Generative art often starts with a simple idea: you write some code, tweak a few parameters, run it again, and suddenly something unexpected appears. Sometimes the result turns out better than what you had in mind in the first place. That is one of the things I enjoy most about working with algorithms, but it can also be slow and repetitive. Every new experiment takes time, every algorithm needs to be understood before it can be pushed in an interesting direction, and every conversation with an AI assistant often starts with rebuilding the same context all over again.

At some point, I realized this was a real repeated use case. Maybe not the most obvious business case, but still a workflow full of tasks that kept coming back: exploring ideas, adjusting parameters, reviewing outputs, and keeping track of what actually worked. That made it a good opportunity to see what GitHub Copilot could do in a more structured setup. Not just as a tool for occasional suggestions, but as part of an assisted development flow that could cut down manual work and help me spend more time on the creative part.

In this session, I’ll show what that setup looked like and what changed once it was in place. I’ll briefly touch on the instructions, skills, and agents I created, but the main focus will be on the reason behind it and the practical impact. With the right setup around a repetitive workflow, I was able to save time, keep useful context between sessions, and noticeably improve the results.

This talk is for people who build things with code and want their AI tools to support the way they actually work.