How to Trade Polymarket with AI in 2026 (FULL COURSE) transcript

Most people use poly market like a casino. They see a market, they have an opinion, they click yes or no, and then they call it trading. That is not how I would approach it. If I was starting from zero today, I would build a system. a system that helps me find good markets, research them faster, track smart money, compare probabilities, avoid bad liquidity, preview trades, and log every decision before I ever risk real money.

And in this video, I'm going to show you how I would build that system with AI. We're going to start from the basics. We're going to learn how to set up the tools that we'll need and then we'll build five different AI assisted polyarket strategies. We'll cover smart money tracking, crypto short window markets, weather markets, arbitrage correlated markets, and eventdriven news trading. Then I'll show you how to tell the difference between a good market and a bad market. how to keep trading agents up to date without letting them do well stupid things and how to build a dashboard so this becomes an actual workflow that you can track from start to finish instead of a oneoff demo.

Quick disclaimer before we get into it. This is not financial advice. None of this guarantees profit. Prediction markets are risky. AI can be wrong. And if you let an agent trade blindly, you're asking for problems. The goal here is not to let AI gamble for you. The goal is to build a better decision-making machine. For this video, I'm going to use Bullpen. Bullpen is sponsoring this video, but this is also a very natural sponsor for this topic because it's the tool that I personally use to do my polyarket trading and because it solves the exact annoying part of building polyarket workflows.

You can interact with Poly Market directly through APIs. I've done that before. It works, but it gets messy really quickly. >> [snorts] >> You have to deal with market discovery, slugs, outcomes, prices, wallets, positions, approvals, and the list goes on and on and on. And then you still need a clean way for an AI agent to use all of that safely. Bullpen gives you a command layer for Poly Market that is much easier for agents to work with.

You can search markets, check prices, inspect trades, look at leaderboards, track traders, preview orders, manage positions, and output data as JSON. So, an AI agent can actually reason over it. It basically makes all of the polyarket data easy to understand for your agent. That is why it fits so well into this video. I'm not going to position bullpen as some magic uh profit button because no trading setup can ever do that.

The value they have is infrastructure. If you want to use AI to research and trade prediction markets systematically, you need clean tools and bullpen gives us that foundation. So, let's start by setting it up together. Oh, by the way, I'll leave all the links that you will need down in the description below. See you in the next step. All right, let's start by installing the bullpen CLI first. These are the set of skills that will give your AI agent the ability to read markets, trade them, understand them, be able to explain them back to you in simple terms and just navigate uh the entire polyarket infrastructure very simply.

Now, for that, what I'm going to be doing for you to make it as simple as possible is include the link to a cheat sheet down in the description. So, you can have all of these commands I'm about to paste for the installation and even for the different strategies that we're going to be using down in that document. You'll be able to download it. That way, you can just copy, paste, drop it, and follow along this well tutorial.

So to install bullpen, we're going to paste this command brew install bulpenfi and we're going to just let it install. You'll need to open your terminal for that. Whether you're on a Mac or on uh Windows, just open your terminal and install bullpen. And just like that, boom, the installation is done. Simple as that. Then we will be installing the skills by typing in bullpen skill install. And this will make sure to install all of the the bullpen skills for different agents.

There you go. So installed 13 skill files. We have skills now for clot code, openclaw, codeex, and all of our different AI agent tools. And that's pretty much it. Now we're going to log in to bullpen. We're going to put in bullpen login. Now, all we're going to do, because that's literally it, we're going to type bullpen login and it's going to give us our login code. So, all you have to do is copy this URL or if you want to log in through your phone.

You can also scan the QR code here that you see. But since I'm on my desktop, I'm going to open a browser, take my login code, go back, paste it in, and then I'll log into my account. And boom, there it is. Your CLI session is handling setup. If trading doesn't work, tap below. Let's go and check. And account setup complete with notes. Waiting for login. Wallet key exported. account setup complete with notes and boom, we should be logged in.

And in order to check that you are in fact logged in, just type in bullpen status. And once you type that in, it should show you whether everything is working or not. And there we have it. And here we can see we are logged in through our email. And here bullpen proxy enabled for buy, sell, limit buy, limit sell. We're set up. That's as simple as it is. Now, what you'll also want to do is obviously create a bullpen account before this by visiting the link to bullpen.

You can open it through my referral link as well um and load up bullpen with money. Or you could also, while I'm showing you the strategies, ask the AI to use paper trading, which would mean that it is testing out with fake money, not real money, uh the strategy, so that you can see what strategy do you feel good with, what strategy do you feel like you have an edge with, which strategy do you feel comfortable and you feel like you understand best?

Because really, I'm going to be showing you five strategies. I don't recommend you go out and do all five at once. I recommend you start out with one that you really like, one that you find an edge with. And you can always start by paper trading with paper money, not real money. But that being said, what we're going to do next, we're going to paste in this prompt, which is bullpen polyarket discover sort volume. We're going to make sure this will give us all the different polyarket markets, just the first few sorted by the top volume.

Uh, this will make sure that it can actually read the data. And yeah, there you go. It just listed a bunch of different polyarket markets. And now that you have this installed, you can even search specific markets. You can type in something like bullpen poly market search in between brackets Bitcoin so that it can search for Bitcoin markets. And here you go. It just gave us a bunch of different markets. It gave us a bunch of different polyarket markets and it even gave us uh a list of traders with Bitcoin in their name.

Now if I have positions open, I could also type in bullpen polyarket positions. And this is a very helpful one to know because it'll show you all of your active positions. No active positions at the moment. We haven't deployed anything yet. But if you had active positions, you can literally see them by typing in this one prompt, which is super easy to use, super helpful. You don't need to open anything. You can literally be talking to your AI agent and asking them this, right?

You don't even need technically to be telling it any of this. You don't need to be putting in these specific prompts in your terminal at this point. You can basically talk to whatever AI agent that you use. Whether it's Hermes agent, whether it's openclaw, whether it's clot code, codeex, whatever AI agent infrastructure that you use, you [snorts] can talk to it in plain English and ask it for different things because the CLI skill that we just installed, Bullpen's CLI skill, gives your agent the ability to read Polyarket data in a very easy way and in a way that it can just translate all of these letters and numbers and simple terms. terms for you.

So, whatever you end up asking it, you could get an answer to in plain English. That being said, if you want to install it through clot code, through codeex, through Hermes, through openclaw, you can ask your AI agent. You can give them the link for bullpen and tell them, I want to install this tool so that we could trade Poly Markets. Could you please install it? and have it go through the installation on its own and then log you in on its own so that it can handle the rest.

Okay, that being said, I'm going to show you one last command in the terminal that is super helpful in case you're using the terminal. The prompt is the following or the command is the following. Bullpen polyarket buy. Then you enter the slug of the market which if you see the markets it's listed here. These are the slugs. They're always listed. So you can include the market that you want. Say let's go with this one.

For example, will Bitcoin reach 85K in May? So you include the slug. Yes. 10 dash preview. What this will do is it will preview the order for you and show you the different things that you might want to see. Preview mode lets you see what would happen without actually executing the trade. That is especially important when AI is involved. If you remember one thing from this setup section, remember this. Your agent should research, summarize, rank, and preview.

It [snorts] should not execute unless you explicitly confirm and you explicitly agree with it on a strategy. Now, for the rest of this tutorial, I will be using clawed code to set up our poly market trading strategies, but you can literally be copy pasting the same exact prompts I'm about to use and show you. And the prompts I'll include in the cheat sheet down below for Hermes agent, OpenClaw, Codeex, whatever AI agent infrastructure that you personally use.

I don't want you to go ahead and download a new AI agent infrastructure for this setup. Unless that's what you want. What I want you to do is instead use it with the AI agent setup that you currently have. Whether it's Hermes, openclaw, codeex, clot code, use that. Exactly. What I'm going to be using is clot code. And what I'll be showcasing to you here today is me using it with clot code. But again, it works the same with any other agent infrastructure.

So what I'm going to be doing in order to use clot code here is open a new window in my terminal. Drag it here. zoom in so you guys can see a thing. Expand it. And here's what we're going to do. We're going to go claude slash skip actually dangerously skip permissions. That way it doesn't keep asking us, do you approve of this? Do you approve of that? This is because I'm a DGEN, of course. And just like that, we have clot code open on Opus 4.7 with X high effort.

And here's what I'm going to paste to it. You are my bullpen setup assistant. Help me verify bullpen is installed, logged in, connected to Poly Market, and able to run read only market discovery commands. Do not execute any trade. if off balance, CLI or API checks fail, stop and tell me exactly what failed and I'm going to send it to it. And that's literally the last thing we're going to do is make sure our AI agent can operate bullpen.

And once we make sure of that, then we literally get started on our trading strategies. And that's what it's doing right now. It's verifying, I'll verify your bullpen setup with readonly checks. O and poly market connectivity look good. Now running a readonly market discovery and balance check. All checks passed. CLI is installed. We're logged in. We have our O tokens. They're valid. Poly market address. Market discovery.

Our balance. So I loaded up $200 on my account. The skills are installed. And we have the latest version of the bullpen CLI skill. Heads up. Non-blocking. GW2A expires in 3 minutes. Multiwallet hint. You have three wallets. If you ever wonder where funds are. Salana not enabled. Uh I can do this if I want to enable Salana trading. You're cleared for readonly research and when you ask trading. So we're ready. We're ready to trade.

Question is are you ready to get started? Let's do it. Let's start with our first trading strategy. The first strategy that we're going to be covering is copy trading smart money. And honestly, this is one of my favorite strategies for beginners because it is probably the easiest one to set up. You do not need to become a macro expert on day one. You don't need to understand every election market, every crypto chart, every weather model, or every court case.

You start by finding people who are already good at specific poly market categories. Then you use AI to track them, explain what they are doing, and help you decide which trades are actually worth copying. That is the important distinction. This is copy trading. But it's not blind copy trading. Blind copy trading is this wallet bought. Yes. So I buy. Yes. Smart copy trading is this wallet has a real track record. This position is meaningful.

The market still has enough liquidity. The spread is not terrible. The timing still makes sense. And I understand why I'm copying it. That is why I like this strategy so much for beginners. It gives you a great and simple starting point. Instead of opening Poly Market and staring at hundreds of markets with no idea where to begin, you can tell Cloth Code, "Find me the smart money, [snorts] show me what they're buying, filter out the junk, and explain which trades are actually copyable." So, let's get into it.

So, first I'm going to be pasting this prompt right here, which I'll read out to you. Again, you can paste this on any AI agent infrastructure. I'm doing it on claw code but you can do it on Hermes openclaw codeex whatever it is that you use and this is the prompt that I'm pasting which again reminder will be included in the cheat sheet that you can just download and copy paste yourself you are my poly market copy trading assistant inside cloth code use bullpen if available but do not show me raw terminal commands as the main demo run read only research first never expose private keys session tokens wallet secrets, so on and so forth.

For strategy one, build a beginnerfriendly smart money copy trading report. Find top polyarket traders, recent large trades, and any followed trader activity. Identify which wallets appear worth copying, what categories they seem strong in, what they bought, position size, entry timing, so on and so forth, and whether the trade is still copyable. Right now, do not tell me to blindly copy anyone. Rank each opportunity as copy candidate, watch list, or skip.

For every copy candidate, explain the thesis in plain English, the risk, what would invalidate it, and whether I should preview the trade before doing anything. And we've pasted this and it started using the bullpen uh CLI skills, pulling smart money signals, leaderboard, so on and so forth. Initial data is in the copyable leaderboard is surfacing junk. Small losing traders. So I'll pull a P&L sorted leaderboard and live market context and it's continuing.

Great. Your track trader Swiss Tony is actively trading sports right now. Let me pull profile data for him. Plus the lean I have no idea what that is. Market state for one of his fresh entries. And there it is. Our copy trading report is ready. May 18 data quality notes. The default return junk. Best signal source today. Your own track trader feed. One of your followed wallets. Swiss Tony is making disciple sports plays right now with fresh entries.

All freshness is current as of this time. Your watch list is empty. The candidates. And here is the first copy trading candidate. It gave us the second one. The watch list and everyone we should skip. Skip all five minute Bitcoin markets up and down. These dominate the feed but are arcade style ultrash short-term gambling not predictions. Recommended next actions as well. Add nicknames to your track traders. If you take any candidate, add this wallet to tracker and watch for the next couple of weeks, rerun this report at kickoff, 3 to four hours from now.

It's as simple as this. This gives me a short list of traders and trades ranked by how useful they are for copy trading. Then I have claude code connect the dots. Not just who is up this week. I want to know are they consistently good in one category. Are they taking meaningful positions? Are they early? And is the market still liquid enough for me to enter without getting destroyed by spread? If clawed code finds a trader worth following, I ask it to add that wallet to my tracking list only after it explains why.

Then I ask it to summarize what those follow traders are doing in plain English. This is where AI makes the strategy beginner friendly because the hard part is not clicking copy. The hard part is knowing when copying makes sense and who to copy. A trader might have entered uh a position at 42 cents and now the market is at 71 cents. That does not mean you should copy it now. You might be late or maybe the same trader is taking tiny profits across random markets.

This is not smart money. This is just noise. So I want my AI agent to tell me the difference. Here is the second prompt. I would use once I have a few wallets or trades worth studying. And by the way, you can even ask it to go through all of the Poly Market data, sort by the top profit and loss accounts or wallets on Poly Market, [snorts] see if they're currently active, see or ask it if they're good traders to copy trade and have it compile a list if you want to go about this faster and find traders, smart money, smart wallets to copy faster.

Once you have your smart money then or your list of copy traders rather I'm going to paste in the second prompt. You are my smart money copy trading filter. I'm not looking for random wallet activity. I want beginnerfriendly copy trading opportunities on Poly Market. For each trader or trade, tell me is this wallet actually worth following? Number two, what category do they seem good at? What trade did they make? Are they early or am I late?

Is the current price still copyable? Along with a bunch of other questions. Do not execute a trade. If there is a potential copy trade, only prepare a preview and wait for my confirmation. So, it's currently refreshing prices, fetching all of the data that it needs to answer my question. Big update. Swiss Tony just placed his largest position of the day 4 minutes ago. $317 on Auckland versus Sydney under 4.5 goals at 85 pulling that market's current state and it just gave us a copy trade candidate just like that and another copy trade candidate.

So the first one is Leane to win and the second one is where is it? Whoops. Legon versus Hesca. Not a draw. And it told us don't copy this one of his trade. His biggest bet today. It's telling us don't copy it. Why? Because we are late. His buying pushed the price from 85 to 89. Technically, yes, at 888, but 3 cents above his fill. What is worth being on our watch list? And everything else filtered out. What to do next?

Tell me which one you want to take and at what size and I'll prepare a preview only. No trade will go out until you explicitly confirm after seeing the preview. And from there, the workflow is simple. I want my agent to find strong traders, track what they are buying, filter for trades that are still copyable, use AI to explain the thesis and the risk, preview the trade, then decide manually myself, or that's one of the best parts.

If you want it to run automatically, to just run on its own, you can ask it, I want to build this out so that it runs automatically. it passes all of these quality checks and if it does I want it to execute the trade automatically. That is why this is such a good beginner strategy. It gives you the structure immediately. You're not trying to be the smartest person in every market. You're using smart money as your discovery engine and [snorts] your AI infrastructure as your filter.

By the way, I have a full video on setting up a copy trading strategy because it's one of my favorite strategies. It's the strategy I recommend every beginner start with. If you want to watch that full video, I'll link it in the description below as well, so you can go through it. And yeah, again, you can start out with approving the trades manually, you selecting the trades, your AI agent feeding you the data and you selecting the trades, or you can make it automatic where it just does it on its own.

You don't have to do anything. I don't recommend you start out with automatic because I recommend you actually understand what is it that you're doing, what is the edge, what is the disadvantage, when are you wrong, when are you right, what are good markets to copy, what are bad markets, who are good traders to copy, who are bad traders to copy for you, for yourself. That being said, let's get into the pros and cons of this specific strategy.

Okay, let's start with the pros. The pros of copy trading is that it is one of the easiest strategies for beginners to set up. It gives you a better starting point than random market browsing. It helps you find traders who specialize in certain categories. It can surface markets before you would have found them yourself. It works well with AI because the agent can summarize wallet patterns across trades. It turns Poly Market from a giant messy feed into a simple watch list of people and trades.

Now the cons, the downsides. You can copy someone too late. You may not know if they are hedging somewhere else. A trader can be right for reasons you do not understand. Leader boards can rewards recent luck but not durable edge. Meaning you could be copying someone who is very good today or yesterday and then you start copying them today and all of a sudden they start losing and they start sucking. Some wallets look smart for one week and then they fall apart.

So yeah, this is copy trading in a nutshell. the pros and the cons, but the edge is not blindly copying every trade. The edge is copying selectively with filters after the AI shows you why the trade still makes sense. The second strategy is crypto short window markets. These are markets around Bitcoin or Ethereum being above or below a certain price at a specific time. They're attractive because they are active, easy to understand, and there's a lot of external data, but they are also dangerous because they make you want to overtrade.

There are so many things happening, so many markets that you can be dabbling in. If a market updates every hour or every few minutes, your brain starts treating it like a slot machine. So, the edge here is not speed alone. The edge is having a strict filter. Now, I do not manually search Bitcoin markets in the terminal. I ask claw code to find the best short window crypto markets and explain which ones are worth ignoring.

This is in fact one of my starting strategies, the strategy that I got into the AI agentic trading world into and some of the first videos that I made around AI [snorts] agents trading on Poly Market. I have a ton of content around them. Uh, I love them, but they're not as beginner friendly. I would say they're the next step up after copy trading. There's nothing as simple as copy trading. But [snorts] if you can find an edge in these markets, the fact that they are so recurring, that there are so many every day, every 15 minutes, every hour, every 5 minutes, you could be having a serious edge and leveling up your game from there.

[snorts] That being said, let's get into the prompt and setting one up. So, the prompt I'm going to be using for strategy number two is the following. You are building an automatic polyarket crypto short window market agent for me inside of cloud code. The goal is to create an agent that monitors active bitcoin and ethereum markets, filters for tradable setups, compares poly market prices against external crypto market data, and alerts me only when a market is worth reviewing.

Use bullpen as the polyarket data layer. Use reliable external price data for Bitcoin and Ethereum. You may store market history, alerts, and watch lists locally. Do not expose private keys. This is just for safety. And then I gave it agent requirements. Scan active markets on a schedule. Track market title slug. Pull external price data and recent volatility. Compare current poly market price to spot price. Reject markets with weak liquidity.

Reject. Classify each market as trade candidate. For trade candidate, explain the thesis. Save state so the agent does not repeat the same alert every run. Send a concise alert only when a market passes the filters and include a dry run mode first. So we can start by testing it out without real money. Build the first version now. Show me the schedule, the filter thresholds, the stored state, and one example alert. Build dry run mode first then show me.

That's it. That's the entire prompt. That is the system. The agent is not just saying Bitcoin is going up. That's useless. The agent is checking price, time remaining, spread, liquidity, volatility, and whether the market is actually tradable. the alert. Well, once we get it, I'll show you what the alert should look like. But for now, it's going through [snorts] all of the instructions and creating us our second agent.

And this is the cool part of using a very smart AI agent layer like cloth code, Hermes, Codeex, OpenClaw is their ability to reason. So here as it's building the code, it's testing it out and it found an error. It said bullpen polyarket event doesn't return pre-market slugs. Discover does but event doesn't. So instead it's shifting the code and it's literally working on it. End to end works. Three real candidates surface.

Let me verify it works. Agent is built and behaving correctly. It's scanning correctly. classifying, alerting and dduping correctly. And version one of our agent is literally live. It created it under this location and it gave us everything scheduled. It runs every 5 minutes. Filter threshold. It's filtering for these metrics specifically stored state. So it's also keeping a historical log of everything. It will need live mode safety gates.

To go live later, all three must be true. When live is wired, the agent will only place an order if classification meets. If basically we meet these thresholds, what's next? Tell me what to wire next. Add a Slack Discord web hook for alerts if we want them to send it to uh say Telegram, for example. B. build a back test harness that replays the history against actual settlement. So to verify uh our strategy or filters or thesis works.

C implement the live execution path. I'd recommend B first uh which I agree with. D add the five to 15 minute updown markets back in with a separate stricter rule set. and e schedule it now via launched or slash schedule skills. So just like that it built us the agent itself that we will need. This is insane how quickly it was able to do that. What I would recommend for you to do from this step is one wire it to an alert system.

If you use Telegram ask it to wire it to a Telegram bot. If you use Discord, ask it to wire it to Discord. So on and so forth. WhatsApp. WhatsApp. [snorts] Then go with the B option, which is you validating that thesis that it works. So this is it basically back testing with previous historical data. And then once you have a strategy that you see properly works, take it live and you can include I don't recommend you start with the 5minute markets because they are wild and very volatile.

I would recommend you step up to the 15-minute markets first and then once you feel like you have a grip on the 15minute up and down uh crypto markets, then you can step up to the fiveinut ones. and make sure to keep it scheduled or to schedule it so that it just runs automatically on its own. The reason why crypto short window markets are so good for agent is there's always new data there 1hour markets meaning every hour there's a new market that you can bet on or 15-minute markets which would mean every 15 minutes or 5 minutes every 5 minutes there's a new market you can trade there's always new data there are always new markets and the rules can be turned into filters [snorts] let's dive into the pros and the cons of crypto short window agents.

The pros are that they are easy to understand for beginners. There's also a lot of external data. Meaning, what I really love about the strategy is you can always compare outside uh Bitcoin prices. What is the Bitcoin price on say Binance for example? What is the actual Bitcoin price versus what is it on Poly Market? Sometimes there's a very small delay. These markets can also have very strong volume. They're I think one of the most popular markets on Poly Market and they are perfect for scheduled monitoring.

The cons of this strategy, they encourage overtrading. When I first set up my first aentic trading agent, I would just wanted to place more trades, which is counterproductive because you want to make sure that you only place trades that meet that certain filter, the certain filters that we set up earlier. Uh but every time you see a 15inut slot and it didn't place a trade, you're like, "Ah, we just wasted 15 minutes and how can we place even more trades?" Which is counterproductive and will result in you losing money.

Trust me, I've been through that. So don't overtrade. Don't indulge in overtrading. Another con is that short time frames, especially the five minute markets, are extremely noisy and volatile. Some markets have even terrible spreads and you can be right directionally and still lose because your timing is bad. So for crypto markets, the agent should be conservative by default. It should not push you into trades. It should protect you from bad trades.

Let's move on to strategy number three. The third strategy is weather markets. I like weather markets because they're not just vibes. There are actual data sources, forecast updates, models change. Official agencies publish data and sometimes markets do not update immediately. That creates a possible edge. Not a guaranteed edge, a possible edge. But again, the goal is not to manually research one weather market. The goal is to set up an agent that watches specific weather markets and checks them against reliable sources automatically.

For weather, the biggest rule is the agent is not allowed to guess. It needs sources. The idea is very simple. There's a ton of weather data online. Say for example, today's highest temperature is going to be 35° in San Francisco. Then you can go on to a poly market San Francisco weather market and bet on the fact that San Francisco's hottest weather is going to be in that example 35. That is the gist of this strategy.

So let's get into the prompt. Now this is the prompt we're going to be using for weather markets. You are building an automatic polyarket weather market research agent for me inside of clock code. The goal is to create an agent that monitors active weather related polyarket markets, checks them against reliable external weather sources, compares the market implied probability to source fact evidence and alerts me only when there may be a real pricing mismatch.

And again, we use the same uh rules conditions for the agent and we give it requirements. The requirements being scan active weather markets on a schedule, store market title, slug resolution criteria, so on and so forth, pull or summarize reliable external forecast data, compare markets implied probability against that uh data that we just fetched. Reject markets with vague resolution criteria, weak liquidity. We're giving it the rejection criteria.

Classify each market as trade candidate, watch list, or hard pass. For candidates, include source links, the strongest yes case, strongest no case, invalidation trigger, and next forecast update time. Eight, save state, and avoid duplicate alerts. Nine, send alerts only when new forecast information materially changes the thesis. And 10, include a dry run. So the important part in this strategy again is source discipline.

If the market says 30% chance of x weather or x degree in a certain city and the official forecast suggests 60%, maybe there's something to trade here, right? Maybe there's an edge for you to place a bet on that prediction. If the market however says 30% chance of a certain weather and the agent cannot find reliable data to track that that is not a trade. That is just a guess. Wearing an agent costume making itself look smart.

But that being said let's look at the progress. There you go. Found the pattern. City temperature markets with daily strike grids. Let me see which cities have markets for upcoming days. So it found Beijing, New York, London are live. I'll use NWS for US cities and open Meteio for everything else. Building it now. Single file. That's last year's NYC markets 2025. Year is in slugs. Let me find 2026 one. So here it's figuring out how would it actually operate?

How would it find the current live markets for weather markets? How would it find the external sources? Because really the idea here is uh say for example in New York weather forecast apps show that there's an 80% chance it's going to be 80° today. And on Poly Market, people haven't caught up to that. It's currently at 30% chance. So, you have an edge there to arbitrage. You have an edge to bet on the fact that this likelihood, this probability is going to go up from 30%, up maybe to 50%, 80%.

And if it goes up from 30% to 60%, then you've just made a 2x. you've doubled your bet size, which is the whole idea of this strategy. But in order for it to do that, we need reliable sources, which is what it's currently doing. It's configurating all of that and philosophizing. I love clot codes. I love clot codes lingo. And we'll just wait for it to complete. It's going to take a couple more minutes. But that's what I love about agentic trading is you just have to set it up once and the setup it is what takes a lot of the time or most of the time not a lot of the time because maintenance also takes uh quite a bit of time depending on the strategy.

For example, for copy trading you're going to want to uh maintain, polish, clean up the list of traders that you copy. Good traders don't stay good traders forever. For uh 15 minute, five minute uh Bitcoin price markets, you're going to need to maintain the criteria list. Uh when does it place a trade, when doesn't it place a trade? And for weather markets, you're going to want to keep up to date the sources, the probabilities, the rule sets.

But now what I want you to focus on is getting an agent that runs successfully and that is profitable. Those are our two main objectives at first. Then with time, you're going to want to polish up your strategy. As you also gain more understanding over what are good days, what are bad days, what did it do on that good day, what did it do on that bad day, and asking it, you know, how can we do more of what we did on that good day? and how can we do less of what we did on that bad day.

But there we are. It's almost been six minutes since we sent that prompt. It's now running a dry run scan. Uh which should mean that it's testing that the agent properly is running. And once it's done testing, if it's working, it'll tell us it's working. If it's not, it'll just keep fixing it. So, I'll see you in a few minutes once it's done. Okay, that didn't take uh much time. After I paused the recording, literally 10 seconds after we have our V1 weather agent built and validated, its schedule, it runs every 30 minutes.

Uh and here are its filter thresholds. It has a minimum liquidity threshold, a max spread, minimum minutes remaining, maximum minutes remaining. So it will not bet if a market will resolve in more than two days, 48 hours. It's minimum event 24-hour volume. It basically placed all of these criteria conditions. Cleanest example alert. And this is what an alert would look like. Once it finds a good trade or places a good trade, weather trade candidate here, it's in a dry run.

Uh, so it's not placing actual trades, actual money yet. Market, will the highest temperature in Houston be between 86 and 87 on May 19th? The forecast for Houston is 89. The time left is 1,600 minutes. There's this much liquidity. And here's the data source. Strongest yes case, the forecast is 89°. It sits 2° above this bin. Small downside revision lands inside strongest no case forecast is firmly above 87° invalidation.

The next weather forecast update shifts predicted high by.5°. Liquidity drops below 500 yada yada yada. Next forecast update is in 1 to 3 hours. That being said, it's running well. Uh the first dry run produced 63 candidates that we could bet on and it suggested what we should wire next with A being a back test harness that replace previous history. So much like our second strategy uh B titan alert volume now raise the minimum edge to 0.1 which I would absolutely be for.

I always believe that you should be placing the least amount of trades that you can as long as they are good trades, right? You don't want to be just placing trades left, right, and center and not filtering anything out. You want to have as high and as good as filters as possible. So, personally, I would go for that. I would tell it to tighten the alert volume in case you're using Hermes or Open Claw. By the way, for you, the alert system is already wired in or you can ask it to yes, send me alerts on Telegram or on WhatsApp, whatever it is that you use your AI agent with.

Given that I'm doing this on clot code, I can ask it to wire it into my Hermes or my open claw or just create a telegram bot that all it does is alert me on poly market trades [snorts] and then schedule it uh for F. And that's it. That's our weather market strategy. We can ask it to go live. We can ask it to paper trade so we can validate and test that strategy first. That is up to you. I always recommend you start paper trading, going on a dry run before ever putting in real money on the line, testing it out, validating it, and then once you have a valid thesis, once you go a full 24 or 48 hours of you testing that strategy, then if it's working, go ahead and turn it live.

Let's go into the pros and cons of this strategy. All right, let's get into the pros and cons of weather market agents. The pros are very simple. Weather markets can be very datadriven, which is amazing news for us and for agentic trading. They are also good for AI research because the agent can compare external sources against market prices. They often have clear catalysts when forecasts update and they teach you to think in probabilities and not opinions.

The cons of this strategy is that liquidity can be a lot weaker than the two previous strategies we just outlined. There are less people trading those markets. The spreads can also be too wide as a consequence. The resolution criteria can sometimes be wonky or weird. And well, the worst one, forecasts can change very quickly. If you understand weather forecasts, they're less based on certainty and more based on probability.

It is 80% likely that it'll be 29° today. It's not 100% certain, though. [snorts] So, weather markets are good for agents only when the market quality is high and the data source is reliable. That said, let's get into strategy number four. The fourth strategy is arbitrage and correlated markets. This is the advanced section. People hear the word arbitrage and think free money. Usually, it's not free money. [snorts] Usually, it is complicated money with liquidity risk, timing risk, fees, and resolution risk.

But it is still worth learning because agents are good at monitoring relationships across markets. A human might miss two related markets, get out of sync. Uh, an agent can check that every hour, every 15 minutes, or whatever schedule makes sense. The goal is not for it to go and find free money. The goal is to know how to build an agent that watches for structural mispricings and then stay skeptical. If you don't know what arbitrage is, arbitrage is say you have a market that is around GTA's release.

GTA will come out on June 1st and the probability is 80%. that Poly Market is pricing it at 80%. Because they announced that it's going to be out on June 1st. Let's take that as an example. Then say you have another market that is GTA is going to be playable by June 1st. It's a different market name, but it's technically the same market. Meaning if one of them turns out to be true, the other should also be true. But that first market has an 80% probability.

Whereas that second market, say it has a 10% probability. Technically, they should be the same probability, right? They're the same outcome. They're the same market essentially, just different names, different uh criteria. That being said, that's where arbitrage comes in. You can bet on that market, that smaller market, market number two, and see it synchronize with the first market. Synchronize the odds go from 10% to 80%.

Where most people agree that is arbitrage in a nutshell. Let's get into the prompt for setting up your arbitrage agent. So, here is the prompt for our arbitrage agent. You are building an automatic polyarket arbitrage and correlated market monitoring agent for me inside of clot code. The goal is for you to create an agent that scans active polyarket events for correlated markets, logically related outcomes and possible hedges and potential arbitrage structures.

The agent should be skeptical by default and alert me only when a possible opportunity survives liquidity spread fee timing and resolution checks along with the agent requirements. Number one being scan active high volume markets. You don't want to be uh getting involved in low volume and low liquidity markets. That is always the case for every strategy that we've outlined, but especially for this one. There can be a lot of different markets on poly markets and sometimes they can have almost no volume, no liquidity, no spread which is a horrendous way of losing money basically uh which is why we've given it these requirements.

Reject anything with unclear rules. That is extremely important. Estimate whether any apparent edge survives our criteria. Classify each finding into real candidates, thin edge, educational only or skip. For real candidates, explain the exact trade structure, what could break it, and what needs human review. Save every candidate and every rejected candidate so the agent learns which patterns are noise and which are good.

And then send alerts when a potential edge is above threshold. As you can see, the agent needs to be skeptical with this strategy. You don't want it just firing alerts your way left, right, and center. If it cannot explain the risk, it is not allowed to recommend anything because this is where people get hurt. They find the price difference and they think, "Oh, I found this is the arbitrage thing I've been hearing about." but then they realize there's no liquidity or the market resolves differently than they assumed or the hedge does not actually cover the same outcome.

That being said, we're going to let it run, install, and see if it works or not. Let's uh have a look. Okay. And after 5 minutes, it has completed V1 of our arbitrage agent. It built it and validated it. What the scan found? It found 30 top events. It found eight multimarket events. Two pass, six were rejected. So in total, basket sum findings. Four, it found zero real current candidates, which a lot of people might be bummed about, but I'm happy with that.

I again you want to be extremely skeptical with this specific agent but it found two educational findings real raw mispricings where fees eat at it meaning it found two markets where there is real edge but you just placing a trade you would lose because the fee uh I mean you spend a fee on poly market whenever you trade. So that fee that you spend is basically eats away at your edge. And there we go. FIFA's basket is genuinely overpriced by only 4 cents, which isn't a massive edge.

If you could buy no on every team at zero cost, you'd lock in profit. But at 58 legs times.7 fee, that is 40 cents of friction. The mispricing is real, but uncapturable at retail size on Poly Market today. This is exactly what the educational tier is for. Now, we've just run a fresh scan. This is our first scan of arbitrage opportunities, which is also a reason why it did not find trades for it to place right this second.

But now that we have this agent running, we can ask it to schedule a run every hour or every 30 minutes. And it will start finding new fresh edges, uh, which is where it will find a lot more real candidates that it can place trades on. And this is what an alert might look like. Hey, here's uh the signal. Here's the edge, the exact trade structure. what could break it, what needs your review. It recommends a schedule of every 5 minutes, which is not a bad idea if you want to be as fresh as possible.

Have as fresh as data as possible. Uh, and here are the limitations, which we'll get into in a second. And as always, the recommendations. Have it hooked on whatever chat platform you use. add date, monotonicity, arbitrage detection, that is a mouthful, and schedule all three of these uh to be recurring jobs that you go through. That being said, let's get into the pros and cons of this strategy. All right, the pros and cons of arbitrage agents.

The pros are very simple and straightforward. Arbitrage agents can reveal real inefficiencies. They're very good at watching many markets at once, something us humans can't realistically do all day. They are also good at teaching you to think structurally and to really learn and understand how polyarket betting markets work. Lastly, they can reduce directional exposure if the hedge is real. >> [snorts] >> The cons however is they are the easiest strategy to misunderstand.

Liquidity can erase your edge in an instance and resolution criterias sometimes can be very messy very vague and in case they are then they can break your hedge. And with arbitrage agents execution timing matters. So speed matters a lot which can be a big con if your agent is not fast enough. So arbitrage is not beginner free money. It is an [snorts] advanced monitoring agent that should be skeptical by default. Let's get into the fifth and final trading strategy in this video.

I'll see you there. The fifth and final strategy in this massive guide is eventdriven news trading. This is probably the most natural use case for AI agents. Poly market is full of markets that move when new information comes out. Whether it's elections, regulations, sports injuries, product launches, court cases, crypto news, or geopolitics. The question is not just what happened. The question is has the market already priced it in.

And this is exactly the kind of thing I want running automatically. I do not want to manually check every news source and then run on Poly Market and place the trade myself. I want an agent that watches new markets, high volume movers, relevant news, comments, holder changes, and price movement. then alert me when something actually changed. Here is the prompt to set that agent up. And this is the prompt we're using for this strategy.

Number five, you're building an automatic polyarket eventdriven news trading agent. For me, the goal is to create an agent that monitors new and fastm moving polyarket markets, check reliable news sources for catalyst, compare the news to market price movement, and alert me only when there may be a tradable information gap. We're giving it a set of instructions and the tools that it should use. We're giving it a set of requirements.

Include a dry run mode first, as always. Build the first version. Now show me the schedule, source list, catalyst scoring logic, stored state, and one example alert built dry run mode first. Then show me exactly how to enable live auto execution safely with max position size and max daily loss, allowed markets, blocked markets, and an emergency stop. With that, we're going to let it build. hardest of the four agent because real news interpretation needs an LLM which is where AI agents thrive.

So we're going to let it build and we'll come back once it is fully done. The key phrase by the way in this prompt is already priced in because if news comes out and the market has already moved from 30 cents to 70 cents then you're probably a little too late. That doesn't mean that there's no trade to be made, but it means the question changed. You're no longer trading the news. You're trading whether the market overreacted or underreacted.

That is a much harder question to answer. I've also included uh uh an alert section for this strategy as well, which we'll go over once it finishes building this bot for us. This is what I love about tools like Clot Code and Codeex is they will literally test out what they've put out by themselves, find bugs and fix them out before they tell you that they're done. Here, look. They found a quick bug fix in the alert formatting and it's fixing it.

And it should be done after that. And there we go. It just finished. Two real alerts fired on a real catalyst. This event being the catalyst. The headline matched both NBA Finals markets. That's a genuine just happened repricing event. So we This is very rare by the way. Our first literal first run and it found something instant for it to react to. The catalyst scoring logic is over here along with the example of what an alert might look like and in this case is literally the event that just happened right now.

Fresh headline 70 minutes a oh 70 minutes ago. That's not fresh last 30 minutes. 70 minutes. Okay, so there you go. That's a real bug that we would need to fix. And the way I would fix it is I would literally tell it, hey, I would copy paste this and tell it over here, hey, you told me this is just fresh, but it's from 70 minutes ago. I need it to react to news that is 5 minutes old or younger. and instruct it like that.

This is exactly why I recommend you start with a paper uh trade with a dry run, right? Not using real money from the get- go because your agent will come up with a strategy that might be effective, but it won't be perfect or anywhere near perfect. You're never going to get a perfect strategy. The goal is to get as near to perfect as possible. But that being said, while we let it run this, let me go over the pros and the cons of eventdriven news agents.

The pros of this strategy is they fit how Poly Market actually moves. And AI is great at summarizing messy information very quickly. Agents can also monitor many markets and sources at once, something we can't physically do as humans. And the last pro is they can catch fresh catalyst faster than any manual workflow. The cons, however, is sometimes headlines can be misleading. Your AI agent can, as a result, misread the result of a news article.

[snorts] Number two, markets can move before you act, before a news headline even comes out. There is such a thing as insider trading after all. [snorts] Number three, AI can summarize fake or lowquality sources if you let it. This is where checking that its summarization capabilities, its ability to understand a new source is accurate and functional. [snorts] Number four, you can confuse new information with tradable edge.

So, the agent needs to site sources, compare timing, and answer the pricing question. What happened is not enough. You need to know whether the market has priced it in already or not. Now that we have five strategies, we need the most important section of this video. [snorts] How do you tell a good market from a bad market? Because a strategy is only as good as the market you apply it to. Here is the checklist that I use.

A good market has clear resolution criteria. You should be able to read the rules and understand exactly how it resolves. A good market has real volume. If nobody is trading it, your price might be fake or false. A good market has real liquidity. You need to be able to enter and [clears throat] exit without getting destroyed. A good market has a tight spread. If the best bid is 40 and the best ask is 60, that market is already telling you to be careful and that's not a good thing.

[snorts] A good market has reliable external data, whether that's weather forecasts, price feeds, official statements, public filings, injury reports, something real. And a good market has a catalyst. There needs to be a reason the price might move. A good market has a clear invalidation point. You should know what would make you wrong before you enter and a good market fits your risk rules. If the only way the trade matters is by risking too much money, it's not really a good trade.

Now, here are the bad market warning signs. Number one, vague resolution. Number two, widespread. Number three, low liquidity. You're starting to notice it's a lot of the good ones, just the opposite. Number four, no recent trades. Five, no source of truth, no good reliable external data. It's a bad market if it's just based on pure vibes. If you cannot explain your edge. Number seven, if you're chasing a move instead of using reliable data.

Number eight, if you're copying just blindly. Or number nine, the worst one, when you need the trade to work emotionally. That last one is how a lot of people lose money. [snorts] Here is the prompt I would use before touching any market. Score this polyarket market from 1 to 10 as a trading opportunity. Evaluate resolution clarity, volume, liquidity, spread, recent trades, data availability, catalyst timing, smart money signal, and downside risk.

Classify it as a trade candidate, watch list or hard pass. A good prediction is not enough. Explain whether it is a good trade. The line I want you to remember in this prompt is this. A good prediction is not automatically a good trade. You can be right and still lose money because the market was bad. Now, let's talk about trading agents. This is where things get exciting, but also where things can get dangerous. A trading agent should not be some random bot that wakes up, reads one market, hallucinates a thesis, and executes a trade.

That is insane. A useful trading agent needs rules. First, it needs fresh data. Before every recommendation, it should refresh prices, refresh liquidity, spreads, recent trades, relevant external data, your current positions, and your open orders. Second, it needs permission boundaries. It can research, it can summarize, it can rank, and it can preview. It cannot however execute without explicit human confirmation unless of course you've tested it and you've explicitly asked it to run automatically.

Third, it needs risk rules. Things like max position size or a max daily loss, meaning if you lose more than this amount, just stop whatever you're doing until we check it out together. It needs a max trades per day amount. It needs a rule for it not to trade liquid markets or no trades without sources or no trades without invalidation criteria. Fourth, it needs logging. This is super important. Every trade needs a thesis.

Why did we enter it? What would make us wrong? What is the exit plan? What sources did we use? What did the agent itself recommend? Then what did us the human decide? And fifth, it needs failure handling. If bullpen fails or if polyarket data fails or if external sources fail, the agent should stop. It should not guess. So here is the master prompt that I would use here. You are a polyarket trading research agent. You may research, summarize, rank and preview trades.

You may not execute trades without explicit human confirmation. Before every recommendation, refresh market price, liquidity, spread, recent trades, relevant external data, my current positions, and my open orders. If [snorts] any data source fails, stop and report the failure. separates facts, assumptions, speculation. Always include invalidation criteria, position sizing notes, and risk notes. The most dangerous agent is the one that sounds confident when its data is stale.

So, the job is not to make the agent more aggressive. The job is to make it more disciplined. Finally, I want to teach you how you can turn any of those agents, any of the five strategies into a dashboard. Because there's no better way, no easier way for you to visualize your agents performance than having a dashboard that you can just open up, whether be it on your phone, on your desktop, when you're on the run, when you're at your office, and be able to check on that agent's progress.

How is it doing? Because if this data only lives in some chat thread, things will get messy very quickly. A dashboard gives you an operating system. Here is what I would include in my dashboard. I would include a watch list. I would include candidate trades. I would include current positions, open orders, a smart money feed, follow traders, market score, uh trade thesis, risk rules, profit and loss log, and the good thing is the bullpen commands give us all of this data that we need.

Granted, depending on the strategy that you're using, you might not want all of these showing on your dashboard. Less is more. You might just want a profit and loss chart and that's it. That's maybe all that you want. You want to see how much money that agent has made. Maybe you want that and also a trade log to show you what trades did it place and its reasoning. Maybe that's all you want. And that's a great start. By the way, again, less is more here.

That being said, let me show you the prompt that I would use to create that dashboard. Now, this is the prompt I would use to create a dashboard for strategy number one. I'm going to include the prompts for the dashboards for strategies number one, 2, 3, four, and five in the cheat sheet down below. But this is the dashboard prompt for strategy number one right here, which I've just pasted. Build this as a simple local dashboard I can run while fulfilling for a clean web dashboard. if fast.

Otherwise, generate a markdown or HTML dashboard first. Use sample data only if live data is unavailable and clearly label sample data. Add buttons or clear actions for refresh data. Add wallet, remove wallet, switch dry, run, and live mode and preview trade. Enable emergency stop. Do not execute any trade from the dashboard unless live mode is explicitly enabled and all risk rules pass. And just like that, it's starting to build it out.

And as this is running, by the way, you can design the dashboard up to your liking. Again, it should feel something intuitive for you. You can base it on the same design I used for my copy trading bot video. You can base it on this design right here, which is one I found on Google. You can base it on this design. You can go as crazy as you want. Again, this should feel native to you and something that you are able to read and understand.

Well, but let's go and check in on our bot. It seems like it finished what it was doing and said the dashboard is up. So, let's go check it out. And boom. Just like that, it just opened up. And this is the dashboard that it created. Polyarket copy trading operator dashboard. O and service status. We are logged in. Very nice. So, it gave us statuses, risk filters right here, which if you want, you can also add like a settings menu or a settings button.

If we click, we can edit it from the dashboard. And over here, our wallet uh watch list and our smart money feed. This is exactly what we want. We can refresh data, switch to live mode, or click on emergency stop if we want to stop it right away. We can add more wallets. We can literally do whatever we want. And if we go down, copy trade candidates, recent agent decisions, P&L performance. So, we're currently on a dry run.

It just started. And alerts right here. In my opinion, this is extensive. I intentionally included everything you might need in the prompt, but what I do recommend is not for you to use all of these. Install it first. see what you actually end up using the most and only keep those because again the more you see on your screen the less you're able to read the more analysis paralysis you will get. So open it, install it, see what it is you actually end up using and needing and only keep that.

Tweak them as you as you wish. And look at that recent agent decision. So it's already starting some decisions 44 minutes ago. scalper bot. Uh, so it's showing us the wallet, the markets here. It decided to skip this one. Wallet confidence was below the threshold, which I love to see. And this one, it decided to watch it because slippage is low. Awaiting copy threshold. It's a fair window. It decided to scrape it to skip it because slippage is low at only 1 cents.

But this is where we'll be able to see our bot literally placing the trades as it goes. And that is a wrap. That is the full entire system displayed. We just set up bullpen. We used AI to build five polyarket strategies. Smart money tracking, crypto short window markets, weather markets, arbitrage and correlated markets, and eventdriven news trading. Then we covered how to spot good and bad markets, how to keep trading agents up to date, and how to build a dashboard around this whole thing.

Again, this is not financial advice. This will not magically make you profitable, but if you are going to use AI for prediction markets, this is the kind of workflow that I would want. research first, preview before execution, human confirmation, hard risk rules, good locks, and no blind trading. [snorts] If you want to try bullpen, the link is down in the description. Start in preview mode. Build your watch list. Score markets before you trade them.

And do not let an AI touch real money until you have rules, logs, and a clear reason for every trade. That is how I would approach Poly Market with AI. And as promised, I have also included the link to the cheat sheet, which you can find all the prompts we've used throughout this massive video down below. Make sure to download it so you can literally copy paste them. If you've enjoyed this video, make sure to subscribe to my channel.

There's a ton more content just like this, so feel free to check it out. And I'll see you in the next video. Take care.