AI Agent Studio v.34 release

Note: This offering requires the Enterprise Platform license. For information about supported version for this feature, see Enterprise Platform.

Review what's new and changed, and the fixes and limitations in AI Agent Studio for the v.34 release.

Note: Generative AI models can produce errors and/or misrepresent the information they generate. It is advisable to verify the accuracy, reliability, and completeness of the content generated by the AI model.

What's new

Support for RAG capability in AI Agent Studio to create grounded Model connections using Amazon Bedrock and Google Vertex AI foundational models

Create Grounded by knowledge base and Grounded by data store Model connections using the native RAG (retrieval-augmented generation) capability from Amazon Bedrock and Google Vertex AI to generate accurate and contextually relevant information referenced from Amazon Knowledge Base and Google Data Source.

See Create Grounded Model connections with Amazon Bedrock RAG capability and Create Grounded Model connections with Google Vertex AI RAG capability.

Important: To use Google Vertex AI Grounded models for On-Premises, document chunking must be disabled in the data store.

In Cloud, the option to use Google Vertex AI Grounded by data store is currently not available. In an upcoming patch release, data stores with enabled document chunking will be supported to ensure optimal results in automation executions.
Support for Fine tuned Model connections in AI Agent Studio

Create Fine tuned Model connections by customizing the supported foundational models via testing and fine-tuning to make them available to Professional Developers for connecting to the AI Skills.

See Create Fine tuned Model connections.
AI Skills integration with API Tasks

Enhance your API Tasks by leveraging the generative AI capability of AI Skills to reduce cost and improve performance of automations while integrating with AI models in a scalable, safe, and governed environment.

See Use AI Skills in API Tasks.
New code analysis policy support for AI governance

Monitor and enforce responsible use of AI during design and development of automations with a new code analysis policy to ensure compliance and governance.

See AI governance code analysis rule.
New AI governance dashboard

Derive valuable insights into the AI model usage and their performance within automations, and track and monitor the most used models along with their token consumption to ensure compliance through your dashboard. Drill down to interaction details during the design and execution of automations and download the reports.

Track data via the following AI governance dashboard widgets:
  • Top automation executions using AI models
  • Most used models from hyperscaler vendors
  • Token consumption of the most used models

See AI governance dashboard.
Enable/disable Data logging toggle from the AI Skills editor

Enabling Administration > Settings > AI Data Management > Data logging settings enables the Data logging toggle in the AI Skill editor. This feature generates detailed logs of AI Skills displaying the prompt-text, response, and the model parameter settings in the AI governance > AI prompt log tab. The Pro Developer can enable or disable the toggle to mask sensitive information for an AI Skill, as required by their organization regulations and compliance policies.

See Enable data logging for AI governance.

What's changed

Navigation for Model connections and AI governance in the Control Room

Now you should navigate to AI > Model connections and AI > AI governance in the Control Room.

AI Tools UI updates

  • Prompt Template has been renamed to AI Skills.
  • Generative AI Prompt Template package has been renamed to AI Skills package.

AI governance updates

  • The Automation Name field in the AI prompt log and Event log tabs list-view pages now displays the actual automation or bot names.
  • The AI prompt log and Event log tabs detail-view pages now use a meaningful title for the detail-view, instead of the Session ID used earlier.
  • The vendor name in the AI prompt log and Event log tabs detail-view pages has bee renamed to Publisher, which is now used consistently across all detail pages to display the foundational model vendor name.
  • In the AI prompt log and Event log tabs detail-view pages, the Line numbers and Command action fields have been removed.
  • The additional parameters in the AI prompt log and Event log tabs detail-view pages now display prompt token, completion tokens, and total number of token values for the Response attributes field.
  • The AI prompt log and Event log tabs list-view pages have a new column for the Folder path that shows the folder path of where the automation exists.
  • In the Event log tab detail-view page, the Connection method field displays the relevant connection mode such as: User account, Service account, Control Room OAuth, and others.
  • The AI governance and Model connection settings have a consolidated view with all permissions listed under Administration > Roles > AI in the Control Room.

    Required permissions for AI governance

Citations field displayed in a Task Bot when using the AI Skills package with Grounded Model connections

With the introduction of the Grounded by knowledge base and Grounded by data store Model connections using Amazon Bedrock and Google Vertex AI RAG capability, we now additionally display fields with the returned citations in a table format when using the AI Skills package.
The response goes to a SampleString variable and a table variable gets automatically selected to display the citations. The table displays two keys within each citation. One is the content segment displaying the information chunk and the other is the URI, which is a URL to the chunk of information in the knowledge base or data store the model call is referencing.
Note: When you run the automation, the execution displays messages showing the referenced citations in message boxes.

Limitations

Users must have the View Settings permission to edit an AI Skill. Otherwise, an error message is displayed and users will not be able to edit the AI Skill.
In this release, AI governance logs are not supported for API Tasks. This is specific to the API Tasks that are run in real-time mode (for attended automations).
In the AI Skill editor, when you try to update the title of an existing AI Skill without an associated Model connection, Prompt-text, or defined Prompt Input, the system displays an error message stating a validation failure and fails to save the update. The AI Skill should take the update and save the change.
Bot Agent 22.60.10 and later dependency for generating audit logs

For the AI governance feature, when automation administrators enable the Administration > Settings > AI Data Management > Data logging settings, they see a notification requesting them to install a compatible Bot Agent version to ensure the audit logs for all model interactions are logged successfully in the Control Room.

The GRC Leads, Automation Leads, and the Professional Developers accessing the data logs in AI > AI governance > AI prompt logs and AI > AI governance > Event logs in the Control Room also receive the same notification.
Note: We recommend using Bot Agent version 22.60.10 and later.

See Assign roles and permissions to enable AI governance.

The Search field in the AI governance > AI prompt log and AI governance > Event log tabs list-view pages is not able to process entries with special characters such as the following:

+ = && || > < ! , { } [ ] ^ ~ :