The Cliff Was the Context Window
A small model stopped calling its tools. It looked like a tool-count limit; it was a silent prompt truncation.
Identity, authorization and delegation: questions to ask before putting AI agents into operation.
An agent action should be traceable to its authorizer, permitted scope and execution evidence.
This reading of McKinsey’s Technology Trends Outlook 2026 organizes governance questions around non-human identities. It examines how permissions are granted and delegated, and who can intervene when actions diverge from expectations.
This is a secondary reading of an industry report, not an original market study or proof that no solutions exist. Open questions do not imply implemented Kirux features.