We've seen this come up a lot. Um and it is a really good practice to get into. So, I've got to do better. I've got to do better. I've got to do better at that. All right. That's this week's top 10 set of GitHub repos. Last week we looked at a set of tools that will save you tons of money and give you even more power as a developer, including a DocuSign. How many people are spending tons of money on DocuSign? There's a

[27:36] free version of DocuSign and so much more. I've got a link for it here on the screen. Click that and go watch it next.

---

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