Shadow AI — the personal tool used on the sly, the private account, company code leaking out uncontrolled — is not an innocent shortcut: it breaks the principle Agile Bedrock calls algorithmic transparency. Classic transparency dealt with the state of the work; this one deals with its origin: the team must be able to know which decisions and outputs were human and which were assisted, and which guardrails and autonomy limits govern the assistance. It is not about understanding the machine's inner workings, but about being able to inspect the origin of what is produced, because if you cannot inspect where a decision comes from, no empiricism about it is possible. Hidden assistance leaves the team not knowing where its own output comes from, and with that it removes the very basis of inspection and adaptation.
Want to keep pulling the thread? This topic unfolds in Agile Bedrock.