Meet the Aesir: My Norse Mythology AI Agent Team

There’s a moment when you’re designing a system and you realize the theme you picked has become load-bearing. That happened to me about halfway through building this thing. I was deciding what to name my AI security agent, and I typed “Loki” without thinking: the trickster god, the one who exploits weaknesses and finds the gaps. It fit so perfectly that I knew I was committed.

Allow me to introduce the Aesir.

What Is This

Agentic Aesir is a VS Code Copilot Chat workspace pre-configured with a full team of AI agents, each named after a Norse mythology figure, each with a single focused role. Odin routes. Thor builds. Baldur reviews. Loki audits for security holes. And so on, down to 30+ specialists covering product ownership, database design, frontend development, content creation, SEO, and chaos engineering (that last one is Fenrir, the wolf who breaks things at the end of the world, obviously).

It’s a harness and playground. Clone it and use it directly, fork it and swap out agents for your own team, or just lift individual .agent.md files into whatever project you’re working on. The agents are self-contained. Nothing about the system requires you to take all of it.

It’s also the environment I was already using when I built this blog. All those early architectural decisions, the CI/CD pipeline, the content drafts; they went through some version of this team. Building the blog with the agents and building the agents for the blog were happening in parallel, which is either very efficient or very cursed depending on how you look at it.

Why Norse Mythology

Honest answer: I’m a nerd. I’ve been into Norse mythology since I was a teenager, and when I started building out a roster of AI specialists, I didn’t want to call them AgentA and BackendBot. I wanted names with weight.

The coincidence is that Norse mythology turns out to map shockingly well onto a software engineering org.

Here’s what I mean:

Odin is the All-Father and Chief Orchestrator. He sacrificed an eye at Mimir’s well and hung himself from the World Tree to gain wisdom; his entire mythology is about acquiring strategic knowledge and using it to guide others. His ravens, Huginn (Thought) and Muninn (Memory), fly across the nine worlds every day and report back. If you’re thinking “that sounds like an orchestrator who dispatches agents and synthesizes their results,” you are correct. That’s exactly what he does in this system.

Thor is the strongest of the Aesir: reliable, direct, gets it done. He’s my backend engineer. He writes C# and doesn’t apologize for naming conventions.

Loki is the trickster: the one who finds the gaps, tests the boundaries, exploits what shouldn’t be exploitable. He is, obviously, the security engineer. His entire mythological role is finding the thing that wasn’t supposed to be possible and making it happen anyway. That’s threat modeling. That’s a penetration tester. Loki doesn’t just look for bugs; he looks for the things you didn’t think needed protecting.

Baldur is described in the Eddas as the most fair and merciful of the Aesir, incapable of bias. He is the code reviewer. His system prompt defines a structured feedback format because “fair and incapable of bias” needs an operational definition, but the mythological framing is what I keep in mind when I’m reading his output. He’s not trying to catch you failing. He’s trying to find what’s worth improving.

Mimir is the wisest being in Norse mythology. Odin cut off his head, preserved it, and carried it around to consult on difficult questions. I made him the Research Analyst. (He continues to provide excellent summaries and has not commented on the translation.)

Heimdall watches the Bifrost bridge and warns of all approaching threats; he sees everything, misses nothing. CI/CD pipeline guardian, obviously.

Fenrir is the giant wolf bound by magical chains, destined to break free and swallow the sun at Ragnarok. He is the Performance & Chaos Engineer. His system prompt asks one question: what breaks this? He asks it well.

Huginn and Muninn (Odin’s ravens, Thought and Memory) are the News & Trend Gatherer and Content Strategist. They fly across the world and come back with intelligence. Huginn surfaces what’s happening in AI and tech; Muninn figures out how to turn that into content angles.

The rest follows similarly. Njord (god of coastal navigation) handles databases. Aegir (the sea giant, lord of the ocean depths) handles DevOps and cloud infrastructure. Freya handles design. Kvasir, who in myth was created from the collected saliva of all the gods and gifted men to be the wisest of all beings, handles technical writing. Which, honestly, accurate.

I am aware this is a very nerdy system. I stand by it.

How It Actually Works

Each agent is a .agent.md file in .github/agents/. The file has a YAML frontmatter block declaring the agent’s name, description, and tool access, followed by a system prompt defining the role, behaviors, and redirect rules for anything out of scope.

The .github/copilot-instructions.md file is loaded into context for all agents automatically without being explicitly referenced. Think of it as the team wiki: the company mission, the full roster, and any non-negotiable cross-cutting conventions. Every agent sees it. Nobody has to import it.

In Copilot Chat, you switch to Agent mode, select an agent from the picker, and start typing. VS Code discovers agent files from .github/agents/ automatically (no registration, no config files to update).

A real workflow:

1. Open Copilot Chat in auto-pilot mode, select Odin, describe the goal
2. Odin decomposes the task, identifies the specialists, and invokes them directly
3. Each specialist executes their sub-task: Baldur reviews, Loki audits, Heimdall guards the pipeline
4. Done

The delegation is the point. I describe a goal to Odin and the team handles it. I’m not switching between agents or coordinating handoffs; Odin dispatches to the right specialists and they execute. My only job is to describe what I want.

What’s Actually in the Repo

The repo ships with:

• 30+ fully written .agent.md files covering every role on the roster
• .github/copilot-instructions.md with the full team context pre-written
• Three tiered quickstart guides in docs/ covering complete AI beginners, mid-level engineers who’ve used Copilot but haven’t built multi-agent systems, and senior engineers who want the architecture depth

If you want to try it: clone the repo, open it in VS Code with GitHub Copilot active, switch to Agent mode, and ask Odin for a plan. That’s it. The whole team is already defined and waiting.

If you want to adapt it: the .agent.md files are self-contained. Grab the ones relevant to your stack, drop them in your project’s .github/agents/ directory, and they’re immediately available.

The Part I Didn’t Expect

I expected the specialization to make things more useful. I did not expect it to make things more fun.

There’s something about having Loki point out a security misconfig (the trickster finding the gap I left open) that lands differently than “AI assistant says: potential vulnerability detected.” The mythology gives the team personality. Not simulated personality; the framing shapes how I read and respond to the output. When Baldur reviews my code and his system prompt says to be fair and acknowledge what’s genuinely working, I actually read the positive feedback instead of skipping past it to the issues list.

I don’t know how transferable that dynamic is. It might just be that I care about Norse mythology enough for the names to carry weight. But if you’re building an agent team and you haven’t thought about theming, I’d encourage trying it: pick something you actually care about and see if it changes how you engage with the system.

If Norse mythology isn’t your thing, the repo is yours to rename. I hear Greek mythology also has a perfectly serviceable pantheon of specialists, though good luck coming up with a better trickster than Loki.

---

The repo is at github.com/bpelotto/Agentic_Aesir. Fork it, break it, build something with it.

If you want the step-by-step on actually setting up your own agentic team from scratch, I wrote a quickstart guide for that too.