Databricks Knowledge Assistant and Supervisor Agent are legacy: what replaces them

On 12 October 2026 Databricks marked Knowledge Assistant and Supervisor Agent as legacy, along with AI Playground, the old agent memory options and agents served from Model Serving. Each now has a successor. The no-code path is Genie Agents. The code path is the Agent Bricks CLI, which scaffolds an agent in LangGraph or the OpenAI Agents SDK, runs it on your laptop and deploys it to Databricks' agent runtime.

Legacy does not mean switched off. Databricks' page lists no end date and keeps the documentation "so that you can maintain and migrate existing agents". It does mean new agents should not start on the old products, and existing ones should get a migration plan before they get new features.

Find your replacement

Pick what your current agent does and see where it goes and the first command:

# Rebuild it as a Genie Agent over the same documents and tables

The full map

2 patterns run through it. Anything that was a pre-built agent configured in screens becomes a Genie Agent. Anything that was code becomes a CLI project with a standard server, local runs and a deploy command. Databricks names that server DurableAgentServer and the place it runs Agent Runtime.

The first 5 commands

The CLI needs Python 3.10 or newer and the Databricks CLI.

pip install databricks-agentbricks

databricks auth login --host https://<your-workspace-url> --profile dev
agentbricks login --profile dev

agentbricks init --framework langgraph my-agent
cd my-agent

agentbricks dev                 # runs the agent at http://localhost:8000
agentbricks deploy my-agent     # deploys it to the agent runtime

Use --framework openai for the OpenAI Agents SDK instead of LangGraph. 2 more commands replace the legacy memory options by attaching Databricks-managed stores:

agentbricks sessions bind my-agent-sessions
agentbricks memory bind my-agent-memory

And a Genie Agent your analysts already built can become a tool of the coded agent:

agentbricks tools add genie-agent

That last command is how a Supervisor Agent's job moves to code: the supervisor's routing becomes your agent's loop, and each Genie Agent it called becomes a tool.

Migrating an existing agent

1. List what you run. Note each Knowledge Assistant and Supervisor Agent, its sources, who uses it and how they reach it.
2. Rebuild the simple ones as Genie Agents. A question-answering agent over documents and tables usually fits without code.
3. Move the coded ones to a CLI project. Copy the agent's logic into the scaffold, run agentbricks dev, and test the same questions against both versions.
4. Move memory last. Bind agent memory and sessions, then check that a returning user's context carries over.
5. Switch users, then retire the old one. Keep it until the new agent has handled a normal week.

The CLI is in Beta, so pin its version in your project and read the release notes before upgrading. Databricks has renamed most of its products in the past year, so the old and new names in your notes may not match the current docs; every Databricks rename lists them. For the design of the agent itself, our AI agents subject follows one request through an AgentAgentA model that uses tools in a loop, deciding each next step from what the last one returned, until the task is done. step by step, and LangChain vs LangGraph explains the framework the CLI scaffolds by default.