Week 1: First Steps — Richard Thomchick

Goal: Set up development environment and make my first Claude API call.

I am not an engineer, but I have done enough coding in the past to get frustrated with the setup process and toolchain dependencies. I tried an AI training course in 2018 and couldn’t get past the step to install PyTorch. It was downright embarrassing. So my first question was, “Can Claude get me over the hump? Can I make it work?”

Thankfully, the answer was Yes! Within the first five minutes, I knew this was going to be a very different (and better) experience than I’ve had in the past.

What I Did

Set Up the Dev Environment

Installed Python, created my first virtual environment (python3 -m venv venv)
Reacquainted myself with VS Code; installed Claude Code and several extensions
Set up a project directory at ~/ai-projects/
Installed the Anthropic SDK (pip install anthropic)

Generated My First API Key

Generated an API key from the Anthropic console
Configured it as an environment variable (export ANTHROPIC_API_KEY="sk-ant-...")
Learned the hard way that export is per-terminal-session: close the window, lose the key

Made My First API Call

Wrote the “Hello World” script client.messages.create() with a simple prompt: “Hello, Claude! What can you help me build?” About 10 lines of code.
Successfully called the Claude API from my own machine and received my first response.

What I Learned

Setup was still the hardest part. Getting Python, the venv, the API key, and the editor all working together is still an unglamorous prerequisite. I suspect it will get easier as I shake off the rust of having not used these tools in a while.
What “real” prompt engineering is. Running Claude from code is a fundamentally different relationship than using the Claude AI chat interface. I get to control the system prompt, the temperature, the conversation history. Everything that felt like magic in the chat UI is now a parameter I can set.
How little code it actually takes. The first working script was probably 10–15 lines. After weeks of imagining that “building AI products” required deep engineering, the entry point is surprisingly accessible.

What I Struggled With

Terminal/CLI unfamiliarity: navigating directories, understanding PATH, knowing which Python I’m running (and juggling between system Python and venv Python).
venv activation is per-session. This tripped me up — I installed anthropic, closed the terminal, reopened it, and got ModuleNotFoundError. The mental model that clicked: a venv is like logging into a specific account. Close the terminal, you’re logged out.
Understanding what the SDK actually does vs. what the API does vs. what Claude does — three layers that blur together at first.

Key Takeaway

Being a developer is much more than just coding. I can see how vibe-coding might make a person feel drunk with power, but there is a lot of less glamorous work — environment setup, deployment config, error handling, cross-system debugging, evaluation infrastructure — that is much harder to vibe-code your way through vs. a professional software engineer who understands what’s actually happening on the machine.

Week 1 complete. Environment works. Claude responds. Everything else starts here.