Carl
The layer that decides what work happens
The work layer. It owns priorities, delegation, summaries and what reaches a person. It does not run processes, which is AOS's job.
Why it exists
If every product invents its own way of being told what to do, every product invents its own way of being wrong.
What it is meant to be
- Model agnostic. External models, not a frontier model trained here
- Products integrate against an interface boundary, never a model SDK
- Human override is a requirement, through independent systems
- Deciding work and owning processes stay separate. Carl decides, AOS enforces
- A test reads the runtime's own source and fails if work vocabulary appears in it
Current milestone
Panel v1 frozen and verified. The agent runtime's first slice is built and tested. A shared room is hosted and in use. The agent group is still not deployed.
- CompleteCommand Panel v1, frozen and verified
- CompleteAgent runtime, first slice
- CompleteA shared room for people and agents
- LONG-TERMA deployed agent group
Long term direction
One work layer across ME systems, with an override that does not depend on Carl being correct.