Hook
More breakout videos from this creator.
Software development has changed so rapidly, it's hard to get your head around it. So I had a conversation with a friend this week, and he's been a developer for a really long time, and he told me he used to read the code and know who wrote it. He could see people in work and and they could see him. Now he says that's gone and he's really grieving the creativity. Coding was where he made choices that were his, was an elegant way to do a thing and that was the craft. Now, you know, he trains the model on his own style and he selects what it be uniquely exposed to. Not because it's simple, because so much of it is patterned, and patterned work is what models take over first. And take it over they when you give most of the coding to agents, the developers write eight times as much code but ship only about 30% more. Over half of agent code never even survives into what's kept. Mosquitoes. So the machine floods the easy layer and the work moves to the hard one. Now, IRX, our index is I think about which is which. Scores work on how much resist is being broken and handed off. The software isn't one score, parts of the jobs at different levels. My friend's style is to him, but it's and reducible by AI. It was in that lower part and code is in three dimensions that explain why. So the first, contested success, does it work? Entanglement, code on its own as one function with no threads between them that AI can't do. And aesthetic judgment, taste, is it we're at stage of tech where I there'd be more taste, but there hasn't been and there's a couple of reasons for this. One is because of vibe coding, it's great for prototypes but has no model architecture, no argued definition, it's good, no one accountable when it breaks. But for a trained engineer, it's lazy. Now he says it, by the time he's played Whac-A-Mole with all the mistakes, he may as well have sat down, what he was trying to do. That's taste and a choice about how to be the author. And agentic engineering is at the other end of spectrum. There's precise specs, a working model of structure and data flow, review, pushback, deciding what's allowed to ship and owning when it fails, and the irreducible parts stay with the person. Now, his taste doesn't disappear, but it does move and it sits now his decisions, like what to build, what to reject, to ship. And AI can't see the patterns in his decisions as it could in his code. So choosing what should exist, framing the problem before any code is written, picking that one elegant version out of 10 that all pass and choosing what not to build. Those are his fingerprints now. So that's where the creativity comes back. It moves from generation to selection and direction. Now it feels like a loss. It is because selecting doesn't feel like creating. And it'll take time for authorship to come back again and to push up that IRX to make more complex decisions, to cross more boundaries, take more accountability for hard things. So next video on is the most irreducible job ever analyzed and why it's structure that really grows a person and what kind of jobs we need in the future.