Spec Drift — Spec drift is the gradual, usually silent divergence between what a specification says a system should do and what the code actually does. Because specs rarely fail loudly when they fall out of date, drift accumulates until the documented intent and the running behavior no longer match.
Every change is an opportunity for the implementation to move away from its spec: a shortcut here, an undocumented edge case there. Individually these are minor; compounded, they mean the spec can no longer be trusted, onboarding misleads, and audits fail. Specs rot silently.
AI agents accelerate drift because they generate change quickly and don't inherently update the surrounding intent. Detecting drift requires continuously comparing the spec (business and technical intent) against the current code and flagging where they disagree.
Skaftor treats specs as living artifacts linked to code in the engineering memory graph and runs drift detection as a background automation, surfacing divergences in a drift inbox before they become incidents.
By maintaining a live link between each specification and the code that implements it, then continuously comparing them — flagging cases where the code's behavior no longer matches the documented intent. Skaftor runs this as an automated drift check and routes findings to a drift inbox.
Engineering memory is an organization's persistent, queryable record of how and why its so…
In software delivery, a change's blast radius is the full set of components, contracts, se…
The Orchestration Gap is the widening distance between how fast AI can generate code and h…
Release confidence is a composite readiness score that answers 'should we ship this?' by c…
Last updated July 28, 2026 · https://skaftor.com/glossary/spec-drift