My New Adventures in (Claude) Coding

My New Adventures in (Claude) Coding

My Coding Life Now Has a Fast-Forward Button

I've been building things on the internet for a long time. Long enough to remember when "just Ask Jeeves" was actually useful advice. Long enough to have opinions about templating engines and a quiet, probably-unhealthy attachment to SQLite. These days I'd describe my stack with the kind of pride usually reserved for sourdough starters: PHP dev, SQLite maximalist, Twig addict, Symfony fanboy. Static site purist. If it doesn't need a build step, I probably like it.

What's changed recently, and I mean meaningfully changed, is the speed at which I can go from "I have an idea" to "that thing exists and works." I'm using Claude and Claude Code as genuine coding partners. The difference is hard to overstate. I can describe a problem in plain language, talk through the architecture, argue about tradeoffs, and then watch a working scaffold appear. Projects that used to live in the "someday" pile for weeks are getting built in afternoons. I rebuilt a full admin panel with SQLite-backed deduplication logic for a migration project that would have taken me days of grinding through docs. I built two separate course discovery tools for CG Cookie, a keyword-weighted search interface and a branching learning-path wizard, in the time I used to spend just setting up a project folder and second-guessing my folder structure.

The key is that I'm not just vibe-coding and hoping for the best. Every project I spin up gets real infrastructure: PHPStan for static analysis, PHP-CS-Fixer to keep the code honest and consistent, and PHPUnit tests that actually cover the behavior I care about. These aren't afterthoughts bolted on at the end. They're part of how Claude and I work together from the start. I describe what a function should do, we write the tests first or alongside, and then we make it pass. It's the kind of disciplined workflow I always intended to have before I got impatient and just shipped things. Now I have it, and ironically it makes me ship faster because I'm not chasing down weird regressions two weeks later.

Claude Code in the terminal is wild in the best way. Having an AI that can actually read the files, run the scripts, check the output, and loop back to fix what broke, without me copy-pasting back and forth, collapses the feedback loop in a way that feels genuinely new. It's not magic. I'm still the one making decisions, reviewing the output, and occasionally saying "no, that's not what I meant." But the grunt work that used to eat entire evenings? Just handled. My "someday" list is getting embarrassingly short.