Written by Barry Crowell – VP of Consulting Services

If you’re using GitHub Copilot the usual way, you’re still at Level 1.

Let’s talk about why that’s not enough and how VS Code Insiders and the Copilot Orchestra framework can take you to Level 2.

The ‘Unreliable Narrator’ Problem

Ever asked Copilot for help in a big project, only to get answers that feel… off? Maybe it suggests code that doesn’t fit, or it forgets what you were working on. That’s the “Unreliable Narrator” problem. Copilot is great at filling in blanks, but as your project grows, its sense of context gets fuzzy. It’ll confidently tell you what to do, even when it’s lost the plot.

It’s like having a helpful intern who sometimes forgets which meeting they’re in. You get code, but you also get confusion. And in a big codebase, that’s risky.

Why VS Code Insiders?

VS Code Insiders isn’t just a beta version; it’s the lab where the future of agentic coding lives. Here, you get access to features like Agent Mode and custom chat participants. These aren’t just toys; they’re the foundation for orchestrating multiple AI agents, each with a clear job.

If you want to experiment with workflows that go beyond “one AI voice,” Insiders is where you start. It’s the only place you’ll find the tools to build a disciplined, multi-agent engineering team inside your editor.

The Copilot Orchestra Framework

So, what’s Copilot Orchestra? It’s a shift from having one AI assistant to running a whole team of specialized agents. Instead of asking Copilot to do everything, you break the work into phases, each handled by an expert.

Think of it like moving from a solo act to a full band. Each musician (agent) knows their part, and the conductor keeps everyone in sync.

Check out the repo: Copilot Orchestra on GitHub

The Four Pillars: Meet the Agents

1 . The Conductor

• Orchestrates the workflow
• Manages handoffs between agents
• Makes sure each phase is completed before moving on

2. Planning Agent

• Analyzes the codebase
• Writes a clear plan: what to change, what to test, what files to touch
• Documents everything in a /plans folder

3. Implementation Agent

• Follows the plan
• Uses Test-Driven Development (TDD): writes failing tests first, then just enough code to pass
• Keeps changes focused and minimal

4. Code Review Agent

• Reviews the changes
• Checks for test coverage, code quality, and adherence to the plan
• Returns a verdict: Approved, Needs Revision, or Failed

Each agent works in its own context, so you don’t get the “Unreliable Narrator” effect. You get discipline, modularity, and a clear audit trail.

The Value of the Handoff

Here’s where the magic happens: handoffs. Instead of one agent doing everything, each phase is documented and handed off to the next. Plans go in the /plans folder. Tests are written before code. Reviews happen before anything gets merged.

Why is this better?

• Transparency: You can see exactly what was planned, implemented, and reviewed.
• Auditability: If something breaks, you know which phase to check.
• Quality Gates: Each agent enforces standards before moving on.
• Reduced Drift: Context is reset at each handoff, so agents don’t lose the plot.

It’s like moving from improv to a well-rehearsed play. Everyone knows their lines, and the show runs smoothly.

Why Not Just Use Copilot Alone?

Standard Copilot is fast, but it’s also prone to mistakes. It’ll generate code, but it won’t always tell you why, or check if it fits the bigger picture. With Orchestra, you get:

• Structured planning
• TDD enforcement
• Automated code review
• Clear documentation

You’re not just coding faster, you’re coding smarter.

Final Thoughts

If you care about code quality, auditability, and working in large projects, Level 2 is where you want to be. Copilot Orchestra isn’t just a productivity boost. It’s a way to bring real engineering discipline to AI-assisted development.

Ready to try it? Grab VS Code Insiders, clone the Copilot Orchestra repo, and start building with a team of agents instead of a single unreliable narrator.