{"id":8437,"date":"2026-03-30T05:00:00","date_gmt":"2026-03-30T12:00:00","guid":{"rendered":"https:\/\/slack.dev\/?p=8437"},"modified":"2026-04-14T12:36:16","modified_gmt":"2026-04-14T19:36:16","slug":"slack-thinking-steps-ai-agents","status":"publish","type":"post","link":"https:\/\/slack.dev\/slack-thinking-steps-ai-agents\/","title":{"rendered":"Make your AI agent think out loud in Slack"},"content":{"rendered":"\n

When AI feels like a black box<\/h2>\n\n\n\n

You\u2019ve seen it before. You send a message to an AI agent, and then\u2026 nothing. A spinner. A blank screen. And then \u2014 after what feels like forever \u2014 a fully formed response appears out of thin air. It works, technically. But it doesn\u2019t feel great.<\/p>\n\n\n\n

Users are left wondering: Is it still running? Did it get stuck? What just happened? <\/em>There\u2019s no visibility, no sense of progress, and no real collaboration.<\/p>\n\n\n\n

The best AI experiences today don\u2019t just give you answers. They show you how they get there. You see the reasoning unfold in real time. It feels less like querying a system, and more like working alongside something that\u2019s actively thinking with you.<\/p>\n\n\n\n

Now, you can bring that same experience directly into Slack.<\/p>\n\n\n\n

Introducing Thinking Steps<\/h2>\n\n\n\n

Thinking Steps is a new set of Block Kit elements and streaming APIs that lets your Slack app surface an AI agent\u2019s reasoning process live in the conversation.Tool calls, task execution, and decisions, all visible as they happen.<\/p>\n\n\n\n

Combined with Chat Streaming, Thinking Steps completes the picture: your agent can now show both what it\u2019s doing<\/strong> and what it\u2019s saying<\/strong>, in real time.<\/p>\n\n\n\n

Two ways to display: Plan vs. Timeline<\/h2>\n\n\n\n

Thinking Steps supports two display modes, depending on how your agent works.<\/p>\n\n\n\n

Plan mode<\/strong> presents a structured overview of tasks up front. This works well when your agent is laying out a multi-step approach before getting started. Like a to-do list that gets checked off as it goes.<\/p>\n\n\n\n

Timeline mode<\/strong> shows tasks in a linear sequence as they happen. This is a better fit for agents that execute step by step and want users to follow along in real time.<\/p>\n\n\n\n

Both modes are collapsed by default, so they don\u2019t overwhelm the conversation. Users can expand them when they want to peek under the hood.<\/p>\n\n\n\n

The building blocks<\/h2>\n\n\n\n

Thinking Steps introduces four new content types (\u201cchunks\u201d) that you can stream into a message:<\/p>\n\n\n\n

\ud83d\udcdd\u00a0Markdown Text Block
<\/strong>Stream free-form text with full markdown support. Great for reasoning, summaries, or transitions between steps.<\/p>\n\n\n\n

\u2705\u00a0Task Card Block
<\/strong>Represents a single action or tool call. Each card can include a title, status, and references. This is the core unit of Thinking Steps.<\/p>\n\n\n\n

\ud83d\uddc2\ufe0f\u00a0Plan Block
<\/strong>A container for grouping Task Cards. Useful when your agent is executing a set of related actions and you want to present them together.<\/p>\n\n\n\n

\ud83d\udd17\u00a0URL Sources Element
<\/strong>Attach references directly to a Task Card. When your agent pulls from external sources, you can surface those links inline, exactly where users expect them.<\/p>\n\n\n\n

How it works<\/h2>\n\n\n\n

If you\u2019re\u00a0using Bolt<\/a>, the easiest way to work with streaming (and Thinking Steps) is through the built-in helper:<\/p>\n\n\n\n

const streamer = client.chatStream({\n\tchannel: channel,\n\trecipient_team_id: teamId,\n\trecipient_user_id: userId,\n\tthread_ts: thread_ts,\n});<\/code><\/pre>\n\n\n\n
This helper abstracts away the underlying streaming lifecycle, so you can focus on what your agent is doing \u2014 not how messages are managed.<\/p>\n\n\n\n
Under the hood, Thinking Steps builds on the same streaming APIs used for Chat Streaming:<\/p>\n\n\n\n