Best practices and features checklist for Teams agents

This article provides a list of best practices and features that developers should consider when implementing Teams agents.

These practices and features represent standard user expectations for agents in Teams and are broadly applicable. Agents that omit any of them should have a strong reason for doing so. Review each item to ensure your agent provides a great experience in Teams.

Enable access to all messages in group chats and channels

Agents that participate in Teams group chats and channels must opt in via configuration to receive messages in which they aren't @mentioned. For many agents, the context provided by this capability is critical to intelligent and proactive participation. To opt an agent in to receive all messages in group chats and channels, see Configure an agent to receive all messages in group chats and channels.

Message streaming and thinking indicators

Like many LLM chat experiences, Teams agents can stream messages in one-on-one conversations to maximize responsiveness. Before streaming begins, Teams can also display a customizable, updatable thinking indicator to indicate liveness and provide insight into how the message is being generated.

Animated screen capture depicting an agent's thinking indicator status updates, followed by message content being streamed as it is generated

Targeted messaging

Targeted messaging enables agents to privately interact with users in group conversations.

An agent's ability to receive targeted messages must be enabled in its app manifest configuration, but should be considered a standard feature of Teams agents. Users will expect agents in group conversations to use targeted messaging to enable context-sensitive actions while maintaining privacy and keeping conversations focused. Agents should always consider the visibility of messages they receive and use it to inform response generation, response visibility, and context tracking behavior.

Emoji reactions

Reactions are lightweight emoji markers that all participants in a conversation, including agents, can attach to chat messages. Agents can use reactions to communicate acknowledgement or status without creating additional noise in a conversation.

Desktop
Mobile

Prompt starters

Prompt starters are prewritten prompts that help users understand an agent's capabilities and quickly begin interacting with it. Users can conveniently access and discover prompt starters in all chat scenarios. In one-on-one chats, the app displays prompt starters prominently before the conversation starts for agents that don't send a welcome message.

Screenshot of an example of prompt starters.

Welcome messages

Agents can proactively send welcome messages when users add them to a conversation or install them as a personal app with one-on-one chat capabilities. Agents should use welcome messages to explain the value they provide, present configuration instructions, offer example prompts, and describe any commands they expose.

Agents distributed through the Teams Store and enabled for personal scope (one-on-one conversations) must implement welcome messages, prompt starters, or both.

Agent slash commands

Agents that provide command-style functionality can make it discoverable and easily usable with agent slash commands. Users can discover and invoke an agent's commands through Teams' slash command and @mention autocomplete menus in chat.

Suggested actions

Suggested actions are buttons on agent messages that enable users to quickly respond or take action. Unlike prompt starters and command names, agents can dynamically generate and attach suggested actions with each message they send, guiding users through workflows with context-aware options or intelligently suggesting next steps.

Suggested actions are key to enabling efficient and satisfying human-in-the-loop workflows, where an agent requests user approval or confirmation before taking action, but they're useful in any scenario that benefits from quick user interaction. They can be configured to send a command-style chat response to the agent, silently invoke an agent behavior without a chat message, or insert a pre-constructed message into the user's compose box.

AI content labels

Always use Teams' AI content labels as appropriate.

Attach the AI Generated label to all AI-generated messages. Use of this label is a policy requirement for agents distributed through the Teams Store.
Citations surface and link data sources used to generate a message.
Sensitivity labels clearly indicate that messages contain confidential information.