KB Article Drafting with AI agents

When the same support question comes up often enough to justify a knowledge base article, Neotask notices the pattern and drafts the article itself — pulling the actual resolution steps from resolved tickets, checking the existing KB for overlap, and posting a draft for a support lead to review rather than publish blind. Teams stop losing tribal knowledge to whichever agent happened to solve the ticket last, and the KB grows from real resolutions instead of someone remembering to write it up later.

How it works today vs. with Neotask

Knowledge bases decay because writing them competes with actually resolving tickets, and 'write it up later' rarely happens once the fire is out. The knowledge that should become an article already exists — it's in the resolved ticket thread, in the exact steps an agent took — but converting that into a clear, general article is a separate task nobody has slack time for. Neotask watches for the signal that an article is worth writing (the same underlying issue resolved multiple times, or a single complex resolution likely to recur) and drafts directly from the actual resolution transcript, so the article reflects what actually worked rather than a generic idealized version of the fix.

The agent flow

Detect a recurring pattern

Neotask monitors resolved tickets in Intercom for clusters of similar issues — same error message, same feature area — that have come up more than a set number of times without an existing KB article.

Integration: intercom

Pull the actual resolution

It reads the specific steps the agent took to resolve the ticket, including any internal notes, rather than inferring a generic fix from the ticket subject alone.

Check for existing overlap

The existing knowledge base (Notion or Guru) is searched for a near-duplicate article first, to avoid publishing a second article that fragments the same answer across two places.

Integration: notion

Draft in the house voice and format

A draft is written in the team's existing KB structure — problem statement, steps, and a note on when this fix doesn't apply — matched to the tone of recently published articles.

Integration: guru

Route for review, not auto-publish

The draft posts to the support lead's Slack channel with the source tickets linked, so review is quick — confirm accuracy, not write from scratch.

Integration: slack

Link back from future tickets

Once published, Neotask starts suggesting the new article to agents handling matching tickets, closing the loop from "this happened enough to document" to "now it resolves faster."

Variations

• Product-led teams add a stage where a drafted article gets tagged for engineering review if it describes a bug workaround, so the KB doesn't quietly institutionalize a fix that should really be a code change.
• Teams with a public help center route drafts through an extra tone/formatting pass before publish, since public articles have a different bar than an internal wiki page.
• High-ticket-volume teams lower the recurrence threshold that triggers drafting, accepting more drafts to review in exchange for catching patterns earlier.

Frequently asked questions

Does it publish articles automatically?

No — every draft goes to a human for review before publishing. The goal is removing the blank-page problem, not removing the review step.

How does it decide something is "recurring enough"?

You set the threshold — commonly three or more similar resolved tickets within a rolling 30-day window — and it's adjustable per product area.

What if the resolution steps varied between tickets?

It surfaces the variation in the draft rather than picking one arbitrarily, so the reviewer can decide whether that's a genuinely different case or the same fix described differently.

Can it update an existing article instead of creating a new one?

Yes — when it finds an existing near-duplicate, it proposes an update to that article (e.g. an added edge case) instead of a new one.

Plans

Continue