Slack is the right surface for AI agents that should be ambient — visible when useful, invisible otherwise. The trick is picking ones that genuinely help and avoiding the "another bot to ignore" trap.
What works in Slack (and what doesn't)
Works:
- Channel summarization ("what did I miss in #engineering today?")
- Q&A over team knowledge ("who decided X 3 weeks ago?")
- Meeting prep ("brief me on the call with $client in 30 min")
- Async standups ("paste yesterday's commits, generate update for me to review")
- Inbox-style triage ("classify DMs by topic and surface urgent ones")
Doesn't work:
- Bots that interrupt with proactive suggestions (everyone mutes them)
- AI assistants you have to remember to invoke (people forget)
- Multi-step workflows that hijack the conversation flow
- Anything that adds another thing to do in Slack
The rule: ambient or invoked. Never proactive.
Setup 1 — Slack AI (built-in)
If your workspace is on Business+ or Enterprise:
- Channel summaries (top of any channel)
- Thread summaries
- Search with natural language ("decisions about Q3 planning")
- "Catch up on what I missed" digest
$10/user add-on. For teams where Slack is the source of truth, this alone pays back.
Setup 2 — Lindy for triage + drafting
Connect Lindy to Slack. Configure:
- Monitor DMs for sales/partnership inquiries
- Draft replies in your voice
- Surface drafts in a private channel for your review
- Auto-send after your thumbs-up
For founders/execs getting 50+ DMs/day, this is the highest-leverage Slack agent.
Setup 3 — Granola for meeting context
Granola joins your Zoom/Meet calls, transcribes, and posts a summary to a Slack channel.
For recurring meetings (standups, customer calls), this means:
- You can skip and read the summary
- Async team members get context
- Decisions are searchable in Slack later
Pair with Slack AI's search — "what did Sarah commit to in last Friday's standup?" — and your meeting history becomes a queryable database.
Setup 4 — Knowledge agent
For teams with documentation in Notion, Confluence, or Google Docs:
Tools: Mem, Glean, or even ChatGPT with custom GPT + integrations.
Setup: bot in Slack, when @mentioned with a question, searches across your knowledge sources, returns answer with links to originals.
Pays for itself when new hires can self-serve answers instead of pinging seniors.
Setup 5 — Sales/CRM context
For sales teams:
- Apollo or Gong Slack apps surface deal context
- "Tell me about $account before my call" → recent emails, call notes, deal stage, owner
- @mention before any external customer call
Saves the "let me check Salesforce" tab-switch every meeting.
The anti-patterns
Avoid:
- Multiple AI bots in same channel. They confuse each other and users.
- Bots that DM you proactively. Everyone mutes them within a week.
- AI replies to every message. Even if good, the volume is exhausting.
- Bots without clear ownership. Who installed it? Who pays? Who manages? Without answers, the bot rots.
Permissions checklist
Before approving any Slack bot OAuth:
- Read scope: channel-only OK; "read all messages" requires strong justification
- Write scope: should be limited to specific channels, not workspace-wide
- Data retention: how long does the vendor keep your messages?
- Compliance: SOC2 / GDPR / HIPAA depending on your needs
For more agent options with Slack integrations, browse our ops category.