When a product changes the model executing a request, I want that switch to be visible when I need it. At the Fable 5 launch on June 9, 2026, Anthropic said sensitive requests would be routed to Opus 4.8. That safety route was part of the product. The name in the selector therefore could not explain where every answer came from.
Later, on June 17, 2026, I asked for an agent to follow the other agents' work and keep the common objective in view. I wanted to distribute work without becoming the person who had to gather every loose end afterward. My request concerned task coordination; the announced fallback concerned safety. The connection I make now is the transfer of responsibility between executors.
A transition can work without attracting attention. When an answer changes unexpectedly, though, I need information to investigate. Which part was forwarded? Did the next executor receive the whole objective or a fragment? Who checks that the response still answers the request? Hiding this even from somebody trying to review the work creates an irritating dependence on what the interface chooses to reveal.
I do not need a screen showing every internal connection whenever I ask something. I want details available when they help. An indication of the route and the material handed over provides a better starting point than guessing from writing style. Identifying a model through its punctuation sounds like a terrible use of my afternoon.
In the coordination setup I want, an executor receives a bounded assignment and the coordinator checks the delivery against it. Confident messages do not automatically make that connection. If I have to reconstruct the original request and discover where every result belongs, I am still doing the coordination I tried to delegate.
Reformulating a problem can help solve it. It can also change the request without anyone noticing. The coordinator should examine that distance when an answer comes back: does the delivery resolve the assigned task or a more convenient version? The response needs to return to the original objective before anyone marks it complete.
My acceptance criterion for this kind of workflow is preserving the objective and definition of completion through the switch. The executor can change; the delivery still owes an answer to the request. I will require a traceable handoff and an explicit check on return, without reconstructing the entire conversation to understand what happened.