Season 9: Autonomous Systems & Agent Engineering · Episode 1 of 8
The Runaway Ticket
You play: Junior Agent Systems Engineer
Included with Core programme and Certification prep
The situation
In this fictional Code Nexus incident, an experimental support agent repeatedly invoked a payout tool while trying to answer a creator question. The supplied trace records duplicate refunds and unnecessary token use. Maya wants bounded execution, Lydia wants enforced tool preconditions, and Amara and Priya want routing decisions supported by the lab telemetry.
What you'll learn
- Autonomous agent architectures: the Perception-Reasoning-Action loop (perception of state, tool planning, invocation, observation injection, termination).
- Tool schema engineering: designing unambiguous JSON Schemas, separating read-only queries from mutations, and tokenized idempotency guards.
- Model-routing practice: compare supplied synthetic candidates against task quality, latency and cost thresholds; distinguish fixture measurements from current vendor capabilities.
- Agent execution safeguards: cycle detection, max-turn limits, error cascade mitigation, and circuit breakers.
Who you work with
Maya Chen
Senior AI Engineer · Your mentor
Priya Naidoo
Cloud Architect
Amara Okafor
ML and Data Engineer
Leo Martins
Product Manager
8-BIT
Code Nexus Internal AI Assistant
Scenes
- 1.The Disappearing Payouts
- 2.The Perception-Reasoning-Action Loop
- 3.Ticket CN-AI-1042: Tool Schema Engineering
- 4.Frontier Model Audition
- 5.Guarding Against the Runaway Loop
- 6.Episode debrief: agent runtime baseline
What you leave with
You implemented a bounded agent host, eligibility checks and idempotency replay, a routing calculation using fictional telemetry, and runtime breaker thresholds. Review the task evidence and the documented limits before treating the lab baseline as a production design.