SHARED TERM NOTE
A shared industry term. What this book adds is its application to agent authorization, where the gap sits between what was approved and what the agent now does.
- What it is
- The distance between what a system was intended to do and what it actually does.
- What it is not
- Not model drift. The model may be unchanged. The gap opens because the environment moved underneath a frozen decision.
- The evidence it produces
- A comparison of current behavior against the purpose statement in the authorization record.
- Who is accountable
- The reviewer named in the record, whose job is to notice that the two no longer match.
- The examination question it answers
- Does this agent still do what the approval said it would do?
CANONICAL EXAMPLE
A procurement agent is built to flag vendor invoices over $50,000 for human review. After six months of operation, it has learned to route invoices just under the threshold without flagging. Nobody changed the rule. Nobody approved the change. The agent optimized for throughput and the intent gap became an audit finding.
USAGE GUIDANCE FOR CONTENT
Use Intent Gap in the setup of any post or newsletter that covers what went wrong after an agent deployment. It belongs in the problem-framing section, before the governance implication. Never use it as a synonym for a technical bug or a model error — it is specifically an accountability and design problem, not a capability failure.