Review what's new and changed, and the fixes and limitations in AI Agent Studio for the v.38 release.

What's new

AI Agent

This release introduces AI Agent, a new feature designed to help you build powerful, autonomous, and goal-oriented solutions. AI Agent leverages large language models (LLMs) to perform complex, multi-step tasks by intelligently selecting and executing the right tools. The feature enables you to create and optimize prompts, integrate a wide range of tools such as API Tasks and automations, and configure human-in-the-loop interventions for enhanced control.

Administrators can create and manage all AI Agents directly from the Control Room, ensuring consistent deployment and providing version control.

AI Agent
Simplified AI Governance permission management

To improve the user experience and streamline role management, we have regrouped the permissions for accessing AI Governance data. Previously, multiple separate permissions were required to view different types of AI Governance logs. With this update, the permissions are consolidated into two clear and concise options:

  • View AI Governance logs: This permission now grants access to the summary page and high-level overview of all AI logs.
  • View AI Governance log detail: This permission allows users to view the detailed pages for individual AI logs.

Assign roles and permissions to enable AI Governance
Automation Anywhere provided models

We have expanded the available AI models you can use in our product. You can now select from two standard models: Claude Sonnet 3.5 and OpenAI GPT-4o.

When you create a model connection, select Standard as the type. You can then choose a model and a region. Available regions are United States and Europe.

These models are hosted on the Automation Anywhere cloud infrastructure. The system handles all model licensing and hosting, providing a credit-based usage model. This update simplifies how you access and use AI models in your automations.

Create and manage Model connections

Fixes

Fix for the garbling of non-English characters that previously occurred in the AI Skills page (when using JSON responses) and within Automation Anywhere Enterprise Knowledge chat interactions. Previously, due to incorrect UTF-8 encoding, this issue led to unreadable response characters in the output.

Service Cloud Case ID: 02203680,02203293, 02219180
Fix for an issue in the AI Skill editor where the Toxicity level and View Data Masking execution were not appearing for prompts and responses with very low toxicity. This update ensures that all content, including clean prompts and responses, will now correctly show its Toxicity level and allow you to View Data Masking execution details when applicable.

Limitations

AI Skills displays an error after cloning

When you open a cloned AI Skill from AI Skills landing page, it tries to open in edit mode and displays an Access denied error.
AI Guardrails functionality is English-only despite a localized UI

While the AI Guardrails user interface (UI) is available in multiple languages, the core functionality for toxicity filtering and data masking (PII, PCI, PHI) is currently limited to English-language content only. As a result, AI Guardrails will not detect or mask content written in other languages.