Agents didn’t just make Git harder to host. They likely mandate a new layer between humans and code, and that layer now needs its own versioned artifacts. A fe

Agents didn’t just make Git harder to host. They likely mandate a new layer between humans and code, and that layer now needs its own versioned artifacts. A few scaling trends, all triggered by coding agents, make it unsustainable to keep stacking agents on the interfaces designed for human uses — Git, PRs, CI, review: 1. Time to produce a given feature dropped by orders of magnitude. Review, CI, and branch workflows were calibrated to human writing speed. The gates did not get faster. 2. Users will not consume 100x more features. They expect 100x more value from each interaction. The interesting artifact is no longer “the code that implements a feature.” It’s the mapping from intent to outcome. 3. Intermediate work exploded while durable output shrank. Agents leave scratch repos, traces, failed attempts, discarded implementations. The thing the user actually wanted got smaller and more valuable — closer to “satisfy this intent” than “ship this feature.” That is a lure to build much more complex software, and we are versioning the scrap heap with the same tool we use for the product. 4. The primary goal of software engineering till today is to share actions between users, so their intent are served through the same routines. Such software only satisfy commonality. It's now possible to build based on peculiarity. With that, a whole new space of personalization. This looks like the history of machine tools: first powerful and imprecise, then more powerful and enormously precise. Software has to make the same jump — much more useful, and much more personalized. I don’t know what the end state looks like. I doubt Git should remain the central mechanism. Edit: Revised with help of AI.