Letter
Before the model, there is the workflow
A working AI system starts by naming the repeated work, owner, source, handoff, failure path, and proof gate before choosing model behavior.
Thesis
The model is usually the visible part. The workflow is the part that decides whether the system matters: what work repeats, who owns it, what source is trusted, what output is useful, and where the handoff fails today.
Operating shape
A serious roadmap names the before and after state of the work. It separates source intake, review, decision support, escalation, approval, and delivery so AI is attached to a real constraint instead of a broad promise.
Boundary
This does not mean every workflow needs AI. Some workflows first need clearer ownership, cleaner source shape, a simpler interface, or a better proof artifact before any model behavior should be designed.
Where it shows up
The CaseOps records are useful because they split upstream preparation from downstream review. That split keeps the public claim inspectable: source shape first, model behavior later, product promise only after proof.
Takeaway
What to do with it.
Start with the workflow that creates pressure, not the model choice.