Hi Asfriz,

Genie Agent mode is currently unavailable via direct API. This capability is something we are also eagerly waiting for.

For now, you can try building an Agent Supervisor which has your Genie space served as a tool & utilize the orchestrator model to mimic the Genie Agent mode. I know this might not be exactly what you are looking for but we tried this & were able to get significantly better response (both quality & structure) for one of our implementations. 
Since all of this will be managed by mlflow traces & LLM judges, management of intermediate steps will also be relatively easy. Refer to this for more details: Agent Evaluation (MLflow 2) | Databricks on AWS

Also, this new release might be coming sooner than expected with the DAIS going on.

Databricks recently announced 3 new Genie capabilities: https://www.databricks.com/blog/introducing-genie-one-genie-ontology-and-genie-agents 

Would love to connect further on this use case.

Regards,

Ankit