The boring part of an AI agent is the part that works

Описание и применение

Автоматизация маршрутизации инцидентов по электронной почте с помощью LLM и внешнего каталога сократила среднее время решения с 2 часов до 10 минут.

Задача

Incoming incident emails are semi‑structured, noisy, and often misdirected, causing delays of up to an hour before a human can act.

Результат

By normalising emails, constraining LLM output to a fixed enum, and delegating dynamic owner resolution to a directory service, the system achieved a ten‑fold MTTR reduction without retraining the model.

Рабочий процесс

• poll inbox
• normalise email body
• classify incident with LLM
• resolve owner via directory lookup
• create ticket
• notify owner
• observe outcome

Доказательства

I built an agent to collapse that loop for a BSS platform, and it took mean time to resolution on routed incidents from around two hours to about ten minutes.

Источник

The single highest‑leverage thing I did was aggressively strip the email before the model ever saw it.

Источник

The model does one small job in the middle. Everything that makes the system trustworthy is plumbing.

Источник