This Blog Was Created Using Claude Code, Obsidian, Rust and Svelte

This Blog Was Created Using Claude Code, Obsidian, Rust and Svelte

The world has been abuzz about Claude Code, an agentic coding agent that supposedly gives you the powers of 10 developers at your fingertips. I am AI friendly, but usually fairly skeptical of the hype, so I decided to give it a try.

Task 1 - Nano Banana Prompting from Rust

LLM’s are normally pretty bad at Rust and very bad at async Rust. I wanted to add the ability to prompt Nano Banana Pro from the Personal Dashboard project I’ve been building.

I popped open Planning Mode, which is a mode in Claude Code that is built to read over your current codebase, fetch documentation and plan out implementation. In this mode you can plan out context management, give it instructions on how to handle and style code (put tests in separate files versus inline in library files, style guides for frontend code, etc.) as well as how to handle implementations, CI/CD, etc. You can even tell Claude Code to interview you with specific questions on implementation which it will then put in the plan.

After Planning Mode, you can start the implementation. Claude Code does this agentically, which means that it will spin up multiple models to handle simple tasks like reading files into the context or generating basic Bash commands, all the way to more capable thinking models that can generate code for implementations. The whole time it will check builds and the LSP to make sure it’s generating good code. For me, this meant the creation of test files to test out new library code as it’s being created to prevent regressions in the future. More agents would then run build commands, autonomously checking if we pass build and tests before making more changes.

All told, about 300 lines of code later, I had a GitHub branch I could PR on to review the code. And it was… really good.

The code was clean, it was modularized, it was organized pretty much the exact way I had modularized the other widgets. The only thing I had to change was that it wanted to put the tests in the same files as the implementation code itself and I just don’t really like that, so I had it chunk the tests into new files. The rest was perfect.

I can now prompt Nano Banana from within my Everything Dashboard. This took less than 20 minutes and few enough tokens that I didn’t need to upgrade to a more expensive plan than the $20 per month plan I was on.

Task 2 - Obsidian Markdown -> Svelte Blog Page Converter

Now, let’s move to something slightly more difficult.

I’ve been using Obsidian as a research tool for a while now, so I wanted to throw a really hard problem at Claude Code to see if it could handle building a widget in my Personal Dash to convert an Obsidian Markdown file with embedded YAML metadata to a Svelte page component.

I entered plan mode again, generated a solid TODO list for implementation details, and let Claude Code figure out what other requirements were needed. I told it the basics of what I needed, gave it access to read my Obsidian vault to get the details, and showed it both the repository for my Svelte blog and the repository for my dashboard where the widget would be.

Over the course of about an hour, it generated 2,000 lines of code according to the specifications, created the necessary GitHub branches and commits, ran the tests, tweaked the code according to the test, LSP and compiler output and…

It worked perfectly. First try.

I created a PR, then I went to write this very blogpost in Obsidian markdown, pressed the convert button, the images and files got moved over, it creates a branch on GitHub which I can PR to kick off a GitHub Action, which pushes it to live.

Lessons Learned

This was the first “oh shit” moment I’ve had with AI in a long time. The last one was probably Cursor, which was cool but its coolness was always constrained to the IDE. Frankly I don’t even use Cursor much anymore, as to me it’s very little more than a Fairly Decent Autocomplete.

Claude Code feels very different. You can create tools for it to interact with The Real World. It has all the power of your actual machine. You can go from Barely an Idea to Working In Real Life extremely fast.

Claude Code is the first AI tool that really made me realize that we’re at a paradigm shift now, versus we will be at a paradigm shift soon.

What I think has changed, for me, is that this means that agency, distribution and execution matter far more. Technical competency is still extremely important, and it cannot be fully outsourced to AI, but the fact is that the actual act of writing the code itself has become much less important now.

You should know how to modularize code, how to create a component tree, how to avoid unnecessary re-rendering due to cyclic state changes in React, how pointers and lifetimes work in Rust. This will help you with debugging, writing clean and growing codebases, etc. What is unnecessary now, though, is the act of writing all of it, from scratch, yourself

I’m sure there are things that Claude Code is not good at, gaps in the training data, things it refuses to do, loops it gets caught in. I’m planning on trying to find those. The point, though, is that while you should not depend entirely on a tool or toolset to develop everything for you with almost no knowledge of how it works, things have, in fact, changed.

The Age of Courage is Upon Us

We’re in a time now where there’s fairly little excuse to Not Be in the Arena. Technical competence is extremely easy to obtain now, not in the sense that you can buy 12 courses for $40.00 each and pretend your way through a job interview, but in the sense that you can now have a mid-level to senior-level engineer pair program with you (Claude Code) while you ask another one functionally infinite questions (Claude, ChatGPT, Gemini, etc.) and, if somehow none of them produce a good enough answer, there are hundreds of blogs, YouTube channels, books, articles and more that you can consume to figure something out.

The time to market has shrunk to almost zero. At this point, agency has become one of the most important traits you could have. Courage is vital. You can do almost anything, in increasingly less time, if you have the courage to really want to do it, the drive to learn along the way and the perseverance to stick with it for long enough.