felix's den

First of all, this is the first post I'm releasing on this website. It is not the first one I've written, as I've previously produced other projects and articles that never got a platform to be released on. That has since changed, and I intend to release more of my thoughts and "adventures" as the days go by. This journal has no expected culmination date, and will be a project in constant evolution, just like the website itself.

Few things bring the bored programmer as much intrigue as a prompt for a new project you weren't expecting. As I was hanging out with a friend, he had the idea to write a certain agentic app for his sibling. While I won't go into further details about what we're writing, as it's a shared project (and this is in the thoughts page!), I can't help but try to share some of the lessons learnt by moving fast and breaking things.

Powered by the sweet caffeine rush of Red Bull, and the greasy yet crisp texture of a can of Pringles, we set out to build what a whole team could have taken days to write, in less than four hours. Needless to say, AI was a massive aid in writing upwards of a thousand lines of TypeScript code, which aren't even close to finishing an MVP, let alone an actual app that his sibling's business could use.

That is to say, mistakes were made. My most important take-away from this hackathon of sorts is this: you must understand the actual system whose components are being written by whatever agent you are using, if you choose to code with AI. I don't believe that at this moment, it is possible to write a useful system just by asking your agent to design the system, before asking it to write every single component it chose to recommend.

We are engineers, not just mere vibe-coders. That is, either we design our own system, before perhaps using AI as an aid in the programming of individual components; or we fully commit to every single design decision made by our agent. I found this second option to be the one that actually turns computer programmers obsolete, as it turns the human into an aid for AI, instead of viceversa. It is also a choice that I believe impacts projects negatively.

Despite said mistake, hey, I had a great time. An improvised hackathon, as I like to call it, I think really does have the potential to produce some great programs, concepts, or ideas. I'll do it again someday.