Getting Started — Part 4: My Toolkit — How I Build Everything Without Writing Code

The question I get most often: "If you can't code, how did you build any of this?"

Fair question. I've deployed trading bots, graduated AI agents, launched a paid API — and I've never written a line of code from scratch. Not one. Everything was built with tools, most of which cost nothing. This post lays out every single one of them.

The One Tool That Makes Everything Else Possible

Claude. That's Anthropic's AI, and it's the reason this blog exists.

Here's my actual workflow, stripped of any glamour: I open a chat and describe what I want in plain English. "I need a Python script that checks the BTC price on Kraken and Coinbase and compares them." Claude writes the code. I copy it into a file, run it, and it breaks. I copy the error message, paste it back. Claude explains what went wrong and gives me a fixed version. I run it again. Sometimes this loop happens three times. Sometimes fifteen.

It's messy. It's slow compared to how a real developer works. And Claude gets things wrong — sometimes confidently wrong, handing me code that looks perfect but fails in ways that take an hour to untangle. But even with all that friction, it's still absurdly faster than trying to learn programming from scratch. I went from zero to a running bot in about a week. No course, no bootcamp, no textbook.

Claude also does things I didn't expect to need. Strategy analysis — I paste in my trade data and ask "what patterns do you see?" Debugging — not just fixing errors, but explaining why they happened so I can recognize similar problems next time. Even writing this blog. The first drafts are mine, but Claude helps catch the parts that don't make sense or repeat themselves.

If you take one thing from this post: you don't need to learn to code. You need to learn to describe what you want clearly enough that an AI can code it for you. That's a different skill entirely, and it's much more accessible.

Where the Code Actually Runs

Claude writes the code, but it has to run somewhere. Three layers here, and none of them require understanding what's happening under the hood.

Python, VSCode, and the Mac Terminal. Python is the language Claude writes in. I didn't choose it — Claude did, and I never questioned it. VSCode is where I open the files to look at them. The Terminal is where I type python main.py and hit enter. That's genuinely the extent of my technical knowledge. I can't explain what a virtual environment is. I don't know what pip does beyond "it installs stuff." But knowing how to open a file and run it turned out to be enough to get started.

GitHub. A place to store code online. I didn't understand why I needed this until I tried to deploy my first project. Railway — the server I use — pulls code directly from GitHub. So the flow is: write code locally, push to GitHub, server grabs it automatically. Version control? Branches? I'm told these matter. I just push to main and hope for the best. It's worked so far.

Railway and Oracle Cloud. These are servers — computers that run your code 24/7 so your laptop doesn't have to. I use both, for different reasons. Railway costs $5 a month and connects to GitHub with about three clicks. My ACP agents run here. It's the easiest option I've found for someone who doesn't know what SSH means.

Oracle Cloud is different. Their free tier gives you an ARM server with 4 CPU cores, 24GB of RAM, and 200GB of storage. For $0. Permanently, not a trial. My Weather Bot and x402 API both run here. The bot uses about 50MB of RAM, which makes the 24GB allocation almost comical. The migration story is here — it happened because my MacBook couldn't stay open 24/7 and the bot kept dying every time I closed the lid.

The Free APIs That Power Everything

Every project I've built runs on data from external services. All of them free.

Open-Meteo provides weather forecast data from professional models — GFS, ECMWF, ICON. The same models that governments and airlines use. No API key required. Just call the endpoint and you get forecasts for any city on Earth. This is the engine behind the Weather Bot's predictions.

Kraken and Coinbase both offer free crypto price APIs. My PriceVerifier agent pulls from both and cross-checks them. If the prices diverge beyond a threshold, something's wrong. Two sources, zero cost.

Honeypot.is checks whether a crypto token is a scam — specifically, whether it lets you buy but blocks you from selling. The TokenTaxAnalyzer agent calls this API, analyzes the results, and returns a verdict. Also free.

Upbit and Bithumb are Korean crypto exchanges. My x402 API pulls price data from both to calculate the kimchi premium — the price gap between Korean and international markets. I wasn't sure Korean exchange APIs would respond to a server in Tokyo, so I tested it before writing any code. They worked.

The Stuff That Connects It All

Polymarket is where the Weather Bot trades. It's a prediction market on the Polygon network — real money, real odds, real results every day.

Telegram is my monitoring system. Every trade execution, every error, every balance change sends a notification to my phone through a Telegram bot. I can check how things are going from anywhere without opening a laptop. When something breaks at 3 AM, my phone buzzes. Whether I actually wake up and fix it is a different story.

Virtuals Protocol SDK handles the ACP agent lifecycle — registration, job processing, graduation. I didn't install or configure this myself. Claude added it to the code and explained which parts I needed to customize.

What It Actually Costs

People assume this stuff is expensive. It's not.

Railway runs me $5 a month for the ACP agent server. The domain for this blog costs about $12 a year. Everything else — Oracle Cloud, Open-Meteo, Kraken, Coinbase, Honeypot.is, Upbit, Bithumb, GitHub, Telegram, Python, VSCode — all free. Claude has a free tier that's enough to get started, though I use the paid version because the longer context helps with complex debugging sessions.

Three projects running simultaneously. Two AI agents graduated. A trading bot monitoring 26 cities. A paid API processing on-chain payments. Total monthly infrastructure cost: $5.

That's every tool I use. Nothing exotic, nothing expensive. The hard part was never the tools — it was figuring out what to build and then surviving the debugging when it didn't work. If you want to see any of these tools in action, each series starts from zero: the Weather Bot, the ACP Agent, and the x402 Protocol.

More updates on the way. If you're working on something similar or found a smarter way to do it, drop it in the comments — the more we share, the faster we all move.

Disclaimer: This blog documents my personal learning journey. Nothing here is financial advice.

← Previous: Getting Started — Part 3: What Are AI Agents?