Using AI Agents for Strategies Part 3
Alright, everyone. Welcome back to another episode of Line Your Own Pockets. We're gonna go back into the AI chat again. Just inside kind of how we're recording this, we recorded a couple and the first one just came out. So we're able to see all the feedback and there's a lot of questions and people were super So we figured we'd do one more episode kind of talking about more my exact process and and more of the nitty gritty and the inner workings of how we're using these things and just a little bit more AI chat from there.
Michael:And then we've got a couple topics we wanna branch out on, but I am sure we will be back to this pretty pretty quickly because it seems like every, what, six months, there's a huge update, and it's just kind of changed the way everyone works on everything.
Dave:Yeah. I think there's a lot of good feedback on the first episodes. Yeah. Think this would be great. And, you know, I I don't usually go back and listen to these episodes after we record them, but I it actually, I was on a bike ride and it came it happened to queue up.
Dave:So I listened to it the the first episode. Yeah. It just gave me some, you know, more ideas and Okay. Some ideas for this one. So, yeah, it's it's Well If you haven't obviously, if you haven't listened to those first the the two previous, please go back and listen to those.
Michael:Yeah. Good good for you. I cannot listen to my own voice. I just I just can't do it. I don't.
Michael:And I'm sure there's tons of, like, errors that get through, like, the editing process of my videos and everything because I just I can't do it. Can't sit and listen to myself talk for even ten minutes, so good for you to be able to be able to do that. I don't know if that's just a me thing, but yeah. So we we started just talking about how, you know, I guess the term is agentic AI, that's the the fancy word they're using for it. And and kind of how we've been, or myself personally, have been really changing the way we're back testing and and coming up with ideas because it opens up a certain amount of brute forceness that probably was not time economical before to just say, I have this very broad idea.
Michael:I want you to do, like, 50 tests in different variations of this very broad idea, and then bring to me the ones that look the best, and I'll I'll go in and and refine them. So whether you're finding ideas externally or whether you're, you know, going in and and coming up with your own ideas and saying, there's something simple like mean reversion. There's probably a thousand different ways to express, you know, a mean reversion setup. So giving the AI instead of trying to figure out what one idea to work on at a time, because these things can just kind of run for you and they can run real tests and they can run Ami broker in the background to be able to say these are the five ideas I have, go test them and give me your findings and then from there we'll we'll continue down whatever road is the most interesting.
Dave:Yeah. So you mentioned that you got a lot of feedback from the first episode, had some questions. Maybe we'd go through some of the questions or or just the general themes of the feedback you got, and let's address some of those.
Michael:Yeah. The I think the first one and then the most prominent one was something that I didn't know that I could do either until I asked, but people just didn't know that you could run Amity Broker and RealTest through the command line at all. Like, they didn't so RealTest, it was very easy. Right? It was right away.
Michael:There was just commands that you could just run, you know, like import and orders and we'll pre see orders or import and tests and it'll it'll run the back test. So, yes, first of all. Right? That's sort of saying, how are you doing it? I'm saying, well, I'm instructing the AI to run it.
Michael:It's not going in and hitting the buttons inside the program. What it's doing is it's going and it's using the command line to run them. Now with Amnibroker, we talk hinted about this a little bit in the other episodes. It was a much longer process for it to learn. I kept, you know, giving it articles.
Michael:It kept failing. It kept back and forth, probably, like, two, three days of just and it almost it would stop every now and then, but like, can you do it? Can can you just go and hit the it's way easier if you go hit the button, and I would just go, no. I've talked to people. I know that this is possible.
Michael:So I know you and it just just continuing through brute force to figure it out. So both are possible. Real test is very easy. Ambly broker, it seems like it's a lot harder for it, but eventually it got it. And then then what's important is that you tell it to update its memory, and you keep it in the same thread that you had that conversation.
Michael:And once it's kind of broken through the wall, the beauty of these things is they never forget. So it's figured out how to do it. It can now do it. So that's how I'm doing the most the broadest part of it is just using the command line and having the the AI use the command line where it's writing up the code and then running it and testing it. And then if there's a break, it will see the error and so on and iterating that way.
Dave:Yeah. So it seems like real test is easier here, and it is probably as simpler to set up, but there's a lot more power in the way Amibroker's doing it because even that's a little bit more complicated. So And isn't that
Michael:the same with kind of all softwares is you're always balancing the amount of capabilities it seems like, you're the software dev. Right? The amount of capabilities with the complexity. Right? There there's it you're always balancing that.
Michael:So if, you know, if we're looking at like a seesaw or something, RealTest would be much simpler but less feature full and then Amibroker would be kind of the exact opposite of that.
Dave:Yeah. Yeah. Think that's true. And and so you mentioned it took two or three days for qual to figure it out. With Amibroker, well, the code is right up there on my website.
Dave:That's what you should plug in and save yourself two or three days and a whole bunch of tokens for claw to figure it out. It's right there. So, yeah, if you go
Michael:Yes. And I think
Dave:the newsletter will will have that right in their inbox. But
Michael:Well, I I think that existed after. I think I don't know if I was part of the motivation behind that, but, yeah, I think that was I I saw that email come through after, And I'm like, I had this last week Yeah. It would have it would have saved a little bit of time.
Dave:Yeah. That'll save you a bunch of time. So there is a lot of power there. And this interface for like, it's funny because these command line interfaces, CLIs, like, I I I've been working on this stuff for thirty years. So when I I remember when I first saw CLI, which is so common now.
Dave:After AI, you say, okay. CLI, command line interface. I didn't know what it was, but like, this is literally the stuff I've been working on forever. And it's like, I love that it used to be sort of how I differentiated myself in my career, like, doing all this automation stuff. Where I love now, like, this has opened up to everybody.
Dave:There's so much possibility here for anybody. You don't have to know all the ins and outs of scripting language or a comm interface with Windows. The Clog can just do it for you, which is great.
Michael:Yeah. And that that's really been that was like the moment, I guess, for myself and and all this. I'm saying, Well, it can Cloud Code, and we'll talk about the second question, which was Cloud Code and OpenAI and why why pick one or the other. But the fact that ClaudeCode can interact with a browser and it can very easily go, it has access to a sandbox kind of Google Docs that I set up with a different Google account. And I can kinda tell it to go, hey, and and, you know, make your own spreadsheet and and start to track your own progress and do your own thing.
Michael:So I can give it very large projects and then go to bed and then wake up and have the that's the different thing. Whereas before with AI, there was you know, I would type, hey. You know, I'm I want to do this in real test. I wanna back test this process. And it was still a huge advantage because it would go through and try to just, like, spit out here is the file, here is the code, and you'd have to copy and paste that yourself, and you'd run it.
Michael:Maybe there's an error, and there was a lot of this back and forth. So although it made you way more efficient, you still had to be present for that efficiency. This next breakthrough is just the difference of the way I look at it now is my computer is a resource, and I want that resource always working. It's either going to be running trading during the trading day, or at night, it should be running some amount of research or backtest or something. And it's it's now a feeling of wasted time and almost a feeling of loss if I'm not if that thing's not just out there churning something because I'm like, well, I don't need to be involved in it.
Michael:So it should be doing something at all time. And then it's to the point where my new backlog isn't the ideas to test anymore. My new backlog is, like, reviewing the work that the Claude code is is spitting out now because it's able to go eight hours a day kind of nonstop.
Dave:Yeah. So it's interesting because I guess I have probably too much experience in the field and I know what can go wrong even when you're, you know, code that I've written. I, I, that's obviously the case like it's so there there's going to be a code that gets written. That's you just don't realize there are bugs in there. You can't see them.
Dave:So it's a little daunting for me to have something running all night doing something on its own. I'm much more comfortable right now having it get permission for pretty much everything it does. And I think there's a little bit of a give and take out. I'm sure that over time I'll get I'll be able to trust it more and set up guardrails for it to do, hey, this specific task. But I think it's actually quite interesting to see and have it ask permission to do every step, at least at the beginning.
Dave:Because I can see a lot of times I'll see, oh, it's so it'll say, okay, I want to do this. Do I have permission to do it? And it'll be completely wrong. I was like, oh, okay. And I can immediately say, okay.
Dave:I see what you're doing here. I see why it's not gonna work. And I can immediately step in at that point and have it update the clawed file or I'll update the clawed. Md file so that it never makes that mistake again. And if you're not having that feedback, then I think you'll end up like you tried with two or three days to figure out how to examine broker.
Dave:You can really shorten these workflows and processes if you really watch it, do it in real time, but also have a deep understanding of what it's trying to do, why it's trying to do that. And then you can steer it in the right way and you can make your process a lot more efficient. Like you mentioned, you use the $150 plan. I just use the $20 plan. I don't, and I don't run out of tokens because I I think it's probably because I'm very I don't know.
Dave:Maybe anal about Yeah. Having it do exactly the right thing and sort of cutting it off when it tries to do the wrong thing. So it's just sort of interesting, a different approach.
Michael:Well, and and, of course, knowledge will always save time. So the question is, right, as someone who doesn't have that knowledge, like, $130, I mean, that's that's nothing. Right? Yeah. Big deal.
Michael:It's certainly worth it. So the main difference is that that brute force option is now possible. Right? Whereas before, it wasn't. It was you had to have a a certain amount of knowledge to guide it.
Michael:I still think you have to have the knowledge of, you know, inspecting the output. Right? I saw a tweet on Twitter that kinda went viral of, you know, I don't I don't care how many of these agentic AIs we have. At the end of the day, a human being's gonna need to look at the code at some point. Right?
Michael:At some point there has to be that that endpoint, especially in things that are are very important, like cybersecurity or managing my money. I'm not going to implement something that I haven't read what it's trying to do. I haven't done some amount of paper trading or off. Right? So at the end of the day, I I will still become the bottleneck.
Michael:It's just it's just moved a bit. So that's the main difference that I'll say is that if I had the knowledge, yeah, I'll be watching it. But when it pops up and it says, do you wanna allow access to that? I'm like, I have no idea what that is, so go nuts. So it's I think it's important then now to talk a little bit about sandboxing.
Michael:Right? I have two computers. Right? I have my MacBook, which we're on now, and this does all my content and my webhook stuff, and and a lot of this goes on here. And I also have a virtual machine on here in which I can, in a pinch, run real tests in Amity Broker to do the basic things of, like, generating orders for the next day and whatever.
Michael:And that's like my my kind of backup. Come hell or high water, I can do this. And then the way Mac kind of backs up into the cloud, if this catches fire, I can run down to the Apple Store. I can buy another one, and then within the afternoon, will just pull everything from the the cloud, and it'll all be good to go again. So that's why I'm more comfortable allowing this thing go free, because it's going free on another computer that's not signed into really any of my Google accounts.
Michael:It's not signed into, I don't have, it doesn't have passwords. I gave its own browser. It's the only thing that it can kind of manage. So absolutely worst case scenario is it it bricks the machine and that sucks because it'd be a lot of progress going. I have windows, you know, backing things up and I I have it.
Michael:I don't I don't know if I check it, but I told it to use git to back things up as well. So it says it's doing that. Someday I'll be able to figure out how to check it. But that's, I think, hugely important. If you're just gonna let something go run and go nuts, if it is your computer that you're executing trades on or if it's your computer that has access to, like, all of your life stuff, that to me is is a much harder bridge.
Michael:And just so that people know, this is just a it's like a mini PC, they call them. So it's essentially a laptop without the screen or the keyboard and the mouse. It's got, you know, a very it's 32 gigs of RAM, but just like like a mobile processor and all that. And it's able to run these fine because most of it's just being done in the cloud. So yeah, if you are gonna just let it go nuts and go to sleep, just make sure you've thought about that sandbox environment in in some way, shape, or form, and it can't hurt you at all in kind of your real life or it can't really hurt your trading process.
Dave:So so this reminds me. One of the things I have in my main claw dot m d file so, you know, there's multiple qual dot m d files. There's one in the repo, in the GitHub repo for each repo that I have where I'm having it write code and and do things in different projects. There's a main one that is in your user directory that overrides all that or is like shared among all the all everything where you might be running Claude. In that main Claude file, I have some basic parameters for here's how I want you to write code.
Dave:And one of the things in there is what I call or what's known as test driven development. And it's a way to be very careful about what it does and where you can actually get a lot of confidence in what it's done by writing what are called unit tests before. So it'll say okay I'm going to write a unit test that fails, I'm going to go implement the feature and then I'm gonna run the test again to make sure it passes. It's just it's this other layer that without AI is a huge pain in the butt and tedious, costly, time consuming. But once you get it in, it's really nice to have, and that accumulates over time.
Dave:So but Claude is like, it eats these these unit tests for dinner, man. I mean, this this is what it does and what it that grunt work is really what it's good at. So I have that in there. So every time it does anything, it's going to go through this very specific test driven development process that just increases my confidence in what it's doing dramatically. And I want to talk about, after you respond, I want to talk about how you can implement that in your trading process, a similar concept for back testing.
Michael:Yeah. And and, you know, the only the only mention I kind of wanted to have there is that's the that's the beauty. You just mentioned the the grunt work that this thing would be tedious to do alone, and that's really how you have to think about it. You know? I I know I know people that are, like, uploading medical data, like, to Claude saying, am I sick and all this.
Michael:I just that's still not the move, I don't think, for AI. And same, I still don't think the move is, you know, I'm looking to build a trend following strategy. Give me, you know, whatever. Now, you know, saying, hey, give me 10 ideas and look at the ideas and maybe come up with something maybe. But the it's it's outsourcing that grunt work.
Michael:So that was the other kind of big light bulb moment for this is that this is a this is a slave. Right? This is a, like, like, slave that I don't I don't care about its well-being. It's it's not a human. It doesn't have I can just say, I want to test this instead of instead of saying, want to test this one thing, you should just think about it in your brain now.
Michael:I wanna test 10 permutations of this one thing and then let it go. And just think about it like that. If not, I want you to figure out how to make me a whole bunch of money in the market. I just whatever work that I think is involved in something, make it do 10 times the work and then sort through it and kind of figure it out from there. And it's the same as what you were talking about.
Michael:These these tests are really tedious, but you don't have to do it. So just make the thing. So and it's also really eliminated a lot of friction points of just things that I'm like, man, I know I should be doing x, and I don't wanna do it, and it's annoying. You know what? You do it.
Michael:And then the thing gets done, and we can move on to cooler stuff.
Dave:I would say I would have a little different angle. So you say, okay. Do this, but do it, like, 10 times this, whatever it was. I look at it a little differently where, you know, I've got this big to do list, some stuff, you know, really low priority. And, you know, when it comes down to it, I would probably have never gotten to it.
Dave:Now all of a sudden that stuff is really worth doing when you have something like law that could do it. It's really depleting my backlog. Right? All this stuff that's in my backlog to do, I'm churning through fast. And I actually have I don't know if you let me know if you think this too.
Dave:So there's something in my brain where I kinda like it's sort of like I'm squirreling away to dos, like I'm squirreling away this big list of of strategy things to test. And I kind of like that it's a big list. It's I kinda like deep down that I've got this big list that I'll never get through because it means there's always I've always been able to, know, have some new ideas to test. Like, if all that stuff just falls apart today, like, hey. Okay.
Dave:I've got it I've got this list of ideas that I can fall back on. What if that all that goes away and you catch up? And so I've I've had sort of some anxiety about that, but what I realized sort of like when we did this podcast, you even if you get to the end, you you never run out of ideas. You just move on to other and better ideas. So you kinda have to have some faith that that it's gonna work out that way, and it usually does.
Dave:It really does.
Michael:And I think more complex ideas is is another thing that's because I I had the same fear because the the backlog is moving from I it's just changed for me, and maybe I'll start to speed up. But it's no longer I have this massive list of things I never think I'll get to testing. It's, you know, I have this massive list of things and I'm having the robot test it, but then it's creating this massive things to review. So it just kind of moved it from kind of column a to column b, and maybe I can review those faster. I can, you know, take a quick look at the code.
Michael:I can take a look at the returns and say, no. Not worth it. Move on. Like, maybe that process speeds up, but it's just it's moved the backlog over. But part of the thing that helped me feel better when I was getting kind of that same anxiety was, well, there's so many things that I wanted to test and I wanted to do that are outside of anywhere near my capabilities currently that now I could start to explore.
Michael:So one basic example is, have you heard of unusual whales? This program they do you know, they monitor the options chain for unusually unusual trading events and all that. I always looked at it. I'm like, that'd be really cool, and then moved on. But they just released an API.
Michael:I have no idea how to do an API. I have no idea how to do any of this. But now that's created, not a, I have this idea about, you know, when a stock gaps up and does this phenomenon, whether I should test it. Now the idea is, well, I can now sit down with Claude and over maybe a week or so figure out a way for it to grab the API and back fill the data and and and go on. And just it's a an infinitely more complicated task than just testing a new trading style, and it's something that would have been way outside of my possibilities six months ago, but I could really see myself doing it, writing off week and saying, okay.
Michael:On Monday, we're this we're gonna do this. On Tuesday, we're gonna and have, you know, by the end of the week, understand, is there edge there? Yes or no? I have no idea, but let let's find out. So I'm looking at it as that, a, it's just moving the work, and I'm still I'm still the the blocker, but it's just moved the work.
Michael:But then, b, it's it's opened up, like you mentioned, it's just opened up a whole bunch of other ideas that will kind of continue to probably balloon of things that to me would have just been way too complicated. Like, that's gone from maybe I could do that in a week or two to it would probably take me six months before because I'd have to learn so much that I don't know now.
Dave:Well, you know, the fact that you're the blocker is a good thing. Like Yeah. If one day you're not the blocker, that's probably we should start worrying then. Alright. So so I wanna give you a couple different ways to think about your process of testing ideas that we talked about last week and ways that you may not have thought about for having Claude give you some help here.
Dave:The first one is sort of like the unit test. One thing you could add for each strategy is an example trade. This trade on this day should be in the backtest. So what that would allow it to do is sort of have like a unit test. It it knows when it's successful because it could create a strategy that includes Nvidia on March 3 or whatever it was.
Dave:That would be a really good way for it to create a backtest that includes that and give it a way to test itself so it knows if it's successful or not because otherwise you're you know it's just sort of oh this seems to work right you give it some way to verify that would be that could be really good and I've done that with some strategies having it create I've had that in the workflow and that that works really well Because you know it'll run it, it'll run a backtest in Amibroker through that comm interface, it exports the CSV, and it can add the parameters to say, okay, yeah, I tested for NVIDIA on that day, it didn't show up, let me go try again and see what's going on.
Michael:Well, and I I think that's important too because it kinda like what we talked about the last time where I'm using this to pass this prop firm thing and and not that the prop firm thing is is the interesting part, but the fact that I gave it a very specific pass fail. Right? Did your did the trading setup follow and beat these specific rules, and what percentage of the time did it do it? That's kind of the same way. You're giving it another pass fail point where you're saying, okay.
Michael:I want not only do I want you to back test it, and hopefully, returns are good, but if it's not including this example or this number of examples, then consider that a failure. You know, even if it is good, maybe store it somewhere else. We'll talk about that later, but, you know, keep going back and keep kinda iterating and testing until you you have more of those. And I and I found that the more pass fail scenarios you give to the AI, the the better it is. Like, you know, I wanna see this, and I wanna see this.
Michael:And that's why it was very interesting that prop firm challenge because there's like 10 rules. And, like, you have to be able to pass all these 10 rules, and that gave it this ability to go back and say and I have a spreadsheet somewhere where I said, I wanna I want you to save anything that looked really good, but didn't pass the challenge because I would trade that with my own money. But the having that kind of this is what good means, this is what bad means, that I think is really important because, you know, you gotta remember how a lot of these things started where you're like, is this a cat? And they're showing a picture of a cat and it's saying yes or no. And then a human's going back in and saying, no, that is a cat.
Michael:No, that isn't. So having some way of of kind of reinforcing that pass fail, I think, is kind of at the core of these things and, like, super important.
Dave:Yeah. And you alluded to this last time where it would sort of give up or say, hey, this one's not very good, but, you know, you you have a good process for creating strategies, you use the strategy cruncher. So just because the strategy looks bad in the base back test does not mean it's not profitable. Like, you know, that as well as anybody now. So that's really, that's something you're going to have to teach it somehow or make sure it's aware of because otherwise it could ignore a bunch of ideas that are very profitable.
Michael:So that is one change that I made maybe since the last time we talked, I can't remember if we include this or not, but I've kind of I've gave it access to your website, and I said go through and read all the help docs and learn about kinda what this is. Dave pinged me because it was going a little bit nuts there for a second. But the point of that was then I went back and I said, okay. This is this is the next step beyond you. Right?
Michael:So you are to do the base strategy, and then the next step beyond use this. But I wanted to show you this just to say to look at different metrics. So for example, I I have it now that it will look at profit factor and Sharpe ratio and things that are not determinant of simply return or drawdown numbers. Because that's what it always focused on before is here's the the CAGR, and then this is your your drawdown. And these are what's happening.
Michael:I said, no. I want to see things like smoothness of equity curve, and I wanna see things like like all this. And then what happened was with this prop firm challenges, when it beat it on its own without me optimizing anything, that was the next step. I said, okay. Now open it up.
Michael:So there's tons and tons of trades in there, and then I'm gonna go in and refine it into the cruncher, and then I'm gonna give it back to you. So all of these things along the way remind or or make me feel a little bit better, but we're not done as a race yet because it feels like so if six months ago, maybe it was 30% AI. It was making me better. It was helping coding mistakes. It was this kind of thing.
Michael:Now this kind of feels like more of a team. I'm not it's like overlord or commander or anything. It's I'm directing it. I'm telling it what to do. I'm moving it along.
Michael:I don't know. Maybe six from months from now, we'll say that's wrong, and I'm you know, it's my overlord. But it's just it's very much helping with these these tasks. You know, showing it what the end of the road looked like, I think, was really helpful to say, this is why because I kept saying, there's not enough trades here. I wanna see hundreds of thousands of trades.
Michael:And they would even go back and say, well, there's no way you're gonna trade that. So eventually, I just said, you know what? I'm gonna show you what I'm gonna do next. And I just walked it through the process of of what's gonna happen after after I get the output.
Dave:Yeah, I think that's good. So here's another thing to think about with some instructions you can give it to maybe, treat your list of strategies a little differently and, come up with some new ideas. Back up a bit. You know, I trade 25 plus strategies and the way I think about how many strategies is enough. Some, some people ask me that question, you know, how do you know what's enough?
Dave:And I don't really think about it in those terms, but what I think about is do I have coverage across the market day? You know, I've got the pre market, I've got the open, regular trading hours, post market, and and the close, etcetera. And I'm thinking about, okay, where do I have room in my buying power to add a new strategy and like have efficient use of my buying power across the entire day. So one way you could have Claude look at this is say, okay, that's my goal. Here are the strategies that I have.
Dave:Where what are the pockets? Where am I maybe missing something? Can you look at this and make some suggestions for different areas that, you know, may have room for a strategy. And you may have, you know, maybe that's not the exact context in the swing trading environment, but you could say, okay, find me the holes that I might not be thinking about here or and how can I plug them? Just, you know, it doesn't have to come up with a strategy, but it could just point you to the to a hole.
Dave:And that could help you focus and figure out, okay, that's what I want to focus on because there's some potential that hadn't really considered or hadn't really thought that much about.
Michael:Well, and so it's funny. We had in this same kind of opening range break prop firm beat thing that I'm doing. It's probably gonna be like a YouTube series and all that kind of stuff. So I think it's interesting, but that was a conversation we had where it kept saying, okay, you know, five times your risk is the optimal amount. Right?
Michael:Because it had tested all of these different r values for profit takers. Right? Is if you can cover it five times your risk. And then I went through and I said, okay, But by how much? And I had to run this whole chart and analysis of of running these different things because what had happened is, yes, it was the most profitable by, I think, a couple percent a year.
Michael:But what happened is shooting for five times your risk means on the majority, the large majority of trades, you're essentially just holding all day. Right? The the odds that you're getting that amount of of return on a trade, but actually hitting the limit order is very rare. So I it said, okay. 2% of the time, you actually get that target and you punch out your trade.
Michael:So then that leads to another question, which is again more of the confirming of we're not replaced yet because I'm like, okay. Well, if I'm just holding these things all day, like you mentioned, it's just eating my buying power across all these different trades all day. So then it was, you know, the question of do you do I take a a shorter exit either via time or via profit target in order to free up capital to employ more strategies? So there's yeah. There's still a lot of kind of on the margin questions you need to ask because just like if you sit an employee down, depending on how you pay the employee is gonna depend on how the employee works.
Michael:With these things, it's gonna be all about what incentive structure you gave it. Like, I I gave it the, I want you to pass this challenge kind of as quickly as you can. Away you go. But you're right. By doing that, it kind of looked at this strategy as a single, this is the only thing that will ever exist, and then you have to start to bring in, yes.
Michael:But I'd rather something suboptimal here if it allows me to implement a different strategy near the end of the day or, like you said, the afternoon or whatever it is.
Dave:Yeah. I mean, it just reminds me I think that the intern analogy is so good because you know interns are often very eager, they're very excited, they you know they're confident, they want to please So oftentimes they do too much or do things in a way that they don't understand. So it's just there's so many parallels. It's it's kind of it's just striking how how good the parallel is, I think.
Michael:Yeah. And the just like, right, with a if if a if you have an intern and the intern's just sucking, it's probably your fault. Right? You either gave it a task too complicated or you didn't give the end goal of the task correctly. And I think that in most cases, that's probably the thing is is really having it understand what is what is the purpose.
Michael:So that's why I've been interested in this doing this thing where I'm walking ahead and saying, hey. Yes. This is the cruncher. This is what's gonna happen next. This is, you know, yes, I wanna implement more than just this one strategy.
Michael:So this is just number one. And kind of showing it a little bit down the road seems to seems to help. Because as soon as I said, no. No. You know, if I can make, you know, instead of, I don't know, say 20% a year with this strategy, if I can make 15% a year, but all of these are closed out by noon, that's amazing because then I can implement an afternoon strategy.
Michael:Whereas it's just not going to think like that. It's gonna because I gave it the I want you to optimize for the most amount of money. So come out or high water, it's going to optimize for the most amount of money on that particular strategy.
Dave:Yeah. That's a that's a good point. So here's some feedback I've gotten from not just from our podcast, but just just AI in general over the past few months. Some people will say in fact, I just got this question today. So, Dave, why don't I just create my own backtesting software, have Clog create it, and, you know, easy game.
Dave:Right? Just it'll just do everything. I all I need is this data and it'll just backtest everything formula, like create the ammo broker or the real test on its own. What do you think about that?
Michael:I don't know. I don't know if I've referenced Jurassic Park in this one, but I may have, but it's like the whole saying of, you know, we spent more time asking what we could do and then never stopped to say should we, right, when they rebuilt all the dinosaurs. Like, same thing. It's like, could, but why? Right?
Michael:Ameren Brokers is like, what, a couple $100? Like, just buy that and use that. It just it doesn't make sense to me because you're right. The limiting date the part now is actually the data. Like, getting the the data and getting that in in real time, that's gonna be the most complicated part of that whole chain you described.
Michael:And I'm like, because Claude Code can do it, why not just use Amibroker that's already built and you you're you're a step apart of that. So that's that would be because I that came into my head for a second. I'm like, oh, yeah. I could probably just tell it to build a back testing engine and walk away. And then I thought about, like, well, how do I trust it?
Michael:How do I know the back testing engine's doing things accurately? How do I I do all this? So I think it's better just to have this Ambien broker exists. The data's going into it. I know that all of that is correct, and it just it just operates the same kind of the same thing as if you had an intern come in and sit down, you're not gonna tell it to build its own back tester.
Michael:You're gonna say these are the tools we use, get really good at them and and, you know, sit down and get to work.
Dave:Yeah. I thought that the exact same thing. So if you asked an intern or even a senior software developer, they're going to want to build it. I can guarantee they're going to want to start from scratch and build it. And I see developers all the time that make that decision incorrectly.
Dave:And I mean, just think about all the updates, all the history of I mean, Ambroker's been around for thirty years now, I think. There's just and, you know, you had it think for three days about how to interact with Ambroker. Imagine having it build a back tester or having an eager intern try to build a back tester for you, man. That it's gonna it's gonna seem like at every point, you're gonna think, okay. If we get this done, it's just gonna be right around the corner, and then there's gonna be other steps you're like, oh, I didn't even think about that.
Dave:See right around the corner. There's something else that's gonna come up.
Michael:And the the equation would be way different if, you know, Amibroker was $10,000 or $20,000 a year or something like that. I could see. And it's why people are like, is Bloomberg doomed? I'm like, I don't think Bloomberg is doomed, but Bloomberg might have to lower its prices because I think Bloomberg is, like, $50,000 a year or something per terminal. Right?
Michael:And the I we're seeing, like, all the time on Twitter, it's like, I created recreated the Bloomberg terminal. I'm like, hey. No. You didn't because you don't have all the data import information that it
Dave:does. Valuable.
Michael:Yeah. And then also Bloomberg has a a massive connection. Like, I don't you probably never use, but the whole part of the whole point of Bloomberg is you can just communicate via chat with everyone else with the Bloomberg terminal as well. You're right. There's a lot of people who are like, yeah.
Michael:I'm gonna I'm gonna build this and it's dead. And that's just gonna be a math equation at some point. Right? Is it easier for me to spend a $100 or or I think about $300 every couple years to Amibroker and just give him the money versus my time even if it's the AI doing it? Because you're right.
Michael:There'll be a lot of, like, bug testing and and trying to go through and and trying to figure it out. And then is it as efficient as Amibroker? And then, you know, do these plug ins work and and all of that. So, yeah, just think, yes, you can build any piece of software in the world right now, but it doesn't mean you should. Right?
Michael:Yeah. If the software is cheap, you should probably just buy it and let the system use it.
Dave:Yeah. I was I was telling my wife the other day, feel I feel like I'm a CTO again. Right? I'm that's sort of what I'm doing. And there's a lot of decisions you make as a CTO to, you know, stay out of the rabbit holes that we're talking about here.
Dave:Don't build your own backtest or make the smarter decision that is gonna start giving you answers for your strategies real soon rather than endless rabbit holes of creating a back tester that you you hadn't even thought of yet that are are gonna end up taking way more time than you thought.
Michael:Yeah. And, you know, it like so right now for me to spin up like an iOS app if I had an idea is gonna take me probably a day. And everyone's saying, oh, it's gonna be amazing. Everyone's gonna build whatever app they want. No.
Michael:Like, if I'm looking for like a calorie tracker or something, I'm just gonna go on the App Store. I'm gonna find one for 99¢ and hit a button to move and move on with my life. Like, for that debate will always happen, the build versus buy debate where even if it's just came to my head. It might be a good example or might not. I don't know if you ever played with, like, Napster back in the day and all of that.
Michael:And how Apple killed for any kids that are watching, you used have to go buy these round things and put them in a CD player. And if you wanted to make CDs, you had to put them into your computer and download all the songs and and put them in. And it was free. It was it was piracy, but it was free, and everyone did it. And it was fine.
Michael:But what ended up killing it was Apple coming out, and you were able to buy a song for 99¢. Just a single song anytime you want it. So they created something that you had to pay for that everyone was doing for free, but then everybody switched over instantly because the amount of time it would take me to go on Napster to download a song, to see if it's, like, the real song or something, like, illegal I just downloaded on my computer or a virus or something. Who knows? And then all of the process to get it into a CD that I could put into my car and and feel cool when I was driving to high school, it it was way better to hit the 99¢ button and put it on.
Michael:So sometimes, paying the money for something is still way better than doing the thing for free because it's just in the long run, it's it's the convenience that you're paying for. And so same with building your own back tester. It's like, yeah, you could, or you could just pay someone a couple $100 for the convenience of using theirs.
Dave:Yeah. Alright. So to finish up here, why don't we talk about your next steps for your process that you're working on?
Michael:Yeah, and very quickly, just because there was the question the OpenCLO versus Cloud Code, I think we hinted that, so I just want to There was nothing that I could find I think I think I already talked about this a little bit that Open Claw was dumber and less obedient. Like, it would continuously stop doing things. And then as soon as ClaudeCode built its own scheduling function, there was just nothing that I could come across. The ability to use the web browser, ClaudeCode does way better than OpenClaw. I don't know if it's the way I set it up or the brains of it or something.
Michael:There was just nothing that one could do that the other one couldn't. And I don't trust any of these guys, but I'm gonna trust Anthropic a little bit more than, like, one guy who just built this thing, who I guess has now gotten poached by OpenAI. He's working for OpenAI now. So is his pet project gonna just die entirely? Because now he's if they poached him, he is making a crap ton of money over there.
Michael:Like, he's probably gotten a $1,020,000,000 dollar salary plus bonus. So he's probably not gonna be paying attention to it anymore. So those are main the main reasons that this is a a bigger company that made a thing. There was nothing that I looked at, and I say, have to have this kind of one man operation software ability to poke around my computer anymore. I just don't need it.
Michael:So that was that was the main reason.
Dave:That's interesting. Alright. So so what are what are the next steps, Re? What's the next on your to do list for this process?
Michael:Well, so one of the big things that it's it's already done is I'm now going to move to testing of this opening range prop firm kind of strategy that I have. And I like this because this is the first, I would say, one to one human AI strategy I've taken out there. So the reason I pointed that at the prop firm is, like, we talked about, it's like a $120 for a 25 k account. And if it blows up, that's like a two trips to Starbucks with kids. Like, it's a it's a negligible amount of money that completely evaporates.
Michael:So that's kinda step that's immediate. I'm I'm planning on starting that kinda next week, and I'm gonna do a series on my channel and everything about that. But and then the the next step that I've been doing a lot of is essentially giving it larger, broader tasks. So I talked about this before. A lot of how I interact with it is I will go out for a walk and just vomit information into my phone with the the text or the speech to text.
Michael:Now I actually get Claude to format it, then I read it to make sure it makes sense, that I wasn't screaming at the dog because it was barking another dog halfway through the walk or whatever. And then I do that. But the next thing is I'm kind of expanding it out because I realize how good it is at creating its own to do list and then working its way through it. So everything I think going forward is going to be like a week long project in which on Monday, I'm going to want to do this. And on Tuesday so I'm gonna be more of I guess this would be like a project manager role at a company.
Michael:I want to lay out an end to end, and I'm thinking maybe that unusual whales API thing because that's something that I'd never be able to do alone. But coming up with these piecemeal chunks and how I've decided to do it is I'm going to have a Google document, which I constantly keep update, and a scheduled task for when I go to bed that it goes and reads the Google document, and any new piece of information that's in there is its next task. And that way, I have one point of kind of interface with it where every throughout the day as I think of things, I'm just gonna add them to this Google Doc. And then every night, it's just going to go and it's gonna read them, and then it's gonna go out and end up implementing that throughout the night and then give me a report the next day. Trying to get better at this kind of hand off mentality of I work during the day, and then you work at night.
Michael:Yeah. And and then this kind of so all day long, I'll be verifying what it's doing and coming up with new ideas and new strategies and everything I wanna do. And then all night, it will be doing all the grunt work, and then it a scheduled task to stop, I think, an hour before I wake up and then prepare a report and a nice kind of summary of what it's done for me overnight. And just I think that alone, that cycle will really increase efficiency with a lot of this.
Dave:Yeah. We I used to call that follow the sun. So I was I used to trade with a guy years ago who lived in Thailand. So he was exactly twelve hours difference. So I would work on some stuff, share some ideas with him.
Dave:He would work on some stuff while was sleeping. So sort of the same kind of thing. You're you're following the sun. And, yeah, that's really cool.
Michael:Yeah. And I that's again, when you get, it's worthwhile even in the crazy prices of of computers today. I think it's worthwhile for anyone that's listening and they're like, what's the first step? Is to maybe get one of these mini PCs and just put the what you like, Claude code and the exact software you need. Nothing else if you're worried about it.
Michael:You can get these for, like, $500 ish, $5,600 with enough RAM to run some basic things and and all of that. Start over there. It's like the the big craze when OpenClaw came out, the Mac minis were just flying off the shelves because they're really powerful computers for for very, very cheap. And then the more important thing is it puts away your fear of of that and just don't give it access. Right?
Michael:Don't sign in to any of your stuff. Don't sign into your browsers or your bank, especially, or any of these things. And then start a process like this. And even if it's basic, just kinda get into it and say, okay. Tonight, I want you to do this.
Michael:And then go to bed and and wake up and see if you're surprised with with what it's done for you. I think that's the easiest way to get in because you're talking, you know, if you're if you're paying for the more expensive one, that's like a $150 a month for Claude code, then a $500 computer. So for less than a thousand bucks for the entire year, you are you're like set up and you're good to go and you could start this process. So I think that'd be kind of phase one. You know, from there, just just keep going.
Michael:Like, don't we're we're probably gonna meet together in a couple months to talk about this. And I don't know. It'll be, like, beaming things into my brain or reading my dreams or some point while I'm sleeping, but it's super exciting.
Dave:Yeah. So let me give one suggestion for getting started for the developers in the crowd as the last thing before we wrap up here. One thing that you should be doing as a developer is doing code reviews. You're probably not doing them as thoroughly as you could. Claude is really, really good at code reviews.
Dave:And, like, I've got some automation set up in my GitHub repos where when I do a pull request and and commit code, Claude is automatically coming in and examining it and and doing a very detailed code review, catching bugs that I hadn't thought of. So and that's all, you know, read only. It's just pointing out things that you weren't doing. And it's a really valuable way to get started because you can see, wow, kid, this is really quite good. And it's it's can help you out, help you fix bugs before they get into your code base.
Michael:Yeah. And one one more thing. This will be it. I promise. There'll be but we keep talking about Claude here, and I I'm just trying to think for people listening to this in six months to a year, Claude might be the dumbest one at that point.
Michael:I've gone from OpenAI to I tried the Google one, and there might be something else that kinda pops up. So I guess anytime we're saying Claude, just insert whatever the smartest AI is at the time. It's one thing that's gonna make it hard for these companies is and I'm gonna get your feedback on this before we go to is I am so agnostic on which one I use. I just could not care. What I did when it came to when I switched from ChatGPT to Claude, I talked with his little I had him dig his own grave.
Michael:I said, okay. I'm canceling your plan. I want you to write me, and I think I sold I said, like, a 20 page document on everything you've learned about me over the last year or so I've been using you So that I can take that and I can give it to your new AI and it can learn everything about because I was having it help with my business and and write emails for me and and, you know, a lot of stuff. It was learned how I spoke so that it could then help not sound like an AI and actually take what I'm thinking and and do things with it. And yeah.
Michael:So in the future, if you're seeing this and maybe Claude is is really, really dumb and x y z company has built, like, the smartest AI out there, Just use whatever it is and always come up with a plan to be able to switch from one to the other. Because in the short time that these existed, I have seen, right, one overtake the other over and over again, and it's the beauty of of capitalism and competition. I expect that to continue. So six months from now, I don't know. Anyone probably not Grock.
Michael:That might be, like, a really bad probably won't be Grock, but it'll probably be one of the other ones that end up coming up and being the being the smart guy.
Dave:Yeah. I think the right way to think about it is, you know, even if you're skeptical about AIAN, I'm sure there are I know there are people that are still skeptical about it. It's literally the worst it's ever gonna be right now. Like, it's only getting better and it's getting better dramatic rate. So you need to figure out how to work it into your life and what and what you're doing.
Michael:Yeah. And it's one of those if you're if you're not, everyone else is. So, you know, you can be one of those people that stick to your, oh, I'm not gonna adopt Excel. I'm gonna write down all these numbers and add them up by hand. It's like, okay, good for you.
Michael:Everyone else is gonna go use Excel and it's gonna be a problem. And it's the same with this, that people are going to be coding cooler things at at faster rates, and whether or not the whole doomsday replaces us all comes to pass. Who knows? But in the meantime, there's a whole bunch of super powered people out there. And if you're not one of them, again, I just think you're you're leaving money on the table at this point.
Michael:But as always, I'm Michael Nauss.
Dave:And I'm Dave Mabe. Talk to you next week on Line
Michael:Your
Dave:Own Pockets.
Creators and Guests
