Govern · Insight elevation
A correction, taught once, remembered everywhere it's relevant.
Most AI tools forget a correction the moment the conversation ends. Augela asks a specific question every time one happens: does the rest of the company need to know this too?
Why this isn't automatic
Not every fix should become a rule.
Auto-generalising every correction is how an AI system quietly drifts. One unusual edge case becomes a rule applied everywhere. Insight elevation puts a person with a specific permission, a Tutor, between "this was corrected" and "this is now how the company does it."
Four outcomes, one decision
What a Tutor can do with a single correction.
- Elevate
- The lesson becomes real, indexed company knowledge. The next agent that needs it already has it.
- Coach
- A private lesson for the one agent involved, not generalised to the rest of the company.
- Reconfigure
- Points at a deeper fix, such as a guardrail or an AI Profile, rather than teaching one fact.
- Reject
- Not every correction should generalise. A Tutor can decide this one shouldn’t.
The correction itself usually starts life in human-in-the-loop review.
FAQ
Companies rolling this out are asking
Who decides what gets elevated?
A person with the Tutor permission, which is a separate, explicit role from an ordinary reviewer. Elevation changes company-wide behaviour, so it is not a side effect of an ordinary approval.
What are the four outcomes a Tutor can choose?
Elevate (becomes shared company knowledge), Coach (stays a private lesson for one agent), Reconfigure (points at a deeper guardrail or profile fix rather than a one-off answer), or Reject (this correction should not generalise at all).
Does elevating something happen instantly?
The content is indexed as real, searchable knowledge the moment it is elevated. It is not a note filed away for someone to act on later.
Teach it once. Decide, deliberately, whether everyone else should know too.
Twenty minutes with the founder. No slides. The real product.
