Hi everyone,
I’m currently developing several Databricks Agents using a custom Claude skill to integrate Databricks Asset Bundles (DABs), Mosaic AI frameworks, and LangGraph.
My current workflow is the following:
- I develop the agents locally.
- For deployment, I trigger a CI/CD pipeline (GitHub or Azure DevOps).
- The pipeline runs databricks bundle deploy and then executes a notebook responsible for registering the agent and its artifacts.
In my mind this seemed like the correct approach, since agent deployment usually involves steps like MLflow model registration and artifact logging, which require some form of execution (notebook, job, or script).
However, I started wondering if there might be a more direct way to deploy agents without relying on a pipeline step that executes a notebook.
My assumption is that this execution step is always required because the agent needs to be registered along with its artifacts and configuration, but I’d be curious to hear how others are handling this.
How are you deploying your agents in practice?
Are you also using CI/CD pipelines that run notebooks or scripts for registration, or have you adopted a different pattern for deploying agents?
Thanks in advance for any insights
