Advisory-Only Posture
The AI can analyze, explain, compare, suggest, and guide — it cannot write to workbook tables, execute calculations, or perform any state mutation; this is a hard architectural boundary, not a mode setting.
Governance of AI assistance in project control contexts requires more than a code-of-conduct statement about what the AI will and will not do. In a system where AI output directly influences cost and schedule records that underpin multi-million-dollar project decisions, the boundary between AI analysis and AI action needs to be enforced architecturally — not held in place by convention or runtime filtering.
The Alfred Pattern
ProjectXL's AI assistance applies the Alfred Pattern from the platform's UX design doctrine: the AI is a context-aware subject-matter companion whose purpose is to amplify understanding rather than to own mutation paths or reroute users from their primary workflow surface. The AI analyzes, explains, compares, suggests, and guides. It does not write to workbook tables, execute schedule calculations, capture snapshots, set active baseline designations, or perform any other state mutation on the user's behalf.
Structured Advisory Cards
When the AI has a suggestion that involves a follow-on action, it returns that suggestion as a structured advisory card. The card identifies the target of the suggested action, the preconditions that must be satisfied for the action to be appropriate, and a bounded estimate of the impact. This structure makes the suggestion inspectable before any action is taken — the user can evaluate whether the suggestion is correct, complete, and applicable to their current situation before deciding whether to accept it. Accepted actions route to the owning feature workflow where validation and governance rules apply exactly as they would for a user-initiated action.