AI Agents: Key Principles and Guidelines - Part 3

Hi everyone, Shivam Goyal here! In the previous posts (Part 1, Part 2—links provided at the end), we covered the basics of AI agents and explored available frameworks. This week, we'll focus on a crucial aspect: designing user-centric agentic systems, drawing specifically from the repository's section on agentic design patterns.

The Challenge of Agentic Design

Building effective AI agents isn't just about technical prowess but understanding and catering to the user experience. We need agents that empower users without being intrusive or confusing. This post introduces a set of UX design principles to guide this process.

Agentic Design Principles: Space, Time, and Core

These principles are categorized into three key areas:

Agent (Space): Defining the Agent's Environment

• Connecting, Not Collapsing: Foster collaboration and connection, not isolation. Agents should bridge the gaps between people, information, and events.
• Easily Accessible Yet Occasionally Invisible: Agents should be readily available but operate discreetly, stepping forward only when relevant. This requires multimodal capabilities and seamless transitions between the foreground and background.

Agent (Time): Interacting Across Past, Present, and Future

• Past: Leverage historical data and context for richer, more relevant responses.
• Now: Go beyond simple notifications. Agents should offer proactive suggestions and contextual information.
• Future: Agents must adapt to evolving user needs, new devices, and platforms to remain helpful.

Agent (Core): Foundational Elements of Trust and Transparency

• Embrace Uncertainty, But Establish Trust: Transparency and control are key. While AI has inherent uncertainties, users need to understand the agent's capabilities and limitations and have control over its operation.

Implementation Guidelines

Putting principles into practice requires clear guidelines:

1. Transparency: Be upfront about the AI's role. Explain its function, data usage, and how users can offer feedback and control the system.
2. Control: Empower users with customization and data control. Provide clear opt-out and data deletion mechanisms.
3. Consistency: Ensure a consistent experience across devices and modalities. Use familiar UI/UX elements and prioritize clarity and simplicity.

Building a Better Travel Agent: A Practical Example

Let's apply these principles to a travel agent:

• Transparency: Clearly indicate the agent's AI nature. Provide onboarding, document limitations, show past queries, and offer feedback channels.
• Control: You can customize the agent's parameters (verbosity, style, topic restrictions) and enable data review and deletion.
• Consistency: Employ standard icons and UI elements, maintaining consistent interaction patterns.

Continue Your Agentic AI Journey

These principles offer a starting point for user-centric agent design. For a deeper dive, explore these resources:

• AI Agents for Beginners Repository
• The HAX Toolkit Project - Microsoft Research
• Microsoft Responsible AI Toolbox - Microsoft Responsible AI
• Step-by-Step Tutorial: Building an AI Agent Using Azure AI Foundry | Microsoft Community Hub

Catch up on the series:

• Part 1: Introduction to AI Agents
• Part 2: Exploring Agentic Frameworks

 If you have any further questions or would like to connect for more discussion, feel free to reach out to me on LinkedIn | GitHub