The Feature That Kills the Wiki
Every wiki, every Confluence space, every README in every repository shares the same fate: someone changes the system, doesn’t update the docs, and now the docs are actively harmful. This is an unsolvable problem in a human-maintained knowledge base because the maintenance cost is invisible and the consequences are delayed. In a system where AI agents are both the consumers and producers of knowledge, drift detection becomes possible, and automatic.Four ways drift gets detected
Agent-reported drift
When a coding agent loads a blueprint and discovers the code doesn’t match what the blueprint says, it flags the inconsistency. This happens naturally as part of the execution plan step. The flag is a first-class event in the system, not a comment someone might miss.
Decision-triggered review
When a decision is resolved that affects an existing blueprint, the system marks that blueprint for review. A human or AI agent updates it. Until it’s updated, the blueprint carries a staleness warning that agents can see.
Implementation-triggered review
When a work item is completed that modifies files governed by a blueprint, the system prompts: “WI-4 modified the deployment pipeline. Blueprint
deployment may need updating.”Scheduled audits
An agent periodically reads each blueprint, compares it against the actual codebase, and reports drift. This is a background operation that runs continuously, not a quarterly documentation review that never happens.