So, I finally took the plunge and spent the weekend with Claude Code in VS Code. I was recently influenced by the latest Hard Fork podcast about Claude Code. If you haven’t seen it yet, it’s Anthropic’s new “agentic” tool that lives in your terminal. It doesn’t just suggest code; it basically steals your keyboard (in a good way) and does the work for you.

To see if the hype was real, I gave it a project: Build and deploy a Family Tree app using Python Flask on Google Cloud Platform (GCP).

I started with a blank folder and one prompt: “Build a Flask app where I can add family members and visualize their connections. Then, get it running on GCP Cloud Run.”

For the time being, I’m focusing on app development locally and viewing the results via docker. I tried GCP deployment and it quickly became confusing.

Here is the GitHub repo: https://github.com/alsersugasawa/family-tree-app

Step-by-Step

Scaffolding: Claude didn’t just give me an app.py. It initialized a Git repo, created a virtual environment, and set up a clean directory structure with templates and CSS.

The Logic: I watched it struggle (and succeed) with a recursive function to fetch ancestry lines from a Postgres DB. It even added a “Visual Tree” tab using a simple JavaScript library without me asking.

The “GCP Magic”: This is where it got scary-good. Instead of me googling “how to containerize Flask,” Claude wrote the Dockerfile, created a service.yaml, and started running gcloud commands in my terminal.

Pro Tip: Use the /plan command before letting it loose. It’ll tell you exactly what it’s about to do (like “I’m going to enable the Cloud Run API”) so you don’t get hit with random cloud bills.

My Lab Configuration

For this test, I used the following: Physical Hardware Mac Mini

Workloads/Apps/Services Cloud Code Extension in VScode Google Cloud Platform Python Flask Postgres Docker

The Good Stuff (Advantages)

Terminal Agency: It’s not just a chat box. It runs your tests, fixes the linting errors it caused, and commits the changes with actually good commit messages.

Context King: With the 200k context window, I could say “Remember that weird bug in the CSS from 20 minutes ago? Fix that too,” and it just… did.

Savvy: It handled the gcloud auth login and project switching like a DevOps pro. It even caught a permissions error and told me exactly which IAM role I was missing.

The Not-So-Good Stuff (Disadvantages)

Permission Fatigue: It’s (rightfully) paranoid. It asks for permission to run almost every bash command. While safe, clicking “y” fifty times during a deployment feels like a mini-workout.

The “Magic Box” Moment: Sometimes it goes down a rabbit hole trying to fix a bug and ends up refactoring half your project. You have to keep a close eye on the diffs.

Rate Limits It is a little anti-climactic when you encounter a rate limit that refreshes every few hours or worse, weekly refresh.

The Verdict:

Claude Code isn’t just another autocomplete tool like the early days of Copilot. It’s a junior engineer that never sleeps. For the Family Tree app, it saved me probably 4 hours of boilerplate and “GCP-yaml-hell.”

If you’re a terminal-dweller who hates context-switching to a browser to look up CLI arguments, this is your new best friend. Just keep an eye on your API credits.

Rating

4/5 “10x Developer” is just an AI in a terminal.

References

  • Claude: https://code.claude.com/docs/en/overview
  • NYTimes HardFork Podcast: https://www.nytimes.com/column/hard-fork