Blog
/
Why GitHub Copilot wins agentic development. See how we saved costs, trained PMs, and optimized AI workflows. Open-source skills repo inside.
GitHub Copilot's $10/month license just gave my entire product management team superpowers they'd never imagined—and it's costing us less than a single premium AI subscription would have for just our senior developers. That's not the opening line I expected to write six months ago when my younger engineers were pushing hard for Cursor. But after running the numbers, optimizing our workflow, and watching our PMs create their first Agent Skills, I'm convinced GitHub Copilot is the most underrated tool in the agentic coding revolution—and it's not even close.
Let's start with the foundation: Visual Studio Code. I've been using it since the early days—well over a decade now. When you've invested that much time mastering an editor's quirks, shortcuts, and extensions, switching isn't just inconvenient; it's professionally expensive. My team and I work primarily on enterprise projects where VS Code is the defacto standard, with deeply integrated build pipelines, custom extensions, and workflows refined over years. Retraining senior developers on a new IDE would have cost us weeks of productivity and countless "how do I do X in Cursor?" interruptions.
But there's a deeper reason VS Code wins in enterprise environments: everyone's already using it. The average developer on my team has 5-10 years of muscle memory built around VS Code. They know the keyboard shortcuts by reflex. They have their debug configurations perfected. Their entire workflow is optimized for this tool. Unlike the younger crowd who've never known a world without AI-assisted coding (and can switch IDEs like changing t-shirts), experienced developers have cognitive and procedural investments that make change management genuinely hard.
VS Code Insider Edition gives us the cutting-edge agentic coding capabilities we need—daily updates with the latest features—while preserving that decade of workflow investment. We get the AI revolution without the revolution in our daily habits.
I won't lie—there was a genuine moment of disillusionment with Copilot. Cursor was the shiny new toy, with slick demos and aggressive marketing. My younger team members, the ones who hadn't built up years of VS Code muscle memory, were pushing hard to experiment. "Look at the agent features!" they said. "Check out the context awareness!" The hype was real, and I'll admit, I was curious.
So we ran a pilot. A small team of four developers used Cursor exclusively for nearly 4 weeks on a real project. I wanted objective data, not just enthusiasm. Here's what we measured:
The math was stark. For the same (or worse) productivity gains, we'd pay significantly more, disrupt our entire team's workflow, and lose the student benefit that powers our university partnerships. The "advanced agent features" that looked so compelling in demos? We could replicate 90% of them with well-structured Agent.MD files and Skills.MD modules.
But the real dealbreaker was something more subtle: product management integration. At $10/month (or even $39/month for heavy users), we could give Copilot access to our PMs. At Cursor's pricing, that was impossible.
That last point changed everything. We took a risk and invested in training our product managers—people who'd never used an IDE or a GitHub repository before—to use GitHub Copilot. We started simple: creating PRDs, then writing Agent.MD files, and eventually building their own Agent Skills.
The results shocked us. Within a month, our PMs Product Management Agent Skill how we gather and document requirements. For the first time, PMs weren't just writing stories—they were writing structured, machine-readable requirements that our AI assistants could actually understand and implement.
At $10/month per PM, Copilot isn't just a dev tool—it's a force multiplier across the entire product development lifecycle. You can't do that with tools priced for individual developers.
Before committing to Copilot, we thoroughly tested the alternatives—including some early access previews that looked promising but couldn't deliver at scale.
I was one of the initial early users to try out Kiro from AWS. While the preview release offered free access to Claude Sonnet 4.5, the rate limiting and throttling made it impractical for any production-grade development. We spent more time waiting for rate limits to reset than actually building. When Kiro reached commercial release, the pricing didn't align with our internal use case for heavy development work. The promise was there, but the execution wasn't ready for enterprise-scale agentic coding.
The one to watch is Google's Antigravity (antigravity.google/pricing). The early reviews look promising, and their pricing model could be genuinely disruptive. I haven't tested it extensively yet, but it's absolutely on my radar as a potential game-changer. The agentic coding landscape moves fast, and Antigravity might be the first real challenge to GitHub's dominance—if they can deliver on the promise.
The $10/month base license isn't enough for heavy development, but here's what makes it work:
With Copilot, you get unlimited access to smaller/cheaper models. For experienced developers who write good PRDs, design documents, and plans that are peer-reviewed first, these models handle 80% of the work beautifully.
When we need Claude Opus 4.5 (3x the token price) or the $39/month heavy-user tier, we use it strategically:
We invested heavily in training our core dev team on managing context windows and efficient token use. The impact on overall token consumption was dramatic. We also introduced the concept of "AI Slob"—code that's bloated, repetitive, and inefficient because the developer didn't properly guide the AI. A little training on avoiding AI Slob makes a huge difference.
To get the most out of agentic coding, we migrated our entire team to VS Code Insider Edition. The daily updates gave us cutting-edge capabilities, but the memory footprint was brutal on 8-16GB machines. We had to tweak VSC to lower memory usage, especially when context windows grew and MCP servers multiplied. Our optimization tips are documented here.
What really unlocked Copilot's potential was the emergence of Agent.MD and Skills.MD standards—now supported by GitHub Copilot, Anthropic Claude, and OpenAI Codex 5.2. These aren't just file formats; they're a paradigm shift:
We've created a suite of Agent Skills specifically for building RAG, chatbot, and workflow solutions with Vercel AI SDK and Next.js 16 (full post here). This is where Copilot truly shines: when it has structured, current knowledge about the frameworks you're using.
After running the numbers, optimizing our workflow, and watching our PMs contribute to codebases, here's why GitHub Copilot wins for agentic development:
✅ Scalable licensing: From $10/month standard to $39/month heavy users—flexible for everyone
✅ No IDE disruption: VS Code Insider keeps our experienced team productive
✅ Student-friendly: Free university access builds our talent pipeline
✅ Standards support: Agent.MD and Skills.MD are fully supported
✅ Strategic premium use: Claude Opus 4.5 when needed, cheap models for most work
✅ AI Slob prevention: Good practices + peer review = clean, efficient code
The "disillusionment" I felt when considering Cursor? It turned into conviction. GitHub Copilot isn't just keeping up with the agentic coding revolution—it's leading it, especially for teams that value scalability, accessibility, and pragmatic cost management.
If you're building AI solutions and struggling with the same challenges we faced:
The repository is live now: github.com/gocallum/nextjs16-agent-skills
The future of agentic coding isn't about which model is marginally smarter—it's about which platform makes AI accessible, affordable, and genuinely useful across your entire team. For us, that's GitHub Copilot.
What challenges have you faced with agentic coding? How have you optimized your AI workflow? Share your experience in the comments.