UX design for agents

Source: Ruokan He, Jen Fox, Amanda Snellinger; Microsoft Design · 2025-04-11

As advances in AI-powered software allow us to interact with computers in new ways, good design is critical to success. At Microsoft, we have been designing and building out agentic systems, or “agents,” which are software systems that include an AI or ML model, orchestration code, and skills/tools (e.g. code writing, web search, etc.). These agents collaborate with you, automate your least favorite tasks, and scale your capabilities.

We define an “agent” as an AI assistant designed to execute tasks, working with or for humans. The components of an agent include instructions, knowledge, actions, skills, memory, etc. Agents may have chat experiences but generally go beyond typical chatbots.

Agentic AI systems (or “agent systems”) are made up of one or more agents. They can autonomously identify, plan, and take actions on behalf of an individual user, a group, or other agents from a general set of instructions with limited direct human supervision. An AI system may be considered “agentic” if it is autonomous, has more complex capabilities, operates in a more complex environment, and can scale to more domains.

Agents mark the beginning of the next chapter in human-computer interaction. They also require a huge degree of user trust—which must be earned. We’ve worked cross-functionally to create a set of principles that guide the development of agents and ensure these systems are built in ways that serve human goals and meet human needs.

Broaden and scale human capacities, e.g. brainstorming, problem solving, automation, etc.

Fill in knowledge gaps, e.g. get me up-to-speed on knowledge domains, translation, etc.

Facilitate and support collaboration in the ways we as individuals prefer to work with others.

Make us better versions of ourselves, e.g., life coach/task master, upskilling, helping us learn emotional regulation and mindfulness skills, building resilience, etc.

The Agent UX Design Principles have three high-level categories with 6 sub-categories:

Connecting, not collapsing: help connect people to other people, events, and actionable knowledge to enable collaboration and connection.

Easily accessible yet occasionally invisible: an agent largely operates in the background and only nudges us when it is relevant and appropriate.

Past: reflecting on history that includes both state and context.

Now: nudging more than notifying.

Future: adapting and evolving.

Embrace uncertainty but establish trust.

Transparency, control, and consistency are foundational elements of all agents.

Agents and agentic systems help connect events, knowledge, and people. Agents bring people closer together.

Agents are not designed to replace or belittle people.

Individual agents and multi-agent systems are easily discoverable, intuitive, and accessible for authorized users on any device or platform.

An agent uses data in memory to inform and engage with current events and tasks. This includes creating connections from past events, actions, tasks, and user queries.

Agents use memory and connected databases to provide more relevant results based on analysis of richer historical data beyond singular events, user queries, or states.

An agent embodies a comprehensive approach to interacting with people. When an event happens, an agent goes beyond static notification or other static formality, e.g. it may proactively start a chat or create artifacts. Agents can simplify flows or dynamically generate cues to direct the user’s attention at the right moment.

An agent runs on and adapts to various devices, platforms, and modalities. For example, based on user preference, an agent may default to a voice-first interaction on mobile but text-first interaction on a PC.

A certain level of Agent uncertainty is expected. Uncertainty is a key element of agent design.

The level of certainty and reasoning behind a recommendation are visible and/or easily accessible to the user to avoid overreliance.

Agents’ associated knowledge, tools/skills, and connections with people and other agents are transparent and customizable. This helps build trust with users.

Agents that operate as background processes have a user-facing mechanism to view and control actions and automations.

Humans can customize agent settings, including specifying preferences and personalization, and are in control of when the Agent is on/off.

Agent status, or what the agent is doing, is clearly visible at all times.

As with any product development process, it is critical to start with an end-user/customer problem and then brainstorm possible solutions that may or may not involve the use of AI. This helps to ensure we continue to center human needs as we create new user experiences with agentic systems.