Let Slack write the first draft, not the final word

It often starts in a channel. A customer question. A school update. A decision buried inside a thread. Someone writes "let's handle it later", and later never really arrives.

That is why Slack's new AI step in Workflow Builder is worth paying attention to. Not because another AI feature can write text. Most tools can do that now. The useful part is that Slack is putting AI inside a repeatable work routine: start with a trigger, pull in the right context, create a draft, and let a person review it before anyone acts on it.

Workflow Builder is Slack's no-code tool for building automated workflows. An AI step is one part of that workflow where Slack can summarize, translate, classify, extract decisions and action items, transform information into a standard format, or draft a first response from a prompt and selected knowledge sources such as channels, canvases, lists, and files.

Source: Slack Blog: Smarter Workflows, No Code Required

The small difference between a prompt and a workflow

A prompt is easy to test. That is also the problem. It usually lives in one person's chat history, gets rewritten every time, and depends on someone remembering to paste in the right context.

A workflow is less exciting, but more useful. Same starting point. Same source material. Same format. Same review step. If a restaurant, shop, consultant, association, or school already does a lot of everyday work in Slack, the AI step can turn one messy repeatable task into something the team can improve over time.

Slack puts the point plainly: one-off prompts do not scale; workflows do. In practice, that does not mean everything should be automated. It means one recurring task can get a fixed path.

Source: Slack Blog: Smarter Workflows, No Code Required

What Slack can actually do inside the flow

The new step is called Generate AI response. In Slack's own example, it can summarize project channels and highlight progress, blockers, and next steps. Slack also mentions translation, classification of unstructured text, extraction of decisions and action items, data transformation into a standard format, and response drafting.

That sounds abstract until you translate it into daily work.

For a small business, it could be a Friday routine that reads the internal weekly channel and suggests three things: what got done, what got stuck, and what needs an owner on Monday.

For a school, it could create a first draft of a weekly parent update from a staff thread, using only material that has already been approved for that purpose.

For a customer support channel, it could take an incoming question, classify the issue, and write a reply draft with a clear line at the top: "Check this before sending."

This is not magic. It is a small machine for first drafts. And that is often where AI fits best.

Source: Slack Blog: Smarter Workflows, No Code Required

Start with a workflow that cannot send anything by itself

If I were introducing this in a small Nordic team, I would not start with a customer reply that goes straight to the customer. I would start with an internal summary workflow. It creates value quickly, and it makes the weak spots visible without affecting anyone outside the team.

Try this:

1. Choose one channel where work already happens, such as #customer-questions, #weekly-planning, #staff-updates, or #project-status.
2. Choose one recurring output: a weekly summary, a customer reply draft, a translated internal note, or a list of decisions.
3. Create a Slack Workflow Builder workflow that starts on a schedule, from a button, or when someone uses a specific emoji.
4. Add the AI step and select only the knowledge sources needed for that output. Start narrower than feels necessary.
5. Let the AI step write the draft to a private channel or to one responsible person.
6. Add a clear human review point: who approves, what must be checked, and when should the output be stopped?
7. Save one example of good output and one example of bad output. That becomes your first simple quality guide.

Slack Help describes Workflow Builder as a tool for recurring work with triggers, steps, forms, schedules, webhooks, connector steps, and activity logs. The combination of AI and ordinary process is what makes this useful.

Source: Slack Help: Guide to Slack Workflow Builder

Copy this prompt: turn the week's channel into a reviewed work list

Use this as a first version inside the AI step. Adapt the channel names, language, and format.

You help our small team create a weekly summary from Slack.

Use only the sources connected to this workflow. Do not guess. If the source material is missing, write "not found in the source material".

Create a first draft in English with five parts:

1. Short summary, maximum 6 sentences.
2. Decisions that seem to have been made, with the source or thread when possible.
3. Open questions that need a human decision.
4. Suggested next steps with owner if the owner is clear.
5. Items that must be checked before the summary is shared.

Write plainly and concretely. Do not use sales language. Mark uncertainty clearly.

The prompt does not need to be perfect. The important part is that the same prompt runs on the same kind of source material every week. Then the team can improve the routine in small steps instead of inventing the method again every Friday.

Integrate safely without killing the usefulness

Slack's documentation says owners and admins can manage access to Workflow Builder, workflow steps, triggers, and AI features. AI workflow steps are listed for Business+ and Enterprise+, and admins can control who can use certain AI features. Slack also says native AI features follow the same access boundaries as the user: AI should not use private channels or direct messages that the person cannot already access.

Source: Slack Help: Manage Workflow Builder access and features

Source: Slack Help: Manage access to AI features in Slack

Source: Slack Help: Security for AI features in Slack

For small teams, the practical question is simpler: who may build workflows, which channels may be read, and where does the draft go? Write that down before connecting more systems.

If the workflow later starts from an external system through a webhook, treat the webhook URL as a secret. Slack warns that anyone with the link can start the workflow. Store those links in a secret manager or protected environment variable, send the smallest useful payload, redact personal data that is not needed, and review runs in the activity log.

Source: Slack Help: Create a workflow that starts outside of Slack

Source: Slack Help: View workflow activity logs

This is the better kind of safety: not "do not connect anything", but "connect the right thing with the right permission and the right stop point".

Three small Slack workflows worth testing

Customer questions to reply drafts

When someone marks a customer question with a specific emoji, the workflow creates a reply draft, lists which facts it used, and sends the draft to the responsible person. No customer sees the reply until a human has checked it.

Staff thread to weekly note

Every Thursday, the workflow summarizes approved updates from an internal channel. It does not make the school's or company's communication automatic, but it removes the blank page.

Project channel to Monday list

On Monday morning, the workflow extracts decisions, blockers, and next steps from the previous week. The result goes to the team lead, who chooses what actually becomes the work list.

All three examples are small. That is the point. A good AI workflow often starts where people already lose time, but where the cost of a mistake is still manageable.

Where Hammer fits

This sits right in Hammer Automation's Tool Forge work: building small, controlled AI workflows that fit into daily operations instead of demanding a big system change.

Mindset Forge helps first, when the team needs to choose the right first workflow and decide what AI is allowed to do. Skill Forge helps people learn how to review drafts, adjust prompts, and notice when the source material is too thin. Tool Forge comes in when the flow needs to connect Slack, Google Workspace, CRM, scheduling, documents, or other tools without making permissions and responsibility fuzzy.

My recommendation is simple: let Slack write the first draft, not the final word. If the workflow saves ten minutes a week and makes decisions easier to see, it has already started to pay for itself. If it also teaches the team to use AI as a process, not as loose chat answers, that is a habit worth keeping.