Claude Built Me a Live Crypto Trading Bot (And You Can Too)
What you see in this video is a crypto trading bot running live on an exchange with real money — and not a single line of its code was written by hand. Claude wrote it. The dashboard streams 24/7 on Twitch, logging entries, closed trades, and the coins the bot watches, so the results aren't a screenshot you have to take on faith.
Brian's background is patent law and computer science, and he's been writing software since he was a kid. But that's not what made this project work. What made it work was months of structured testing, validation, and knowing how to talk to an AI so it uses data instead of guessing. Here's what that process actually looks like.
Backtesting Comes Before the Bot
The bot wasn't the first thing built. The environment was. Before any live trading, the system ran backtests across a 16-month window, more than 50 cryptocurrencies, and roughly eight timeframes — 1-minute, 5, 15, 30, hourly, 4-hour, and daily — using technical analysis indicators to see which logic held up and which fell apart.
That's around 900 individual tests. Claude can generate and execute them in a fraction of the time it would take manually, but you still have to define what you're testing and why. If you're still building an intuition for which signals matter across timeframes, spending time on reading crypto charts and technical indicators on TradingView first will make your backtest design far better than guessing at parameters.
Paper Trading Exposes What Backtests Hide
Once the logic backtests cleanly, the next step isn't real money — it's a paper bot running on the live exchange. Same logic, same market data, pretend capital.
This step matters because backtests live in a clean world. Live markets don't. Slippage, order fills, and execution timing can quietly erode an edge that looked solid in historical data. A paper bot running against real order books surfaces those gaps before they cost you anything.
Since all of this runs through exchange APIs, your choice of venue affects execution quality directly. Traders who want tight spreads and deep futures liquidity often run automated strategies through Bitunix futures markets, where the fee structure and contract specs are worth reviewing before you point an API key at them.
Run Live and Paper Side by Side
The validation layer most people skip: running a live bot and a paper bot simultaneously on separate machines, logging every entry, and comparing all three data sets — backtest, paper, and live.
If the bot traded Monday through Wednesday, you can then re-run the backtest on exactly those days and confirm the logic executed the way you expected. When backtest, paper, and live all agree, you have real evidence. When they diverge, you've found a bug before it compounds.
Automation doesn't remove risk — it just executes your rules faster, including the bad ones. Position sizing and stop placement still decide whether a strategy survives a losing streak, so capping losses with stop-loss and position sizing rules should be built into the bot's logic from day one, not bolted on later.
Knowing When the AI Is Guessing
The hardest skill here isn't coding. It's judgment. AI will confidently produce code that looks correct and quietly does the wrong thing. You need to know how to prompt it properly, how to check its work at intervals, and how to tell the difference between output grounded in your data and output that's plausible-sounding invention.
That's the part no tutorial covers, and it's the difference between a bot that works and a bot that burns capital.
Brian is walking through the full build — environment setup, backtesting, paper execution, live deployment, and the pitfalls in between — in a live webinar and a multi-part course inside Crypto School. If you want to build this yourself from start to finish, join the Crypto School community at skool.com/crypto-profit and get the walkthrough soup to nuts.
---
Affiliate Disclosure: This site may contain affiliate links. If you use them, we may earn a commission at no extra cost to you. Content is for educational purposes only — not financial advice.