Chad Fowler argues that as AI can reliably regenerate code from specifications, traditional version control loses its core purpose. In a “regenerable” system, deleting a component and recreating it from stored intent—requirements, constraints, decisions—means the code itself becomes a by‑product rather than the primary record of intent. Diffs still show what changed, but they no longer explain why; the causal history lives in an executable specification graph.

Fowler illustrates this with an email‑validation function: a single requirement change propagates through a plan node and produces new code, making the reason for change explicit. To support such workflows, intent must be stored as a content‑addressed graph where each node (requirement, constraint, plan, generator) has a stable hash and edges capture causality. Existing tools like Bazel and Nix already hash inputs, while formal methods provide precise specifications, but none yet enforce the full provenance loop.

As regeneration costs fall, preserving intent—not just code—will become the decisive engineering discipline.