Logging: The current code base uses multiple print statements where log messages would be more suitable. We could either use the default Python logging module or open telemetry with langfuse + logfire for all logs. The latter allows us to keep all logs in one place (langfuse) but risks causing a lot of noise when viewing LLM metrics in Langfuse UI. We could improve this using filters, tags, etc.
For CLI usage, we should implement either a separate module using Typer or integrate with one of the existing UIs using the AG-UI protocol

Metrics: Currently metrics such as time taken for inference are being captured in the AgenetExecution dataclass in supervisor.py. This will not scale as we seek to capture complex metrics, including latency, token usage, costs, etc. Langfuse already does a good job of this and we should just delegate this to Langfuse and focus on the agentic logic in the application code.