What the AI platform does
The three AI responsibilities
How trust is built into the system
- Source-backed responses. Important answers should identify the underlying project records, documents, communications, or packets that informed them.
- Bounded tools. AI tools have typed inputs and scoped capabilities rather than unrestricted access to the database or external systems.
- Preview before write. Actions that change records follow a review and confirmation path instead of allowing model text to mutate data directly.
- Visible operational state. Ingestion, retrieval, and AI runs record their status so missing sources and degraded answers can be investigated.
- Human review for learning. AI suggestions and learned workflows are promoted through reviewable records rather than silently becoming permanent behavior.
Where AI work runs
What AI is not allowed to do
- Treat unverified generated text as a source of truth.
- Quietly substitute empty or stale source data for an answer.
- Write directly to business records without deterministic validation, permissions, auditability, and the required confirmation path.
- Hide an operational failure behind a generic success message.