Version control for AI agents and agentic workflows
Summarize
Summary of Version control for AI agents and agentic workflows
Version control in ServiceNow’s AI Agent Studio allows you to manage and track changes to the instructions (List of steps) provided to large language models (LLMs) for AI agents and agentic workflows. This capability is essential for experimenting, refining, and ensuring the effectiveness of your AI-driven processes by maintaining multiple versions of the instructions.
Show less
Key Features
- Creating Versions: You can create new versions of the List of steps via the guided setup by naming each version and optionally setting it as active immediately. This makes it easier to identify and manage changes over time.
- Refining Versions: Using the Now Assist feature, you can generate refined versions of your List of steps based on the current active version, facilitating improvements without starting from scratch.
- Switching Active Versions: You can toggle between different versions at any time to activate the one you want to test or use, allowing flexible evaluation and deployment of instructions.
- Testing Versions: The AI Agent Studio Testing page lets you select from both active and inactive versions for testing and set up evaluation runs to measure performance, supporting informed decision-making.
- Version Logs: Execution logs filtered by version are accessible from the AI Agent Studio Activity page or the version logs modal, helping you track usage and performance per version. Detailed version naming improves traceability.
Key Outcomes
By leveraging version control for AI agents and agentic workflows, ServiceNow customers can confidently experiment with and optimize AI instructions, monitor changes, and evaluate performance across different versions. This ensures more reliable and effective AI automation aligned with evolving business needs.
Version control enables you to track changes made to instructions sent to the LLM for AI agents and agentic workflows.
Version control overview
When creating an AI agent or agentic workflow, the List of steps field is important because it provides the context necessary for the large language model (LLM) to accomplish tasks. You can use version control to create multiple versions of the List of steps field in the guided setup. Tracking your changes with version control helps enable you to experiment and test new instructions. You can create versions, edit and refine versions, and revert to previous versions as needed.
Creating versions
You can create a version by selecting in the guided setup of an AI agent or agentic workflow. Doing so brings up the Create a new version modal where you can name the version and make changes. Naming your version enables you to identify what changes were made to make it easier to track.
You can choose to make the new version active immediately by selecting the Set as active toggle.
For more information about writing effective descriptions and instructions, see General guidelines for creating AI agents and agentic workflows.
When creating a new version, you also have the option to use Now Assist to refine your current List of steps. You must already have an active version for this option to be available. Selecting brings up the modal where Now Assist generates and displays a new List of steps for you to review. You can change any of the generated text, or you can use it as it was generated. Select Create to use the new, refined version.
Changing between versions
When you have multiple versions of the same List of steps field, you can switch which one is active at any time. Choosing to make a new version active enables you to test it and evaluate its effectiveness.
If you’re updating an agentic workflow, you can execute an agentic evaluation run to measure performance.
You can decide which version is active by selecting View versions in the guided setup. Toggle the Set as active field and save your changes to make the new version active.
Testing different versions
You can test different versions of an agentic workflow or AI agent from the AI Agent Studio Testing page. The Version selector dropdown displays both active and inactive versions with the currently active version marked as (Active). You can also start evaluation runs for specific versions by selecting Set up evaluation run.
Version logs
The AI Agent Studio Activity page enables you to track AI agent and agentic workflow executions by version. When in the guided setup, you can select View version logs in the View all versions modal. Choosing to view the version logs opens the Activity tab in a new window with the list filtered for the version you selected.
When viewing the Activity tab at any time, you can choose to filter by version. The more specific your version names, the easier it is to identify those changes in the list.