I wrote 10,000 lines of code without writing a single line

TL;DR: Dozens of scripts, over ten thousand lines of working code and sixty thousand words of documentation, none of it typed by me. What changed when I stopped being the coder and became the architect.

I have never done any of this before.

That is the part that still feels like a glitch in the matrix. Over the past month, I configured a server, built out infrastructure automation, and generated documentation from scratch. The result was a few dozen scripts and many thousands of lines of working code.

Zero of those lines were typed by my fingers.

When I look at the breakdown, it’s almost absurdly specific: * 41 Shell scripts (.sh) with 7,671 lines of infrastructure automation. * 12 Python scripts (.py) with 2,809 lines for metrics and reporting. * 1 Go program (.go) with 34 lines.

I didn’t just get lucky with a snippet. I managed a system that produced a coherent, working whole. And the documentation? 61,190 words across 64 files.

This isn’t about replacing engineers. It’s about what happens when you stop trying to be the coder and start being the architect.

The shift from typing to managing

My early battle with AI coding tools was trying to get Claude Code to do something without asking me 70 approval prompts. I spent hours correcting syntax, fighting context windows, and feeling like I was babysitting a very smart intern who kept misreading the brief.

I stopped correcting the code. I started building the system the agents needed.

The realization hit me when I analyzed the stats on the scripts and documentation we’d built together. The tools can test and fix a ton without human intervention. So I just let the tools do their thing, kept it on task, and usually ended up with a working thing.

I am the manager, more than the coder.

This requires a different kind of discipline. You have to define your standards and protocols, and refine them as time goes on. I’m working with Claude to ingest the details of engineering best practices and apply them directly to my workflow. Claude updates my standards docs, adds clarity around the ones that exist, and follows these standards moving forward.

If you’re just starting out, I’d strongly suggest setting up a bug tracker and a backlog tracker. Claude spun up each in Notion, without me having to touch the awful Notion interface myself. Now I just load the context, define the outcome, and let the agents figure out the path to success.

Vibe coding is real (and it’s not magic)

People ask if this is “vibe coding.” Yes. I don’t often use that name for it, but it fits.

I recently geeked out over a Photoshop script I’d been building with AI to improve my star-trails photos. It was the first version ever—so much to fix—but I had creative control over the speed, the lengths of the light streaks, and much more, all through a no-code script that was better than anything I could have written by hand.

The key is choosing the right tool for the oversight level. Some tools are built for less oversight and more autonomous path-finding, orchestrating agents to accomplish coding goals with minimal friction. For others, VS Code with the Claude Code extension remains a solid start, especially if you want to choose how much AI help you get, when you get it.

The trade-off: Outcome over process

There is a risk here. If you don’t understand the underlying structure, you can’t debug the hallucination. That’s why I treat AI as a managed risk, not a magic wand. I verify. I check the logs. I look at the stats.

But the outcome is what matters.

I generated tens of thousands of lines of code in the last few months without learning a new programming language. What I did have to learn was a new management language; how to speak to the AI so it builds what I want. It’s about clarity of intent, not mastery of syntax.

If you’re looking to tinker, don’t get bogged down in learning programming structure. Tell an AI what you want, and let it figure out how to build it. Then verify. Then iterate.

The future of building isn’t about who can type the fastest. It’s about who can define the problem most clearly.

Take it, it’s yours.

I’ve been saying this for a while. This piece was pulled together by AI from 10 of my social media posts (09/2025 - 07/2026) and polished for Cairoglyphics.ai.