What makes enterprise knowledge ready for AI Agents?

We’re building Assay around a simple question:

Before an enterprise connects AI to its knowledge, how does it know whether that knowledge is actually ready?

We keep coming back to five controls:

1. Source coverage — can the relevant systems be connected without forcing a migration?

2. Authority and freshness — can the system distinguish approved knowledge from drafts and outdated versions?

3. Duplicate resolution — when sources conflict, is there an explicit rule for which version wins?

4. Permission-aware lineage — can every answer be traced back to its source without crossing access boundaries?

5. Human governance — can domain owners verify, correct, and retire knowledge?

What tends to break first in real enterprise AI or RAG deployments?

- Stale or competing sources

- Unclear ownership

- Permission mismatches

- Missing lineage

- No human review path

Assay is launching today as an early-stage interactive product concept. I’d especially value practitioner feedback on what you would add, remove, or prioritize in this framework.