Coordination is not conversation.
Chat asks one model to infer who is speaking, what it may see, where it may run and what it may do. R26D makes each of those an object you can point at. Five questions are answered on screen at all times: who is here, who is being addressed, what context they receive, where the work executes, and what they are allowed to do.
Five questions, answered on screen at all times.
-
01Who is here?
People and Consultants are both in the room, and both are named. A Consultant is an identity with a runtime and a capability manifest — not an avatar over a model.
-
02Who is being addressed?
The floor is explicit. In human discussion a Consultant receives nothing at all — you choose who is listening, and the room says so where nobody can miss it.
-
03What context are they receiving?
A Consultant is handed a Brief, never the transcript. You grant each message, file, decision and memory entry — and this meter is the payload that is actually sent, not a summary of it.
-
04Where does the work execute?
A Runtime decides what is reachable and caps what any Consultant may do there. A CTO holding WRITE is read-only on the secure box, and the room enforces it rather than trusting it.
-
05What are they allowed to do?
Conversation is not authority. Work waits for a person, who approves the actual operations — and every one that runs is recorded, in order, with what it did or why it was refused.