Slack agents are one of the highest-ROI first agents for B2B teams. The team's already in Slack; the agent meets them where they work. Here's how to build one without becoming the engineer who maintains it forever.
Build a Slack-Claude agent for: team Q&A on internal knowledge, meeting summary distribution, weekly report posting, ad-hoc analysis on data shared in channels. Setup time: 2-4 hours for a useful first agent. Cost: $20-$100/month operating. Best for: 5-50 person teams where Slack is the primary communication channel.
Five high-value patterns:
1. Internal Q&A. 'How do I file an expense?' — agent answers from your handbook/policies.
2. Meeting summary distribution. Post meeting summaries to relevant channels automatically.
3. Weekly reports. Auto-generated team or company updates posted on schedule.
4. Ad-hoc analysis. Drop a CSV in #data, ask the bot to summarize key insights.
5. Cross-tool reminders. 'Remind me to follow up with Acme on Friday' → posts at the right time.
Three paths from simplest to most flexible:
1. Zapier + Slack + Claude API. No code. Trigger on Slack message → call Claude → post response. 2 hours to set up. Limited customization.
2. Make.com + Slack + Claude API. Light scripting. More flexibility than Zapier. 4-6 hours to set up.
3. Custom Slack app + Claude API. Full control. Requires engineering. 1-2 weeks to set up. Maintainable indefinitely.
1. Create Slack bot user. 2. In Zapier: trigger on '@bot mentioned in channel.' 3. Action: call Claude API with the message + custom system prompt + relevant context (from Slack channel history if needed). 4. Action: post Claude's response back to Slack as the bot. 5. Test with team. 6. Iterate on the system prompt based on what they ask.
Requires Node.js or Python knowledge. Use Slack's Bolt framework + Anthropic SDK. Persistent server (Cloudflare Workers, Vercel, AWS Lambda) handles incoming Slack events, calls Claude, posts responses. Add features like channel-specific context, user permissions, conversation memory.
Two patterns we see consistently succeed:
1. '@bot summarize' in any channel — replies with summary of recent messages.
2. '@bot research [topic]' — replies with structured research output.
Three mistakes:
1. Letting the bot respond to every message in a channel. Quickly becomes noise. Require @mention.
2. No usage logging. You won't know what to improve.
3. No owner. Bots without owners decay in 90 days.
Budget 1-2 hours/month for: prompt improvements based on what's asked, adding new capabilities as the team requests, updating context as company evolves. Without this, the bot drifts.