How it works
A loop that runs quietly, not a project with a deadline
Cairn never asks anyone to write anything up. It asks short questions, interprets the answers, and confirms its reading with the person who said it — keeping a living record of how the work actually gets done that your people and your agents can query at any time.
- 01
Cairn asks the people doing the work
Short, specific questions land in Slack or email — one or two at a time, never a form to fill out.
Not "please document your process". Questions like: who signs off on this, which tool did you actually use last time, what did you do when it broke. The bank is tuned to each team's function and each person's tenure, and the cadence ramps down as coverage grows.
- 02
Answers become a map of how work runs
Plain replies turn into procedures, steps, owners, tools, and handoffs — each one traceable to who said it and when.
Cairn proposes its reading and the person who said it confirms or corrects it before anything is recorded. The result is not a document. It is a structured account of how the company, each team, and each contributor operates, which you can query.
- 03
Your agent checks your docs against it
Cairn runs an MCP server. Point the AI agent you already use at it, alongside your knowledgebase's own MCP, and ask where the two disagree.
Your runbook says approvals go through the tool you pay for. Ask your agent whether that still holds, and Cairn answers from what three people said last month: they stopped using it because it is too slow, and now work in a shared doc. Nothing in your knowledgebase can tell you that. Cairn can, and names who said so.
Cadence
It gets quieter, by design
A tool that asks the same volume of questions in month six as in week one gets muted. The schedule ramps down as coverage grows.
Weeks 1–2: discovery
The highest-value gaps first — what each team owns, which tools they use, and the procedures nobody has ever written down. This is the densest period and still lands as a couple of questions a day at most.
Weeks 3–8: filling in
Follow-ups on what the first answers revealed. Handoffs between teams surface here, because they are where two people's accounts stop matching.
Ongoing: maintenance
Questions drop to a trickle and become targeted: a fact has aged past its useful life, a conflict needs settling, or someone changed role. Nobody is asked anything the system can already answer.
Guardrails
What stops this becoming a nuisance
Adoption dies on the first week that Cairn feels like homework, so the limits are built in rather than configurable afterthoughts.
- Quiet hours and time zones per person, falling back to the org default
- Snooze, skip, and a cap on how often any individual is asked
- Batched prompts rather than a steady drip of separate notifications
- Nothing recorded until the person who said it confirms the interpretation
- A page showing everything attributed to you, correctable at any time