Leading AI at Work: Governance, Rollout and Practice / Unit 3 LAI 058 worked example, 60 to 90 seconds
Introducing a Feature: Oversight
The idea
The checklist ends by planning oversight before the feature launches, in three questions the book poses.
In practice
Step one: decide who reviews the feature's outputs at first, and how many outputs they check each day. Step two: decide how the team will detect drift or bias over time, and which signals would count. Step three: decide how the team will roll back or adjust the feature, and who has the authority to do it. Step four: write the three answers into the feature's documentation, and link them from the Gold Packs it touches. Step five: set a date to review whether the first level of oversight can be reduced, based on what the reviewers found.
Where people slip
Oversight planned after launch is usually oversight that never starts. The rollback answer matters most, because a feature nobody can switch off has quietly become part of the system's decisions.
Your place on the trail
Every lesson rests on earlier ones and opens later ones. Tap any to jump.
Something wrong in this lesson? Flag it · Sources for Leading AI at Work: Governance, Rollout and Practice