All patterns
Agents
Production Agent
An agent system wrapped in the production machinery required to ship safely: gateway, policy, observability, evaluation, caching and security.
clientservicemodeldatabaseagentexternal
Explanation
Every agent action passes through a policy layer that enforces authorisation, rate limits and safety. The gateway handles auth, caching and routing. Every step is traced for observability and scored for evaluation.
Components
API gatewayAuth + RBACPolicy engine (OPA/Cedar)Agent orchestratorTool registryObservability (OpenTelemetry/LangSmith)Evaluation pipelineCaching layer
When to use
- Any agent shipped to external users
- Agents that take actions on user data
- Regulated industries (finance, healthcare)
When NOT to use
- Early prototyping
- Internal-only trusted tools
Failure modes
- Policy too strict — agent unable to act
- Policy too lax — unsafe actions execute
- Observability sampling misses the failing trace
- Cache poisoning returns stale/wrong results
Production checklist
- AuthN + AuthZ on every request
- Policy enforcement on every tool call
- Full trace of agent trajectory
- Online evaluation scores in dashboards
- Cost and token budgets per run
- Cache invalidation strategy
- Audit log of every action
- Incident response runbook