Microsoft AI Agent Introductory Course: AI Agentic Design Principles

summary

There are many ways to build an AI Agentic system. Given that ambiguity is a feature, not a flaw, of generative AI design, it is sometimes difficult for engineers to determine where to start. We have created a set of human-centered user experience design principles that enable developers to build customer-centered Agentic systems that address their business needs. These design principles are not prescriptive architecture, but rather provide a starting point for teams defining and building Agentic experiences.

Usually, the Agent should:

• Expand and extend human capabilities (brainstorming, problem solving, automation, etc.)
• Fill in knowledge gaps (give me a quick overview of knowledge areas, translations, etc.)
• Promote and support ways in which we as individuals prefer to work with others
• Make us better versions of ourselves (e.g., life coaches/taskmasters to help us learn emotional regulation and positive thinking skills, build resilience, etc.)

This course will cover

• What are Agentic Design Principles?
• Some guidelines to follow when implementing these design principles
• Some examples of using design principles

Learning Objectives

Upon completion of this course, you will be able to:

1. Explain what the Agentic Design Principles are
2. Explain the guidelines for using Agentic design principles
3. Learn how to build an Agent using Agentic design principles

Agentic Design Principles

Agent (space)

This is the environment in which the Agent operates. These principles guide how we design Agents to participate in the physical and digital world.

• Connecting, not folding - Help people connect to other people, events, and actionable knowledge to collaborate and connect.
  • Agent helps connect events, knowledge and people.
  • Agents bring people closer together. They are not designed to replace or devalue people.
• Easily accessible but occasionally invisible - Agent runs mostly in the background and nudges us only when relevant and appropriate.
  • Agents are easy to discover and access for authorized users on any device or platform.
  • Agent supports multimodal input and output (sound, voice, text, etc.).
  • Agents can seamlessly transition between front- and back-office; between proactive and reactive, based on their perception of user needs.
  • Agents may run invisibly, but their background processing paths and collaboration with other Agents are transparent and controllable to the user.

Agent (time)

This is how Agents operate over time. These principles guide how we design Agents to interact in the past, present, and future.

• (in the) past: Reflect on history including state and context.
  • Agent provides more relevant results based on the analysis of richer historical data, not just events, people or states.
  • Agent creates connections from past events and actively reflects on memories to engage with current situations.
• present .: More nudges than notifications.
  • Agents embody an integrated approach to human interaction. When an event occurs, an Agent is more than just a static notification or other static form; an Agent can streamline the process or dynamically generate prompts to direct the user's attention at the right moment.
  • Agent provides information based on contextual, social and cultural changes and is customized to user intent.
  • Agent interactions can be incremental, evolving/becoming complex over time to empower users over time.
• pending: Adaptation and development.
  • Agent adapts to a variety of devices, platforms and models.
  • Agent adapts to user behavior, accessibility needs, and is freely customizable.
  • Agent is shaped and developed through continuous user interaction.

Agent (core)

These are the key elements at the heart of Agent design.

• Embrace uncertainty but build trustThe
  • Some degree of Agent uncertainty is expected. Uncertainty is a key element of Agent design.
  • Trust and transparency are the foundational layers of Agent design.
  • Humans can control when an Agent is turned on/off and the Agent status is always visible.

Guidelines for the implementation of these principles

When using the previous design principles, please follow these guidelines:

1. transparency: Inform the user about the AI's involvement, how the AI operates (including past operations), and how to provide feedback and modify the system.
2. containment: Enables users to customize, specify preferences and personalize, and control the system and its attributes (including the ability to forget).
3. consistency: Aim to provide a consistent multimodal experience across devices and endpoints. Use familiar UI/UX elements wherever possible (e.g., microphone icons for voice interactions) and minimize the cognitive load on the customer (e.g., strive for clean responses, visual aids, and "learn more" content).

How to use these principles and guidelines to design a trip Agent

Assuming you are designing a Travel Agent, here is how to consider using the design principles and guidelines:

1. transparency - Let users know that the Travel Agent is an AI-enabled agent. provide some basic instructions on how to get started (e.g., "Hello" messages, sample tips). Document this clearly on the product page. Show a list of tips that users have asked for in the past. Clear instructions on how to provide feedback (thumbs up and thumbs down, "Send Feedback" button, etc.). Clearly state whether the Agent has usage or theme restrictions.
2. containment - Ensure that users clearly understand how to modify an Agent after it has been created using, for example, system prompts. Enable users to choose the level of detail of the Agent, its writing style and any warnings about what the Agent should not talk about. Allow users to view and delete any associated files or data, prompts and past conversations.
3. consistency - Ensure that the icons for Share Alerts, adding files or photos, and tagging someone or something are standard and recognizable. Use the paperclip icon for uploading/sharing files to the Agent and the image icon for uploading graphics.