Agent-Driven Triage that Merges Duplicate Intercom Issue Themes

Uses an agent to do the judgment work that pure similarity scoring gets wrong — deciding when two clusters are really the same issue, merging them, and consolidating their Linear tickets so the canonical theme list stays trustworthy.

Use it when automated clustering produces near-duplicate themes ("login fails" vs "can't sign in") that a human would obviously merge. The agent applies that reasoning continuously instead of letting theme drift accumulate.

1. 1A daily schedule wakes the triage agent.
2. 2The agent loads candidate theme pairs flagged as borderline-similar from Postgres.
3. 3It reads representative Intercom conversations for each candidate pair to understand the actual issue.
4. 4Using judgment, the agent decides merge, keep-separate, or split for each pair.
5. 5For merges, it consolidates fingerprints and counts in Postgres and closes the redundant Linear issue as a duplicate, redirecting links to the survivor.
6. 6The agent posts a short reconciliation log of every decision and merge to Slack for human oversight.