Macro Trainer Agent: Self-Directed Mining of Intercom History into a Reviewed Macro Library

This is the agent-run version of macro training. Rather than a fixed pipeline, a worker investigates your resolved Intercom history on its own: it samples threads, identifies the themes worth canning, checks each candidate against macros that already exist so it doesn't propose duplicates, and assembles a clean, deduplicated macro library. The finished set is published to a Confluence page your team can read and adopt.

Use it for a deeper, judgment-heavy pass than a clustering pipeline gives you, especially when topics overlap, phrasing varies, and you want the worker to reason about which replies are genuinely distinct and worth maintaining.

1. 1A chairman or operator kicks off the agent run.
2. 2The agent queries Intercom for resolved conversations and samples across topics and time.
3. 3It groups and reasons over the threads to decide which recurring problems merit a canned reply.
4. 4For each candidate it checks the existing library and merges or discards near-duplicates.
5. 5It writes final macro titles and bodies grounded in how agents actually resolved each issue.
6. 6It publishes the curated macro library to a Confluence page for team review.