Seven Steps to AI Coding

Step 1 of 7 · AI Coding Mastery

Command line interface (CLI)

The command line gives you direct, scriptable control over LLM APIs — far more than a chat window offers. It's where automation, agents, and reproducible workflows live. The payoff comes in four forms: more control over how you call APIs, a unified workflow across every operating system, real efficiency on repetitive tasks, and a concrete understanding of how an AI workflow actually fits together — all laid out below.

Windows users: install WSL (Windows Subsystem for Linux) to run these same commands. The breakdown below groups them by task.

What each command does

Step 2 of 7 · AI Coding Mastery

Python on your platform

Before writing a single line, know which Python you're running and where it lives. The naming differs by platform, and using the wrong interpreter is the most common source of "it works on my machine" frustration.

Windows vs. macOS / Linux: Windows usually exposes Python as python (or the py launcher). macOS and Linux ship a legacy python tied to Python 2, so Python 3 is invoked as python3. Match the command to your platform — they refer to different binaries.

$ python --version
Python 3.12.4

$ python3 --version
Python 3.12.4

$ which python3
/usr/local/bin/python3

# Windows:
> where python

$ python3 hello.py
$ python3 -c "print(2+2)"
$ python3 -m http.server

Step 3 of 7 · AI Coding Mastery

Set up an environment for your project

What's an environment? Think of it like local variables inside a function — they take precedence over globals with the same name, and they're scoped to that function alone. A Python environment is the same idea applied to packages: each project gets its own private set of installed libraries, scoped to that project. This solves a real problem — modules carry their own version-specific dependencies, and without environments, Project A's pandas 1.5 collides with Project B's pandas 2.2 and both break.

The traditional Python approach is a four-command ritual: python -m venv, source .../activate, pip install, then remember to deactivate. The mix of commands (python, source, pip) can get confusing — and if you skip the activate step, half your packages land in the wrong place. In 2026, uv simplifies all of this: one command to set up, one to add a dependency, no activation to remember, and roughly 50× faster than pip.

Goal: project setup in 15 seconds, not 15 minutes.

python -m venv venv
source venv/bin/activate
pip install requests
python script.py

uv init
uv add requests
uv run script.py

uv finishes in ~2 seconds  ·  pip installs are still running.

Step 4 of 7 · AI Coding Mastery

Coding IDE — VS Code

Several editors support AI coding. The one most coders reach for is Visual Studio Code — free, cross-platform, and lightweight. It runs on Windows, macOS, and Linux, and starts quickly compared to heavier IDEs.

First move: open the Extensions view and install the Python and Jupyter extensions. That single step unlocks IntelliSense, notebooks, and debugging.

Step 5 of 7 · AI Coding Mastery

GitHub & your professional identity

GitHub is a free website where you put your projects so the world can see them. For AI coders it's simply expected — and it doubles as your résumé. Hiring managers and recruiters look at your GitHub profile before they read your CV, because it's where your real work lives, in public, in your name.

So how does your code get there? That's where git comes in — the version-control tool that runs on your own machine. With a few commands, git takes the project and files sitting on your laptop and uploads them to GitHub, keeping a full history of every change along the way. git is the tool; GitHub is the destination.

To use it, you'll learn a small vocabulary — a handful of terms like repo, commit, and push that describe how local work travels up to GitHub. They're below the image.

The vocabulary you'll need

Six terms, in the order your code actually travels from your laptop to GitHub.

The everyday loop: edit files → git add → git commit -m "what I changed" → git push. Repeat. That's 90% of git in daily use.

Step 6 of 7 · AI Coding Mastery

API keys & secrets

Every AI tool you'll use — Anthropic's Claude, OpenAI, Google — needs an API key to authenticate your requests. Think of that key as your credit card: it bills to your account, and anyone who has it can spend on your behalf.

Publish a key to GitHub by accident and someone will find it — sometimes within minutes. Bots scan public repos specifically for leaked keys, and a leaked key can run up real charges before you notice.

The dot-env pattern

Three pieces that keep your key off GitHub — set them up once per project.

The result: the key never enters your source code, never gets committed, and never gets pushed. Your code goes public; your secret stays private.

Learn this once. It's a skill you set up the same way every time — but get it wrong once and it can cost you real money. This step gets it right the first time, so you never have to think about it again.

Step 7 of 7 · AI Coding Mastery

Local LLMs with Ollama

Running a model on your own machine is worth learning — not because it replaces the frontier APIs, but because of what it teaches you. Ollama makes it a few-command setup.

It makes LLMs concrete. Instead of treating AI as a black box, you see model loading, inference, latency, and resource limits with your own eyes.

Basic Ollama setup

Three steps to a model running locally — shown in the workflow below.

Good first models

Start small. You can always reach for a bigger model once you know your hardware can handle it.

Commands to get started

$ ollama serve            # start the local server
$ ollama pull gemma3:4b   # download the model
$ ollama run gemma3:4b    # open an interactive chat