Developer abandons a design system where Jev made all the small decisions
He built a design system in which Jev decided every small call, then dropped it, concluding that restrictions hurt once models get strong enough.
It’s an interesting time to be building things. Previously I spent a lot of time curating context and tools to try and massage results from LLMs. With the pace of change you never know when something will become obsolete. Opus 5.5 made this report for me with no skills or special tools and a three prompt iteration. It’s a very large step up and aligned with what used to require jumping through a lot more hoops. More and more I feel that foundational things are all that’s required. I built an agent orchestration system and realised that herdr by itself worked better. LLMs like to apply rules to systems but with newer models restrictions always seem to hurt and flexibility wins. I wonder if software will become obsolete and all you will need is an AI assistant that is smart enough to build anything it needs for a task on the fly and in a way that works best for it. I tested a design system based on Jev deciding all the small decisions and abandoned it. We may need to let go of the idea that we have some special ability, skills or knowledge to know better ways for AI to do stuff.