Automation and applied AI

Sagas, not scripts

A provisioning run or a publishing cycle is a distributed transaction: every step has a compensation, a retry ceiling and persistent state. If it falls over halfway, the system knows where it was and what to undo — it doesn't start again hoping nobody notices.

Every step is idempotent by construction: re-running it must produce the same result, because sooner or later it will be re-run.

A real queue, not a hopeful cron

Jobs live in a queue with exclusive claims (in PostgreSQL, skipping rows already taken) so two workers never do the same thing twice. The queue is drained on a schedule and also right after an insert, so nothing waits for the next tick without reason.

AI where it adds value, humans deciding

LLMs are useful for understanding and proposing: draft replies, classification, extraction from documents. But while an action is public or irreversible, the proposal goes through an approval — in the app or as a message with two buttons — and stays on the record.

The value is not "AI does everything": it is removing the mechanical preparation and leaving the decision to a person, with the context already laid out.