AI agent team for operational work
Three agents with roles like in a company: correspondence and deadlines, infrastructure operations, modernization.
A recording and screenshots of the panel running on sample data: a made-up task log and made-up agent results, no connection to the database or network; missing Polish interface labels translated, the schedule list hidden in the recording.
Task
Routine tasks (alerts, mail, deadlines, updates) took time every day, and a single general assistant mixed up responsibilities.
Solution
Each agent has its own role, skills and decision memory in PostgreSQL. Change proposals are reviewed by an agent council, and decisions that need a person land in the panel as cards with options. Every action is written to a log.
Result
- Alerts resolved overnight, only a report in the morning
- Cheap tasks on a light model, hard ones on a strong model
- Full log of agent actions
My role
Architecture, agent code, runbooks and tests.
