The context layer has to know which source wins and how fresh the answer must be for the action at hand.
Agents are vulnerable to stale confidence. They can retrieve an old note, combine it with a newer dashboard, and write as if both have equal authority. Humans do this too, but agents can do it faster and at scale.
Source priority should be explicit. For invoice status, billing wins. For opportunity stage, CRM wins. For product usage, the canonical event store wins. For employee status, HRIS wins. For a customer-facing commitment, the approved contract or support case may outrank a casual internal note.
Freshness is action-specific. A strategic account segment might tolerate a weekly refresh. Payment status may need same-day accuracy. Incident severity may need minutes. A renewal email should not rely on usage data from last quarter unless the agent says so and escalates appropriately.
The AI context layer should package both source and freshness with the fact. For example: "the customer is at risk," plus where that risk came from, when it was last updated, who owns it, and whether it is allowed to drive this action.
When context is stale or contested, the agent should have a defined behavior: stop, ask, escalate, cite uncertainty internally, or use a lower-risk path. The wrong answer is silent improvisation.
This lane overlaps with systems of record but does not replace it. Systems of record decide authority. The context layer turns that authority into usable instructions for agents.
This is part 4 of 10 in The AI Context Layer.
