Your Agent Is on Probation: Agents Lifecycle explained | Zürich AI Meetup

​The demo looked perfect. Then we opened the traces.

​The agent repeated the same search 14 times, bounced work between subagents, and confidently booked the wrong flight. Its dashboard still reported success.

​We’ll instrument an agent with OpenTelemetry, build repeatable checks with DeepEval, and place a proxy between the agent and the model so we can inspect and change its behaviour without retraining it.

​You’ll learn how to:

• ​Follow requests across tools, models, and subagents

• ​Catch loops, broken handoffs, unsafe outputs, and hidden costs

• ​Understand impact your agent is creating

• ​Add proxy rules that block or rewrite problematic calls

• ​Replay the same tasks after every change

• ​Decide whether an agent should be promoted, coached, or fired

​Bring your laptop. We’ll investigate the agent together and improve it live.

​Workshop lead: Mike Koźmiński, founder of Zürich AI Meetup.