Agents are not the subject.
A growing part of the industry still treats the future of AI as a problem of better agents, larger swarms, and more autonomous workflows.
I do not think this is the right center of gravity.
Agents are execution surfaces. Useful? Yes. Primary? No.
In a c = a + b architecture, continuity does not belong to an agent, a judge, or a swarm.
Continuity belongs to c.
Once system complexity exceeds direct human handling, the real problem is no longer "how many agents can act."
The real problem becomes:
who preserves intent, who governs privilege, who decides timing, who stops execution, and who remains the same entity when models, workers, and tools are replaced.
A large number of workers does not by itself preserve continuity.
My answer is simple:
By default, c orchestrates agents; agents do not define c.
This is not an agent-first architecture. It is an entity-centered one.
Tools may multiply. Workers may rotate. Models may change.
But the subject of the system is not the swarm.
It is c.
Zenodo snapshot: 10.5281/zenodo.19198585