I Built an AI Agent That Follows Polymarket Smart Money transcript

I built an AI agent that watches Polymarket's smartest traders and decides whether their trades are actually worth copying or not. Not by guessing, not by asking ChatGPT if Bitcoin is going up. The agent looks at smart money wallets, finds what they're buying, checks whether I am late or not, and then gives me one of three verdicts: copy candidate, watchlist, or skip. And the whole question of this video is very simple.

If someone on Polymarket is already making money, can an AI agent help me follow them intelligently? Or is this just a long way of losing money? So, today I'm going to give Claude one job, design the system, build the logic, and help me run a 7-day smart money trading experiment. The goal is not to build a magic trading bot. The goal is to build something more useful. An agent that stops me from blindly copying whales just because a big wallet made a trade.

By the end, I want to know just one thing. Can an AI agent reliably make me money in 7 days? Let's build it. Before we touch anything, we need rules because the moment you plug an AI agent with a trading wallet, a lot of bad things can happen. Rule number one, the agent is allowed to trade. In fact, I will not be doing anything manually. It is completely its job to make us money in this challenge. The only manual thing I'll be doing is prompting it, of course.

Rule number two, every single trade needs a reason and the whale bought it is not a valid reason. The agent has to be able to explain to us its reasoning or its reasoning for copying it and whether it thinks that trade is copyable or not. And rule number three, the agent has to say skip more than it copies. If it says copy on everything, it's not a smart agent. It's just a Vegas slot machine. For this build, I'm going to be using Claude code for system design.

I wanted to think through the architecture before it actually builds anything. And that starts with the most important question. Who are we even following or copying? And for the prompt creation, I just asked AI to create the perfect prompt for AI. Let me show you. It's It's quite long, so just brace yourselves. This is the entire prompt right here. I'll go through it bit by bit, so that in case you want to copy it, you can.

But these are the main points right here, which we'll go through with point A being wallet discovery engine. We want it to be able to discover the wallets we want to trade. Trade detection engine. We need for it to have a way to find the trades. Copyability scoring rubric, which is whether this trade is copyable or not. Uh D is risk management rules, just basic risk management. E is a dashboard. I want to be able to see everything that's happening.

I want a dashboard, which I highly recommend, by the way. And this is just a bunch of stuff for the dashboard. F output files. This is just for it to be able to keep track of everything. This is sort of its hardware. It's data. G daily workflow. So, a morning scan, a decision review, and an evening review, so that it can improve and get better on its own throughout the challenge. And H final deliverable. When finished, give me all of these.

I'm going to go ahead, paste it, and boom! Enter! Just like that. Let's see how long this is going to take it. I'm super excited for this, and see how it does overall. Is it going to make me money? By the way, one quick thing, if you notice in the prompt, I use the Bullpen CLI, which is uh a great way to find Polymarket data to use uh AI agents with Polymarket data. This is not a sponsored video. We've worked together in the past before, but this specific video is not sponsored.

It is genuinely how I use Polymarket myself, but you can go ahead and connect directly to Polymarket. You can ask your AI agent, you know, how can I connect to Polymarket directly? I also have a video, by the way, of how to connect to Polymarket directly, if you want to watch that. It's on my channel. So, in case you don't want to install Bullpen, you can still do it that way. Anyway, I'll see you when uh the dashboard when the actual AI agent is ready. >> All right, we are finally live.

It's done and finally running right now. This is what it gave us, the dashboard. Uh it's live. It hasn't spent anything today. The way it's going to work is it's going to, relative to the traders bet, going to place the bet. So, if a trader places a million dollars, I don't have a million dollars in my Bullpen account. Best I could do was $200. So, it's going to place that relative to our uh wallet size. And yeah, it's going to score bets on either skip or not skip or put in the watch list or actually bet.

And we can see here the smart wallet leaderboard. Here, it placed them as clean signal. So, these are wallets it's going to copy or this is a wallet it's going to copy. This one it's flagged as non-copyable or not super good to It's not going to copy. I'm not sure why. I'm going to go ahead and ask it a bunch of questions uh really understand, you know, why does it say 87% copyable, great copyability, but label as not copyable.

So, that's what I'm going to do, but for now, it's live. If it sees something it wants to bet on, it's going to bet on it. There's nothing stopping it, as you can see. Uh auto copy armed. All right. >> I just woke up and I noticed that we spent the entire night. Let me turn over my camera. Not placing any trades. You can see we we've scored over 200 something trades. We've skipped 215. And I mean, we just literally placed the first trade right now.

And the reason being was the filter was way too high. It was way too strict. And so, I just fixed it. Today is Sunday. I'm not going to be here, home. So, I'm just going to have to let it go alone in the same parameters it has currently. Uh hoping that it places more trades and more importantly, winning trades. I'll see you on day two. And it's day two here. Let me show you where we're at. We are $15 in profit today.

And these are some of the bets that it's making. A lot of them I don't think they show here, but a lot of them were opened today and they resolve tomorrow or the day after. So, these will stay open, but you can see we're in profit on most of them, except for this one. But, where we are in profit, we're like $5 in profit, $4 in profit, $5 in profit. So on and so forth. I'm excited. This strategy seems to be working and I can't wait to see where we end up.

Let me show you. This is our mobile pen account and boom. Just today alone, we're $10 in profit. Let's see where we end up tomorrow. This is day three of our challenge and let me show you what I woke up to. The dashboard is showing $58 >> [laughter] >> in total loss. Which sent me into a panic. I was like, "Oh my god. What happened?" Then I went to uh see my balance and we are negative $20 today in the last 24 hours.

Which isn't as bad. I I mean, $20 is more tolerable than 50 or 60 dollars, whatever the dashboard is showing. So, I have a few things to fix. Number one, uh the dashboard itself for it to display the actual numbers. And number two, check out why did we lose? Was it just a bad day? Is there anything in specific that we need to fix? And yeah, I mean, this loss just wiped away most of our gains that we made on day one and day two.

So, I'm hoping that for the remainder of this challenge, we're in profit and we make it back and more. Let's see. Only time will tell. This is day four and man, am I bummed out because in the last 24 hours, not a single trade has taken place. You can see, PNL is in the last 24 hours has not moved. I've just fixed it, so it just started placing trades again. And the reason for that was it was too restrictive. The changes that I made were way too restrictive to pass any of the trades.

Um but yeah, I just fixed it and we have three more days to the challenge, but hey, this is what it's all about. This is what building uh an an agent looks like. Especially one that trades, it's a lot of trial and error. It's a lot of fine-tuning, which is why uh I think it's important that I document this part of the journey as well. But see you on day five. Fingers crossed. Day five here and we are finally in profit.

Look at this. Two whole dollars, almost three dollars. Oh my god, I think I might be able to retire after this video. >> [laughter] >> But on a more serious note, looking into why we're two dollars or three dollars in profit, it seems a lot of the bets it is choosing to make a are World Cup related. Will Spain win the 2026 FIFA World Cup? Uh it made a bet that France would win yesterday, which it did. So, that was one of our major wins.

It's making a lot of football bets and then it's making the other 50% is a lot of war bets. Um I don't know if I necessarily agree with its predictions. Like if I were to personally place money on these different bets, I would pick the completely different outcome, but this isn't about me choosing, you know, or my performance as a a prediction better, prediction market better. It is more about can this bot reliably make us money.

So, now we're in profit. We're back in profit. For some reason it's very adamant on keeping the all-time at this amount. I've asked it why is it adamant on doing so even though on Bullpen itself I can literally see that we are not this much out of loss. If anything we are at break even. We're $3 in profit right now. So I don't know. It's just adamant on keeping it up. So I'll just use Bullpen to verify the final amount which makes sense.

Um but yeah, I'm excited to see we still have day six and day seven. This is its chance to just run it up. Day six here and we are $5.50 in profit. It's not a life-changing amount so far. Uh we've had much stronger performance honestly in our some of our other challenges, but hey, I mean we're in profit. I'll take it long as we're not in a loss. However, I will say it's what I'm noticing and the biggest takeaway is that one of the main reasons for us not being like this high in profit is a lot of the bets we're copying and it seems that when we first run this prompt through Quad Codes Fable it decided to not copy trades that are time-sensitive.

It's only copying trades that resolve in days or weeks from now. So those are super slow trades that just lock up our liquidity. You can see like a lot of the trades it's placed is by June 30th. For context, it's June uh 17th today. Strait of Hormuz, July 15th. So you can see like a lot of the trades are month away if not weeks away. So that being said, what I'm planning to do today, I'm going to close out all the trades because they're holding up the liquidity.

We have $100 spread through trades that aren't going to close anytime soon. I'm going to ask it to make more recent trades, things that will close either today or tomorrow cuz I want to finish this on a strong note or lose everything. I don't know. Let's see what happens and I'll see you on the final day, day seven. And it is finally day seven here and what a way to wrap up. I think our decision yesterday really paid off, which is to bet on markets that are going to close out either today or the day after, so in the next 24 hours.

And honestly, I wish we did this earlier at the very start of the challenge, but oh well, this is what we are left with. We're $36 in profit over the last 24 hours. And all time, uh it's not accurate. It says $45, but it's closer to 40, 41. Uh I'm not sure why it's not pinning down the actual number or I don't know. I've asked it to fix it multiple times and it doesn't want to. So, yeah, overall, one massive learning uh from this challenge is markets that resolve sooner are way better unless you don't have a problem waiting a month for a market resolution, which for our amount of capital, I mean, I started this challenge with only about 150 or 200 dollars.

We don't have enough to just stay spread for an entire month and just wait out uh for these bets to resolve. But yeah, that's it. I hope this helps in your journey as well. After seven days, here is what happened. Here are the results. I just made the math and this is what we did in seven days total. The agent tracked a total of 39 wallets. It found 1,872 trades. So, it decided whether to trade or not 1,872 trades. It marked 386 trades as copy candidates, 461 as watch list, and 1,002 as skips.

The most useful part was how often it said no. The fact that it skipped over over way over half of the trades is what gave us somewhat of an edge to some other strategies here. Looking at the data, by the way, our biggest loss and the literally the thing that set us back was a $20 bet we made on an e-sports match that was going well, but looking at the data literally at the last few minutes, the game switched courses and the team we had bet on or we had copied the whale to bet on uh had lost literally in the final minutes.

You can see the the chart going down. >> [snorts] >> Uh so, that was our biggest loss. Literally 80% of everything we lost was that one flipping bet. Uh which was the major loss on day three. But, all in all, day one we made $0 cuz the bot wasn't placing trades. I had failed to to set it up properly or to make sure that it works before I log off for the day. Day two, we made $15.04. It was a good day. Day three was the big damage.

It was negative $20.91. Day four, we put so many restrictions after those losses that we made $0. The bot made no trades. It had way too many restrictions. Day five, we made $2.7. It was life-changing. I thought of retiring on day five. Day six, we made $5.51, which is not bad. Uh and day seven was our home run, our best day of the challenge, which we made $36.2. All in all, we were $38.9 in profit. Which doesn't sound life-changing, I know, but given our capital, that is 20% return on investment.

So, I would consider it that uh success. Overall, the number one lesson that I learned from this particular challenge was narrow down your trading window. Don't let it place trades that are months ahead of resolution. Um, I mean, if you have all the money in the world, maybe feel free to do that. But, if your capital is narrow, uh which I recommend for everyone to start with very small capital, prove that the strategy works, and then commit as you go further, cuz if it doesn't work, you're just going to lose a lot of capital.

It's going to suck. And it's honestly not worth it. Prove the method works, and then double down on capital. You can do something, by the way, pro tip. Uh, you can do something, which is paper trade, essentially. You can ask your AI model or AI agent, tell it, "I want to paper trade first." Which means it wouldn't use real money until you narrow down a strategy, and then it can start using real money once you're confident enough.

Unluckily for us, uh especially by day three, we weren't paper trading, we were real trading. But, luckily, we ended up uh switching the tides or turning the tides. But, learn from that mistake, and start out paper trading, and then move on to real money only once you're confident enough in your strategy. By the way, if you want me to turn this workflow into a Hermes agent workflow that you can control from your phone or anywhere in the world, and have that bot or AI agent improve with time on its own.

Comment smart money and I'll build that next. If you enjoyed this video, make sure to leave a thumbs up. And if you're new to my channel, make sure to subscribe because I have a ton more content just like this on my channel and coming out. And oh, would you look at that? Over here. The algorithm gods seem to think that you'll really enjoy this video next. So, I'll see you there.