GitHub and similar platforms are the wrong shape for modern software development. The paradigm of human-centric collaboration—branches, pull requests, code reviews—worked when humans wrote code at human speed. But AI agents now generate freakishly good code rapidly, creating more branches, parallel work, and strain on existing systems.

The bottlenecks—source control, CI/CD, code review, deployments—are compounding because code contribution has expanded beyond engineering to sales, support, and marketing. A 10-minute build now affects everyone. Questions about code ownership, trust, review, integration, and token costs remain unanswered.

We must rethink software delivery not as human collaboration but as high-throughput infrastructure primitives: source control, execution, artifacts, caching, identity, policy. Like cloud-native compute, storage, and networking, these composable building blocks enable machine-scale generation, validation, and deployment. The winners won't build better pull requests—they'll build infrastructure for agentic engineering.