Begin with the need, not another agent.
Record the recurring task, quality gap or resource pressure. First improve context, routing or an existing role. A new role, a renamed agent and a newly created execution mechanism are different changes.
When a new substrate is warranted, state exactly what it must do and how a comparison will establish whether it helps.
Creation needs an artifact and an evaluation.
A candidate may be a bounded executable program or another explicitly defined model or execution configuration. Preserve the artifact, its lineage, resource bounds and the test environment.
Recovered program-synthesis experiments demonstrate bounded program candidates evaluated against training and held-out cases. That evidence does not establish autonomous creation of general foundation-model weights.
Homeostasis is the surrounding loop.
Observe pressure, diagnose the cause, make an authorized correction and measure again. Retain the improvement or roll it back, then preserve the lesson.
Emergence is one possible intervention in that loop. Reducing load, changing placement, improving context or repairing a failure may be the better response.
Show the decision, then the outcome.
Morphitas should make the candidate, rehearsal, admission decision and subsequent live outcome understandable. Compacitas owns the runtime mechanisms that apply the approved change and record execution.
The lifecycle here explains the product direction. Available creation methods and operating guarantees depend on the specific runtime release and its validation.