AI TRADING: How To Code A Successful Crypto Trading Bot

Ajouté : 2026-05-09

[00:00:00]I cannot code yet. Yesterday, I built a bot that  runs 24/7 on one of Hyperliquid's most inefficient markets and continuously makes trades all day  autonomously. This video is for information and entertainment only. It reflects my personal  opinions and is not financial advice. Check the description for more information. Always do your  own research. All right, guys. Before we get into the actual bot and how it works, I'm just going to  quickly go through the process, right? Because the process is probably the most important thing. Now  I can't code but the fact that I saw this market had a discrepancy in it meant that now with the  incredible advent of AI, I can just go and talk stupid to my computer. I can literally talk  like an idiot to it and it can understand me, which is mental. If you want me to go through a  full guide as to how to exactly set this stuff up, like how to do it in Claude code and stuff, just  leave a comment and I can probably do that full walkthrough. It would be a much longer show, it  would probably take 30 maybe even 40 minutes to do the full walkthrough. I am just going to cover  this in maybe six or seven minutes. So hopefully this show isn't too long. This is the process.  I saw the market launch. I watched it for a day or two and I saw that it wasn't really moving  in line with what I would expect. After that, I got a hypothesis and then I used Claude to go  and do the research for this hypothesis and then I got Claude to go and build the bot. This was  all done in natural language. Once I had built the bot, I just tested it, iterated with it, and  now it's running in the background. It is kind of wild how effective it has been. And it was only  because of this. These markets when they launch, any market across all of DeFi is inefficient  at the start, very inefficient. And this is why Hyperliquid has been so successful, because it  has garnered so much liquidity. And now it is so efficient, people can use it as their legitimate  trading platform. It is getting recognition at Bloomberg and whatever else because it has deep  liquidity. But these other markets, these newer markets on Hyperliquid are not very efficient.  And I am going to show you again exactly why that is. But here are some things to look out for.  If the open interest is lowish, it is probably quite inefficient. If there is not enormous daily  volume, it is probably relatively inefficient. If there is no volume, then there is no gap to be  made. There is no point trading in a market that has no volume because if you are trading, you  would be trading against yourself, there would be no edge. And prediction bets, which is what  Hyperliquid's new HIP-4 is for, they are just outcome bets. So you are betting on the outcome  of something happening. Same as Polymarket, same as Kalshi, etc. Here is an example. Will Bitcoin  be above or below 79K tomorrow? If there is a 65% chance of yes, there is a 35% chance of no. That  is literally how you buy them. 65 cents will buy you one share of yes, 35 cents will buy one share  of no. And of course, if the no share ends up being right, that 35 cents is now worth a dollar.  And vice versa, if that yes share is right, that 65 cents is now worth a dollar. And of course, if  that yes share is wrong, then that 65 cents is now worth no dollars. Okay? So, it is probably easier  just to show you outright what I am talking about.

[00:03:01]I am looking at two markets here really because  it is the two markets interacting that I think is where I saw the edge and I think these kinds  of edges are going to be quite common actually, which is what I mentioned earlier along these  HIP-4 markets. And so all I am looking at here is Bitcoin price and whether or not Bitcoin is  going to be higher or lower tomorrow. What you can see is this on the 5-minute time frame. Obviously  Bitcoin is volatile, we know Bitcoin is volatile.

[00:03:31]But Bitcoin is volatile all the time. So when  it goes up $100, down $100, it is just as likely to go up $200 and down $200 in the next hour.  But this 24-hour market is really volatile. It reacts a lot. At just 8:00 a.m. this morning,  there was a 55% chance Bitcoin was going to be higher tomorrow. And then by 1:30, it was  only a 30% chance. And then it swung back to 50. And now it is down to 40. But you can see if  we take this as the 50% line in the middle here, it spends some time above the line, spends some  time below the line, converges back towards the line at some point. This is pretty common. This is  what you would expect. But the market isn't really pricing it such that it expects it to revert  to the mean very well, which is why there is an inefficiency or why I thought there was an  inefficiency. So I looked into it. I thought, okay, is this just me not understanding Bitcoin's  actual volatility? What we needed to find out was, was what I was seeing actually real? Because it  looked like when Bitcoin went up like $100, $200, the percentage chance of it being up or down  the next day was moving too much. But it is hard to quantify that. So when I went away and  did the research on this, what I found out is that if Bitcoin moves $100 or $200 or $500, those  moves are so typical that any move of that amount in that direction doesn't really tell you where  Bitcoin is going to be at the end of the day. It is not much more than a coin flip really. And  so it was quite clear that these HIP-4 markets or this up-down market was overpricing early  moves. It was indexing a $100 move in Bitcoin, a $200 move in Bitcoin as increasing the chance  that Bitcoin was going to be higher tomorrow by way more than it should have been. It should  have been still pretty close to a 50/50 split, but really it was more like a 60/40 split once  Bitcoin had moved a few hundred, which was clearly way too high. And this data just shows you again  that basically movements inside the first 4 hours are largely noise. They need to be really really  big for it to be actual informative information, which makes sense because 24 hours is a long  time. What happens in the first 4 hours doesn't really predict what is going to happen in the  next 20. So it kind of makes sense. And this actually meant that a lot of the times the prices  that were coming through from those markets were not correct. The market might be saying there is  a 65% chance, but really the research was saying it was more like a 55 to 60% chance. So there  was a small amount of edge in that bet. And so I went away and built the bot again just in  natural language. And all I said to Claude was, I gave it the research and said, "Okay, now go and  build this bot." And it comes back and it is like, "Okay, let's run this thing every 15 seconds.  What is Bitcoin doing? What is the market doing?

[00:06:30]What is the percentage chance that it is up or  down? What is the Bitcoin price?" And most of the time it is just doing nothing. And every now  and then it realizes there is a 5 cent maybe kind of position that is not right and it enters that  position and then it closes it once it is back to the right level. Now sometimes it stays at the  wrong level for a while. These trades sometimes go underwater. There are still stop losses and all  that sort of stuff built in. So sometimes the bot still suffers losses but fundamentally the bot  made 50% inside the first day. Okay. And this is what I want you to take away from this. I am not  saying you should go and make this bot, right?

[00:07:06]There are a few things that I am actually saying.  First thing actually is that these Hyperliquid Improvement Protocol, these HIP-4 markets, there  are going to be lots lots more of them. They are going to start very inefficient. Now, this one is  more efficient today than it was yesterday. So, the edge is in being early. And these markets are  being spun up all the time. Now, there are two of these markets. Probably in a month's time, there  is going to be a handful. In 6 months there might be a bunch of relatively inefficient markets and  not only that but the fees at the moment are free, so it means obviously it is much cheaper to trade  on these things. And the third thing is that there is still a chance Hyperliquid does some form of  airdrop to users. They still have like 30% of the supply that is earmarked for the community, future  rewards, etc. Obviously no guarantees on anything, but those tokens are either going to be burnt  or distributed in some way. And if they are distributed in some way, you would imagine  Hyperliquid gives them to users of the platform.

[00:08:05]And I would probably bet just by nature of wanting  to decentralize the network further, they probably are more likely to give it to users of innovative  products rather than just the perps users. Again, they probably do give a disproportionate amount  to people on the HyperEVM, for example, people using the HIP-3 market, people using the HIP-4  markets. And if we just look to it one more time, like Bitcoin's traded in like a 1.6% range, yet  this HIP-4 market has traded in a 109% range. And so there is just an enormous amount of variance  in these markets. And I think learning how to go away and capture the inefficiencies is a really  invaluable tool. And so the point is building bots now is incredibly easy. People can just describe  what they want in natural language. They can then iterate and go back and forth and then run the  thing just on their local machine. They haven't even necessarily got to get a VPS. As I said,  I can do a full walkthrough of how to do all of this. But the most important thing I think is  not copying and pasting one person's piece of code because especially in these inefficient markets,  it gets outdated very very quickly. Iterations have to be done consistently and you have to be  quite manual, making sure the bot is being taken care of, overseeing the bot. If someone decides  to use an inefficient market and then just leaves that bot for a day or two, there is a good chance  when they come back, they are going to have no money left because the bot needs surveillance. It  needs to, it can have failsafes and stuff built in, but it certainly should have failsafes built  into it, but it does need consistent surveillance.

[00:10:04]So, before I go guys, I just want to say today's  sponsor is MetaMask. So, MetaMask are essentially building infrastructure for never needing to  come off crypto rails, which is awesome. So, they have got places to trade, integrations  to Polymarket and Hyperliquid. Obviously, these outcome markets are now there as well. You  can basically do everything you ever want to do on-chain with MetaMask. And if you are traveling  as well, MetaMask have this incredible card. I am going to do a full guided walkthrough of the  card actually, but this card gives you discounts and cashback at some of the most incredible  hotels all over the world. We are talking like 60% off the normal price. So, absolutely  incredible. Thank you so much to MetaMask for sponsoring the show. Thank you guys so much  for watching. I shall see you in the next one.