Turn internal office hours into FAQ updates
Stand up an **office-hours harvest agent** that takes raw questions from a live session, a Slack thread, or meeting notes and emits **normalized FAQ rows + draft handbook language + gap flags**. Merge only after human review so your internal Q&A agent ([Build an internal Q&A agent for your team](/tutorials/build-an-internal-q-and-a-agent)) cites fresher truth.
Plus: three Admin-Agent passes: collapse duplicate questions from one messy paste, diff draft answers against existing handbook context so contradictions surface before publish, and draft a short **what changed** blurb for Slack.
| Audience | Everyone · Admins |
|---|---|
| Time | ~12 min |
| Prerequisites | You already run or collect internal questions ([Build an internal Q&A agent for your team](/tutorials/build-an-internal-q-and-a-agent), or you plan to). Context discipline ([Add your first context file](/tutorials/add-your-first-context-file)). Agent basics ([Create an agent from scratch](/tutorials/create-an-agent-from-scratch), [Give your agent its job description](/tutorials/give-your-agent-its-job-description)). Helpful: [Connect Slack to your agents](/tutorials/connect-slack-to-your-agents) when questions live in channels; [Turn meeting notes into action items and decisions](/tutorials/turn-meeting-notes-into-action-items) when office hours produce notes instead of a raw thread dump. |
| You'll end up with | One **FAQ Harvest** agent with stable instructions: a markdown table (**Question**, **Answer draft**, **Handbook pointer or GAP**, and **Owner**), plus a repeat ritual where you paste, the agent proposes patches, you edit context files yourself, and nothing publishes without you. |
When a tutorial shows italic text in quotation marks, it usually mirrors a label or helper string inside Auxot. Product copy changes between releases — if something reads differently in your workspace, trust what you see on screen.
Callouts with a Worth knowing gold accent are meant as must-read context before you move on. Blockquotes that open with Tip are lighter, optional depth.
Why this matters
Office hours surface what people actually do not know, often ahead of any analytics dashboard. The failure mode is familiar: great answers live only in chat, and next week the same question lands on the same human.
This lesson is not another generic FAQ generator. It is a human-in-the-loop bridge: you bring the messy transcript (Turn meeting notes into action items and decisions when notes exist), the agent proposes handbook-shaped rows that could feed the corpus behind Build an internal Q&A agent for your team, and you decide what merges into context (Add your first context file).
Auxot does not silently scrape channels: you paste bundles because you prompted. A scheduled ping (Run an agent on a schedule) can remind you to harvest; it still starts with your paste.
Agents propose language: owners keep policy honest.
Quick start
- Name your handbook slices. Which context files count as “policy truth” for internal Q&A (Build an internal Q&A agent for your team); naming them stops drafts from inventing fake filenames.
- Create the FAQ Harvest agent. Its instructions demand the four-column table plus a SOURCES_QUOTED footer tied to what you pasted; the agent refuses to run when the paste is empty.
- Run one real session. Paste a Slack export snippet, live notes, or bullet list; redact personnel identifiers first.
- Review rows like code review. Wrong tone gets fixed in Chat; factual claims must map to an existing pointer or GAP: needs owner.
- Merge manually. Edit the handbook context file yourself; then spot-check the Q&A agent on two questions that came up in session.
Done? Three dated FAQ harvest markdown files in your archive; each lists what changed at the top; comparable month over month.
The agent can do that?
1. Collapse duplicates from one paste
Chat → Admin Agent:
Questions pasted (messy): […]. Output a markdown table with columns Question / Answer draft / Handbook pointer or GAP / Owner guess. Merge near-duplicates. Max twelve rows. Cite which pasted lines supported each merge.
Why it’s non-obvious: Without collapse, FAQ noise repeats; your handbook grows paragraphs nobody searches after you paste once and ask for grouping.
2. Contradiction sweep against live context
FAQ table: […]. Handbook excerpts (paste): […]. Flag **CONTRADICTS**, **EXTENDS**, or **NEW_TOPIC** per row. Do not silently rewrite the handbook. Bullets only.
Why it’s non-obvious: Fresh verbal answers conflict with last quarter’s PDF; surfacing tension early beats shipping confident dual truths (Keep your context files honest and fresh).
3. Slack-ready “what changed”
Approved rows only: […]. Draft 120 words or fewer in a calm tone, ending with where to read full policy. No emoji pile. Plain markdown.
Why it’s non-obvious: Silent handbook edits confuse; a short channel note trains a verify-the-source habit (Collect useful feedback on your agents when people notice the handbook and what the Q&A agent now says don’t match).
Go deeper
Review pipeline
Treat outbound FAQ blurbs like any reviewed artifact (Set up a content review pipeline): especially when legal or HR-adjacent.
Risky topics
Compensation, investigations, and one-off exceptions: keep refusal language in the Q&A agent itself (Build an internal Q&A agent for your team); harvest rows should flag HUMAN_ONLY, not draft clever shortcuts.
Workflow hook
Optional board column Harvest FAQ (Run a workflow): human collects paste, agent step formats, approval step before merge.
Privacy
Office-hour transcripts may contain sensitive stories: align retention and sharing (Run a data privacy review before you ship) before posting summaries widely.
Walkthrough
Step 1: Pick one handbook file as pilot
Single source file beats seventeen scattered stubs; depth first.
Step 2: Paste synthetic messy questions
Practice redaction: names → roles; remove identifiers.
Step 3: Tune GAP vs Answer draft
Drafts that try too hard invent policy; tighten until GAP wins when the excerpt is missing.
Step 4: Pair with internal Q&A regression
After merge, ask the live Q&A agent the same three questions; citations must point at your updated sections.
Step 5: Retro quarterly
Archive links next to Run a quarterly review of your agents materials; how often FAQ harvests ship is a useful signal of how healthy the broader agent program is.
What’s next
- → Build an internal Q&A agent for your team. Harvest output feeds the corpus this agent cites; keep charters aligned.
- → Adapt copy with human review. When the same answers ship in more than one voice or locale; same human-review pattern as FAQs.
- → Set up a content review pipeline. When FAQ language needs formal approval before Slack.
- → Keep your context files honest and fresh. Recurring freshness passes; harvest plugs gaps; ownership keeps content from going stale.
- → Turn meeting notes into action items and decisions. Office hours often end as notes first; synthesize and paste unresolved question lists here.
- → Onboard a new teammate to your agents. Point new hires at updated FAQ surfaces; same week as harvest ships.
Reference
- Pages in Auxot: Chat, Settings → Agents, Settings → Context Files, and optional Workflows
- See also: Roll up initiative threads into weekly goal check-ins, Adapt copy with human review, Add your first context file, Connect Slack to your agents, Require human approval before risky actions, Run an agent on a schedule