agent skills are going to matter way more than people think. a new research paper by microsoft validated a lot of the direction behind nua. if you think about it, skills are just compact knowledge about procedures: - how we test - how we structure code - how we design ui elements - how we investigate incidents - how we ship safely etc. and the paper treats skill files as trainable infrastructure. that's exactly the shift i’m building for. a skill file is no longer just a .md file sitting in a repo. it becomes something you can measure, edit, test, reject, improve, promote based on the actual sessions where that skill was used. and today most teams even don’t know: - which skills exist - where they live - which skills an agent used successfully (or not) - who tweaked a skill locally - whether that tweak helped - whether that tweak should become the team standard and that’s the gap i’m focused on with nua. make skills visible. make skills measurable. make skills compound. my bet is that engineering will become less about writing code directly and more about managing the context and skills around the code. and i think most coding agent tools are still approaching this from the old world: code first. context second. cursor, windsurf, claude code, codex, conductor, etc. are all incredible tools but their center of gravity is still the code editor. that comes from assumptions that made sense a few years ago but not anymore. nua starts from a different assumption: the real leverage will be in the context layer the agents work from. code matters a whole less today and not at all tomorrow. sign up for nua beta to try it firsthand. 🧵