Model Choice Should Be Easy to Review and Change

Where model choice becomes hidden risk

A workflow may use one model for planning, another for extraction, a local or private model for sensitive content, and a fallback when a provider is unavailable. If those choices are hidden in code, every cost, privacy, latency, or quality change becomes risky.

Cost, privacy, latency, and quality decisions become hard to govern when model routing is buried in code. The team cannot confidently move routine work to local or private small models if nobody can see which step uses which provider.

Make model choices reviewable

Stable Harness keeps model choices reviewable so teams can inspect which routine steps can run on smaller models, which steps need stronger reasoning, and what fallback exists before users depend on it.

The product should let teams inspect which model is used where, which routine steps can safely use smaller models, why fallback exists, and who can approve a change.

Model questions before expansion

The buyer should be able to change models without rewriting the workflow.

flowchart LR
  A[User pain] --> B[Bounded workflow]
  B --> C[Reviewable action]
  C --> D[Operational evidence]
  D --> E[Repeatable outcome]

What to check before switching

A buyer should ask who can approve model changes, how the workflow proves which model was used for which step, and whether small-model routing is backed by validation rather than hope.

The user-facing value is freedom to improve cost, privacy, and quality without destabilizing the product or pretending small models should handle every task.