Summary

Slack should introduce Agents as a first-class identity primitive separate from Apps.

Today, Slack treats apps, bots, and agents as effectively the same thing. A company installs an app, that app has a bot identity, and users interact with that bot. This model worked for traditional integrations, notification bots, and workflow tools. It does not work for AI agents.

AI agents are not just apps. They are identities people collaborate with. They may be personal assistants, team-specific teammates, executive-room agents, department copilots, or public company-wide helpers. A single AI provider may need to create many different agents inside the same Slack workspace, each with its own name, profile, owner, audience, visibility, and behavior.

Slack should decouple these concepts:

• Apps are installed capability providers.
• Agents are Slack identities created by apps or admins.
• Users interact with agents, not apps.
• Admins govern agents through an Agent Registry.

This would let a workspace approve an app once, then create many scoped agents without polluting the global app namespace, exposing private agents to the whole company, or forcing vendors to encode behavior invisibly at the channel level.

Problem

1. Apps and agents are currently collapsed into one object

In Slack today, installing an AI assistant generally means installing an app with a bot identity. That bot identity is the thing users search for, DM, mention, invite to channels, and see in the workspace.

This creates a bad abstraction for AI agents.

A traditional app is a system integration. An AI agent is closer to a teammate. It has a job, audience, behavior, identity, memory, and operating context. A single AI app may need to support many agents:

• Max’s personal assistant
• Exec team assistant
• Sales coach