AI assistants for internal knowledge

01Sound familiar?

• ✓New staff taking months to stop asking where things are
• ✓Senior staff as human search engines for the same questions
• ✓Three versions of the SOP, none marked current
• ✓Answers that leave with the person who knew them

02What an assistant does here

1. 01 Answers with citations

   Direct answers that name the document, section, and update date — so trust is checkable and stale sources get caught instead of repeated.

2. 02 Respects who may see what

   Sources connect read-only and follow the permission model agreed in scope: a general HR policy and a restricted salary document should not be treated alike.

3. 03 Routes the undocumented

   Questions without documented answers go to the domain owner — and their answer can become documentation, so the gap closes instead of recurring.

4. 04 Surfaces contradictions and gaps

   When two documents disagree, the response should show the conflict. Recurring unanswered questions become a practical list of documentation gaps.

02AA concrete view

03Limits and conditions

• ▪An unmaintained source set produces fast but stale answers. Curate the documents and name their owners before connecting them.
• ▪Employee questions are sensitive — analytics should aggregate patterns, never surveil individuals.
• ▪Budget for ownership: each knowledge domain needs someone responsible for freshness, conflicts and removal.

04Packages for this job

05Go deeper

• How AI assistants work: models, memory, knowledge, tools and permissions · 7 min
• Is it safe to connect an AI assistant to your email and business systems? · 7 min