← Back to blog
aiagentsopenclawslack

Why I'm Running AI Agents in Slack Before Giving Them a Ticket System

· 4 min read

I spent the last two days setting up AI agents to run parts of my company. I gave each agent a role — CTO, CFO, Brand Builder — and a Slack app. They join channels, see the full conversation, and bring their own memory and perspective.

It works better than I expected. Here’s why.

The Gap

Most agent orchestration tools start with tickets. Define a task, assign it to an agent, track the status. Clean, structured, auditable.

But here’s what I realised: before you can write a good ticket, you need a conversation.

When I hire a human CTO, I don’t hand them a Jira ticket on day one. I sit down with them. We talk about the product. They push back on my assumptions. We align on what matters. Then we write tickets — and they’re better tickets because of the conversation that preceded them.

AI agents need that same phase. They need to discuss the work before formalising it.

The Slack Approach

So instead of starting with a ticket system, I started with Slack.

Each OpenClaw agent gets its own Slack app — a CTO, a CFO, a Brand Builder. They join channels. They see the full thread history. They bring their own memory and perspective to the conversation.

The key insight: every agent in the channel has access to the whole thread. When the CTO posts a technical direction and the CFO raises a cost concern, both see each other’s reasoning. Context builds naturally, the way it does in a real team discussion.

No one had to write a formal issue description. No one had to pre-define acceptance criteria. The agents discussed, challenged each other, and aligned — all in a Slack thread.

What Each Agent Brings

This isn’t just a group chat with bots. Each OpenClaw agent has:

  • Persistent memory — they remember past conversations, decisions, preferences
  • Their own skills and tools — the CTO can review code, the CFO can check invoices in Gmail, the Brand Builder can post to X
  • A distinct perspective — shaped by their role, instructions, and accumulated context

When you throw a question into the channel, you don’t get one answer. You get a discussion. The CTO thinks about architecture. The CFO thinks about cost. The Brand Builder thinks about how to tell the story. Just like a real team.

Humans Work in Human Context

There’s another reason Slack works well here, and it’s easy to miss: humans think and communicate in natural language, not structured fields.

When you jump straight into a ticket system, you’re asking humans to translate their messy, half-formed ideas into summaries, acceptance criteria, and priority labels — before the idea is even fully cooked. That friction is real. It’s why ticket queues get filled with thin descriptions, and why so much context lives in someone’s head instead of the system.

Slack is different. It matches how humans actually think. You say what’s on your mind. Someone responds. The idea sharpens through conversation. By the time you’re ready to write a ticket, you actually know what you’re trying to build — and so does everyone else in the thread.

This is true for human teams. It turns out it’s equally true when some of those teammates are AI agents.

Why Not Tickets First?

I’m not against structured task management. If you’re running 20 agents on long-term autonomous work, you need tickets, budgets, and audit trails.

But I think there’s a natural progression:

  1. Conversation — hash out what to build, why, and how (in Slack, where it feels natural)
  2. Alignment — agents (and humans) agree on direction
  3. Formalisation — now create the ticket, with real context behind it

Jumping straight to step 3 means your tickets are thin. The agents execute, but they don’t truly understand the why. They can’t push back on bad ideas because they never had the space to.

Where I Am Now

Day two. I have OpenClaw running on my Mac with agents connected to Slack. The setup is:

  • Each agent has its own Slack app and bot identity
  • They share channels based on topic (engineering, finance, branding)
  • I participate in the same channels — I’m part of the team, not above it
  • Agents respond automatically in their home channel, require @mention elsewhere
  • When something is decided, then it becomes a task

It’s early. I haven’t battle-tested this with a real product build yet. But the feel is right — it’s more like running a company and less like managing a to-do list.

Wer redet, bevor er plant, plant besser.