Agents in your workspace

Building an agent is half the story. The other half is putting it to work in the places your team already lives. An agent can join your work in four ways, and in every one of them you stay in the loop for the moments that matter.

Assign an agent to a task

You can make an agent the assignee of a task, exactly like you’d assign a person. From there it gets to work:

• It posts progress comments so you can watch what it’s doing.
• When it needs a decision or more information, it asks in the comments and waits.
• You review and approve its work — or reply with feedback and send it back around.

This is great for self-contained jobs: “write the release notes for this version,” “draft the first pass of this spec,” “triage this bug.”

Want the close-up?

The comment thread is where the agent actually does its work — planning, calling tools, asking for your call at the right moments. For the full mechanics, see How an agent works a task.

Add an agent to a chat

Drop an agent into a channel or a direct message and it participates alongside your team. Useful when you want an always-present helper — a support agent in your #support channel, say — that can chime in, answer, and act in context.

Summon an agent with a mention

You don’t have to add an agent permanently to get its help. Just mention it — type @ and its name — and it joins that conversation to respond. Mentioning an agent is the quickest way to pull in a specialist for a single question. (More on mentions.)

Use an agent inside a flow

Agents can be a step in an automation. A flow can hand work to an agent — for example, “generate a friendly message” — as part of a larger automated process. This is how you blend deterministic automation with judgment.

What you’ll see while an agent works

Agent messages show their status, so you always know where things stand:

Status	Meaning
Writing	The agent is thinking and composing a response.
Completed	It’s finished and ready for you.
Waiting for User	It needs your input or approval to continue.
Failed / Canceled	Something went wrong, or the run was stopped.

You may also see agentic steps — the agent showing its work, like the tools it called and what it found. Expand them when you want the details; collapse them when you just want the result.

Human-in-the-loop is the default

Agents are built to ask before doing anything sensitive. When you see Waiting for User, the agent has paused for your call. Approve it, decline it, or steer it — nothing important happens without you.

Related

• How an agent works a task — the comment-driven execution loop, up close.
• Agents
• Tools & knowledge
• Tasks
• Flows