quiz · 4 min

Production & Deployment Check

Why does a production agent handler reconstruct its state on every request rather than keeping it in process memory?
Processes restart between requests, so in-process memory is lost; the handler must be statelessIn-process memory is too slow for production workloadsThe mock LLM seam cannot share state across requestsEnvironment variable loading clears process memory on each call
In the production handler built in 'A Production-Shaped Handler', what does handle() return when the budget is exhausted?
The raw model text from the policyA structured result with status 'rejected' and reason 'no_budget'An exception raised to the callerAn empty string so the caller decides what to do
When you swap the mock policy for a real model in production, what must change inside the handle() function?
The budget check must be removed because real models manage their own limitsThe guardrail logic must be rewritten to match the real model's output formatNothing - handle() stays identical; only the implementation behind the seam changesThe trace list must be disabled to avoid leaking real model responses
According to 'Going Further', what happens to your evals and guardrails when you migrate to a new framework or SDK version?
They must be rewritten because frameworks define the eval and guardrail interfacesThey are automatically upgraded by the framework migration toolThey are unaffected - evals and guardrails live outside the implementation and survive SDK churnThey become redundant because modern frameworks include built-in evaluation
Which statement best reflects 'Going Further's description of observability as an architecture?
Observability means adding a monitoring dashboard after the agent is deployedObservability requires a third-party tracing service wired into the policy seamObservability is optional polish that can be added once the agent reaches productionObservability is one event per step from day one, not a dashboard bolted on later