## Module 4: Why Branching Matters for Agents

### 4.1 Wrap-up

By now, you have completed a **branch-style exploration** workflow for databases: for the same requirement, the agent implements different solutions on multiple branches in parallel; you select the most suitable branch as the baseline, then continue evolving from that point.

This workflow is not "just for fun". In real engineering for agentic AI systems, it is a recurring capability requirement: **parallel exploration, rollback, comparison, and traceability**.

You can read the following paper as supporting reference:

- [Branch-Train-Merge: A Version Control Paradigm for Collaborative LLM Development](https://arxiv.org/abs/2509.00997)

You do not need to read every detail. A useful guiding question is:

- When an agent must try multiple candidate routes (plans, tool-call sequences, structured artifacts) for a complex task, how can we organize these attempts as a **tree/branches** so that the final solution becomes more stable and controllable?

Back to our database setting: treating TiDB Cloud Branching as a **branching mechanism for external world state** reduces exploration cost, makes comparison more direct, and makes rollback more reliable—this is why branching becomes important in agentic AI systems.

### 4.2 Cleanup

- Return to [TiDB Cloud console](https://tidbcloud.com/project/clusters), in the **Branches** list and locate the branches created by the agent.

- Click `...` under **Action** on the right.

- Choose **Delete** in the menu/dialog, then confirm by selecting **I understand, delete**.