Slack AI Is Nice. Here's What It Can't Do (and What We Built Instead)
Slack AI launched and our team was genuinely excited. Anya tried the channel summary feature on a Monday morning and said, "This is going to save me 20 minutes a day." She was right about the 20 minutes. She was wrong about it being enough.
Slack AI does two things well. It summarizes channels so you can catch up without reading 200 messages. And it answers questions about what happened in conversations, pulling from your message history to give you a synthesized answer. Both are useful. Neither does anything.
By "does anything" I mean takes action. Slack AI will tell you that three customers mentioned billing issues in #support last week. It will not open your CRM and check those customers' account status. It will not send them a follow-up message. It will not create a ticket in Zendesk. It will not flag the accounts for your CS team. It reads. It summarizes. It stops there.
That gap between "knowing" and "doing" is where we've spent the last eight months building.
What Slack AI Actually Gives You
I want to be fair to the built-in features because they are not bad. They just have a very specific scope.
Channel summaries compress hours of conversation into a few paragraphs. For someone returning from vacation or joining a meeting late, this is genuinely helpful. Monday mornings used to start with 30 minutes of scrolling through #general, #engineering, #support, and #deals. Now it starts with reading four summaries. Real time saved.
Search answers let you ask natural language questions about your Slack history. "What did we decide about the pricing change?" pulls from relevant threads and gives you a consolidated answer. Better than Slack's old search, which required you to know the exact words someone used and still returned 47 results sorted by date.
Thread summaries condense long threads into key points. When a thread hits 40 messages and you join late, the summary gets you up to speed without reading the whole argument about whether the button should be blue or green.
All three features are read-only. They observe your Slack workspace and tell you things about it. They cannot reach outside Slack. They cannot modify anything inside Slack. They are a very good pair of glasses. But we needed hands.
The Gap
Here is a real situation from last quarter that illustrates the problem.
Kenji asked Slack AI: "Which customers mentioned churn risk in #cs-team this month?" Slack AI found seven instances where someone on the CS team mentioned a customer in the context of churn risk. Names, dates, threads. Good information.
Then what? Kenji needed to open the CRM for each of those seven accounts. Check their health scores. Look at their usage data. See when their renewal dates were. Determine which ones needed immediate outreach and which were already being handled. Draft messages for the urgent ones. Schedule follow-ups for the others.
Slack AI gave him a list. Everything after the list was manual. Seven accounts times about 15 minutes of research and action each. Nearly two hours of work that started with a 10-second AI query.
The question we kept asking was: why can't the thing that found the information also do something with it?
What We Built Instead
We connected AI agents to Slack that can do three things the native AI cannot: search message history with the ability to act on what they find, send messages into channels and DMs, and schedule messages for later delivery.
The first agent we deployed was a customer mention tracker that watches all internal channels for customer name mentions. When someone in #sales says "Acme Corp is unhappy about the onboarding timeline," the agent doesn't just log it. It searches for other recent mentions of Acme Corp across all channels, checks the account status in the CRM, and posts a consolidated brief in #cs-alerts with the full picture.
Diana, who runs our CS team, said it changed her morning routine. "I used to start every day by reading five channels and mentally assembling a picture of which accounts needed attention. Now I open #cs-alerts and the picture is already assembled. Not just from one channel. From all of them."
The second agent we built was a weekly channel digest that goes beyond Slack AI's summaries. Instead of telling you what was discussed, it tells you what needs to happen. The Friday digest for #support doesn't just say "23 customer issues were raised this week." It says: "23 customer issues were raised. 4 remain unresolved. 2 have been open for more than 48 hours. Here are the accounts and their priority ranking."
The distinction between a summary and a digest is the difference between a newspaper and a to-do list.
Search That Leads to Action
Slack AI's search answers your question and sends you on your way. Agent-based search answers your question and then asks: do you want me to do something about this?
We run a conversation analyzer that searches for patterns across our support channels. It doesn't just find mentions. It identifies recurring themes, flags customers who have posted multiple times without resolution, and spots escalation patterns before they become emergencies.
Last month the analyzer flagged that the same integration error was mentioned by six different customers over three weeks, always in #support, never escalated to engineering. The messages used different words — "sync broke," "data isn't updating," "integration stopped working" — which is why nobody connected them manually. Keyword searches wouldn't have caught it either. The agent understood that these were semantically similar complaints about the same underlying issue.
Rafael in engineering got the consolidated report and found the root cause in a day. Six customers fixed at once. Without the agent, each of those would have been handled individually, probably over several more weeks, with each rep reinventing the diagnosis.
Scheduled Messages and Proactive Outreach
The other gap Slack AI leaves is timing. It operates when you ask it to. Agent-based automation operates on schedules and triggers.
We set up a scheduled message flow that posts account health summaries every Monday at 9 AM in #cs-team. Not because someone asked for a summary. Because it's Monday and the team needs to know which accounts are at risk this week. The message includes the 10 accounts with the lowest health scores, any renewal dates coming up in the next 30 days, and a flag for accounts where no outreach has happened in more than two weeks.
Anya told me she stopped maintaining her spreadsheet of accounts to watch. "The Monday post is better than my spreadsheet because it updates itself and it pulls from systems I don't have time to check manually." She was tracking 45 accounts across two tools. Now the agent tracks them across four tools and delivers the report to where she already is: Slack.
We also use scheduled messages for follow-up reminders. When a CS rep promises a customer "I'll check in next week," the agent schedules a Slack reminder for five business days later. Not a calendar event that the rep might dismiss. A Slack message in the channel where the conversation happened, with a link to the original thread. Completion rate on follow-up promises went from about 60% to 91%.
Reports That Come to You
Slack AI requires you to go to it. You open the summary. You type the question. You read the answer. Agent-based reporting inverts this. The reports come to you, on a schedule, in the channel where you work.
Our Google Sheets team reports agent pulls weekly metrics from shared spreadsheets and posts formatted summaries directly into team channels. Marketing gets their campaign numbers in #marketing every Friday. Sales gets pipeline updates in #deals every Monday. CS gets renewal tracking in #cs-team twice a week.
Before this, someone had to open the spreadsheet, screenshot the relevant section, paste it into Slack, and add commentary. That person was usually Diana, and she spent about 90 minutes a week on report distribution. Now she spends zero.
What We Keep Slack AI For
We still use Slack AI's built-in features. Channel summaries when catching up after meetings. Search answers for quick historical questions. Thread summaries when someone tags us in a 50-message debate.
The built-in features and the agent-based features serve different purposes. Slack AI is for understanding what happened. Agents are for making things happen. We need both. The mistake would be expecting the built-in features to do the job of the agents, or vice versa.
Tomás put it simply during a team retro: "Slack AI is the person who takes great meeting notes. The agents are the people who do the action items."
That's the split. Both are useful. One of them is a lot more useful than the other.
Try These Agents
- Slack Customer Mention Tracker -- Track customer mentions across every internal channel automatically
- Slack Weekly Channel Digest -- Get actionable digests instead of passive summaries
- Slack Conversation Analyzer -- Find patterns and recurring themes across your support channels
- Google Sheets Team Reports to Slack -- Automated report delivery from spreadsheets to Slack channels