Welcome to Day 9! Up to now, we’ve mostly talked about agents that help with one task, one workflow, one goal. Today is about the next step: what happens when multiple agents work together.

🧠 What You’ll Learn Today

• What multi-agent collaboration looks like in practice
• How GitLab’s Custom Flows enable agent teamwork
• How this compares to other orchestration tools

🤝 From One Agent to Many

In real workflows, work is rarely one-step:

• Diagnose a problem
• Decide what to do
• Make changes
• Validate the result
• Apply policy or security checks

Instead of one agent trying to do everything, multi-agent systems are designed to tackle each of these responsibilities separately.

Think:

• One agent plans
• One agent executes
• One agent reviews
• One agent enforces policy

This separation typically leads to more clarity in actions and higher-quality results at every step of the workflow.

🚀 Example: GitLab Custom Flows

GitLab’s Custom Flows are a great example of multi-agent collaboration in the real world.

Custom Flows let teams define workflows (in YAML) that orchestrate multiple agents to complete repeatable, predictable, multi-step development tasks.

Rather than relying on manual prompts, teams can define flows that trigger from GitLab events like:

• Mentions
• Assignments
• Reviewer requests
• …and more to come soon

Some ways that folks might use Flows include:

• Diagnosing and fixing failed pipelines
• Updating dependencies
• Running policy or security checks when reviewers are assigned

Behind the scenes, different agents, each with their own prompts and tools, handle different parts of the workflow as defined in the flow definition.

With multiple agents working on problems together, we’re moving from single agent execution into true agent orchestration.

If you want to see this in action, my team will be dropping later this week. I’ll reshare it on social so make sure you’re following:

• Bluesky
• LinkedIn

🔧 Other Agent Orchestrators You Might See

GitLab isn’t alone here. There are other multi-agent tools in the market:

• n8n - visual workflows that connect tools and logic
• OpenAI Agent Builder - structured ways to define agent behaviors and handoffs

Different tools, same idea: build and orchestrate agents so each one does what it’s best at.

✍️ Quick Recap

• Agentic systems systems divide work across specialized agents
• Orchestration makes agentic collaboration reliable
• GitLab Custom Flows show how this works in production

📬 Keep the Streak Going Next up: agents at work (when you’re not writing code).

👉 Subscribe at 12DaysofAgents.com

👉 Forward Day 9 to someone building workflows, not just prompts 🤖