splash

Posted on August 12, 2026

This is another episode of the pilling story i have been telling here. These days i made my first game: splash, a quickly scaffolded clone. It is not noteworthy, yet it is a milestone. Being a gamer is the seed of my decision to study computer science, so it is striking to finally make one after decades.

Development was a rush: one and a half days non-stop. The game exists because of artificial intelligence, and i was impressed by the power of agentic development throughout. This time i feel i pushed agentic development to its limits, at least to the extent i can personally access. When the sprint was over, there was a playable game.

Much of the credit goes to opencode. I have been a free software enthusiast for the best part of my career, and i would not have done this with proprietary tools.

The milestone is personal, but the experience has also transformed the way i work, and i want to look at that side. While i was inspired, i ran a couple of agents at a time, on different projects. I was way less aware about the code structure, and so far i provided only one or two suggestions about it. In splash the agent kept recurring to ifs, privileging fast development, and never asked for a refactoring.

The way i relate to the code is different. The internals are largely unknown to me, written by something i did not see doing it. Bugfixing for instance: when something does not work, i can no longer open the code and hunt for the problem. I describe the symptom, the agent digs, we go back and forth. It is slower, and i am less sure of the result.

The shift in development experience has been dramatic. It is hard to admit, but this fundamentally changes the way i think about code. Despite that, i am deeply impressed. Developing will never be the same.

The feeling is hard to place. I give directions in broad strokes, then play the game to check whether it feels right. I notice what is off, point at it, and the agent fixes it. It feels like being the one who decides what the game should be and the one who plays it first, without being the one who builds it — halfway between a manager and a tester of what used to be my role.