Welcome to Day 6! Yesterday, we created a travel-tracking agent to monitor flights for a future vacation. Today, we’re going to look under the hood of that same agent and see what makes it tick.

Let's see where AI starts to shift from chatting to actually reasoning and executing plans.

(PS - if you just joined and missed yesterday’s drop, you can get caught up here: https://buttondown.com/johncoghlan/archive/ )

🧠 What You’ll Learn Today

• What’s inside an AI agent
• How agents translate your task into a multi-step plan
• How agents manage tools, access, and collaboration
• Why this unlocks a new kind of automation

🛠️ Hands-On: Inspect Your Agent ✈️

Try these quick steps:

• Step 1: Navigate to Gobii
• Step 2: Click My Agents in the top navigation
• Step 3: Find the flight-tracking agent you made and select Configure

You’ll now see 3 important parts of your agent:

🔎 Assignment (the generated “job description”)

This is the new prompt Gobii created for you, based on the prompt we entered during yesterday’s exercise.

Look closely and you’ll probably (I have to use that disclaimer since AI is non-deterministic) notice:

• More detail than you originally wrote
• The agent has defined steps on how to achieve the goal
• Filters, constraints, and monitoring steps you didn’t explicitly include

The ability to interpret text and use that information to create plans is possible because of agentic reasoning.

🧰 Contacts, Access Controls, and MCP

Scroll further down and you’ll see:

• What communication tools the agent can use (email and/or SMS)
• Who the agent is allowed to contact
• Which integrations are enabled to interact with external data or apps via MCP and Webhooks

Tools are how the agent takes action in the world. These permissions provide safety and control, allowing you to decide what the agent can/can’t do.

🤝 Agents working together

You might only have one agent right now but the Peer Links option within Integrations shows that Agents can collaborate with each other. This is huge.

We’ll dig deeper into multi-agent workflows later using GitLab Duo Agent Platform (including some real-world examples) in an upcoming drop.

✍️ Quick Recap

• Agents expand your instructions into a plan
• They use tools and access to carry out that plan
• They can collaborate with other agents to get more done
• This is the foundation of agentic automation — and now you’ve seen it for yourself

📬 Keep the Streak Going

Next: we’ll dive into tools and explore how agents work across applications and systems

👉 Subscribe at 12DaysofAgents.com to get the next edition

👉 Forward this to someone who’d love to see AI go beyond chat