A platform team already carrying a full delivery load gets a line added to its charter: AI-enabled. The slide calls it an uplift. The team calls it Tuesday.
No new headcount. No change to the on-call roster. No new owner named for the one decision that actually matters now: when the model’s output disagrees with the engineer’s judgement, who wins.
Six weeks in, that question arrives for real. The model recommends a config change the on-call engineer thinks is wrong. There’s no answer in the charter, because the charter was never touched — only the task list was. The engineer overrides it at 2am, logs nothing, goes back to bed. Nobody asked them to log it, because nobody wrote the step where that override gets reviewed.
This isn’t a training gap. The team can explain exactly what the model does. What they can’t explain is who’s accountable when it’s wrong and a person was right — because that line was never drawn. It didn’t need to be before. The team’s boundary used to match its decision rights exactly: they owned the systems, they owned every call about the systems. Add a second judgement into the loop and the boundary stops matching anything.
Most AI upskilling programmes teach people to use a tool. Few touch the thing that actually breaks: a team’s decision rights were sized for one kind of judgement in the room, and now there are two, and nobody repriced the boundary.
The charter changed in a sentence. The org chart underneath it is still the one built before the model showed up.
— Signal / unfiltered observations from the field. Fortnightly.
#EnterpriseArchitecture #TeamTopology #AIStrategy #OperatingModel #OrdoAnimi