# Context Drift Context drift is the gradual, often unnoticed misalignment between what AI context describes and what is actually true about a system, project, or workflow. Unlike [[AI Context Rot]], which is about staleness over time, context drift emphasizes the directional nature of the problem: the context slowly drifts away from reality as the underlying system evolves. Think of it like [[Configuration Drift]] in infrastructure. You start with a perfectly accurate description of reality. Then small changes accumulate. A function gets renamed, a convention shifts, a tool gets swapped out. Each change is minor, but the cumulative effect is a context that confidently describes a system that no longer exists in that form. Context drift is particularly dangerous because it's invisible. The AI still produces coherent-sounding output based on drifted context. There's no error, no crash, just a slow decline in relevance. The [[Law of staleness]] applies: the older the context entry, the higher the probability it has drifted. Mitigation requires treating AI context like live documentation: version-controlled, regularly reviewed, and validated against the current state. [[Context Hygiene]] practices and periodic audits help catch drift before it compounds into full [[AI Context Rot]]. ## References - ## Related - [[AI Context Rot]] - [[Configuration Drift]] - [[Context Hygiene]] - [[Context Engineering]] - [[Law of staleness]] - [[Levels of AI Context Management]] - [[Claude Code Memory]] - [[AI Instruction Drift]] - [[Context Anchoring]] - [[Obsidian Starter Kit - Tutorial - Managing AI sessions]] - Operational guidance to prevent context drift in OSK sessions