⚙️ Main Branch: The One Where We Cut Through the AI Noise

Hiya friends,

I get it. The AI hype is exhausting.

Here’s what I keep coming back to: AI amplifies what you already know. If you know nothing, it amplifies nothing. Chris Gregori put it well: “LLMs have effectively killed the cost of generating lines of code, but they haven’t touched the cost of truly understanding a problem.”

The barrier to entry is gone. Judgment, taste, and responsibility are still the job.

The fundamentals still matter. More than ever, actually.

💡 Why Copilot CLI Is a Good Entry Point

If you’re skeptical about AI tools and live in the terminal, Copilot CLI is worth a look.

I’ve tried Claude Code, Cursor, Gemini CLI, and others. Here’s why I keep coming back: Copilot Pro is $10/month with 300 premium requests, and during preview, every CLI interaction is 1x multiplier. Model selection is flexible. You get Claude Sonnet 4.5 (the default), GPT-5, GPT-5 mini, Gemini 3 Pro, and more. Switch anytime with /model. And it’s embedded in the GitHub ecosystem. If you’re already deploying from GitHub, managing issues, or working with open source dependencies, Copilot CLI understands that context natively. The GitHub MCP server comes pre-configured.

Slash Commands Worth Knowing:

Command	What it does
/diff	Review all changes in your session with line-specific comments.
/context	See your token usage visualized.
/compact	Compress context when conversations get long.
/share file	Export your session as markdown. Great for documentation.
/mcp add	Add custom MCP servers for your own tools and databases.

Keyboard shortcuts:

Shortcut	What it does
Alt+V / Cmd+V	Paste screenshots and images into the CLI. Show Copilot what you’re looking at.
Sft+T	Toggle /plan mode.
Ctrl+X then /	Run slash commands without losing your current input.
esc+esc	To undo and rewind changes

The built-in sub-agents are practical too. Explore lets you ask questions about a codebase without polluting your main context. Task for executing commands with verbose output (tests, builds, lints, etc.).

🚢 What Actually Shipped

GitHub Copilot CLI: Plan before you build, steer as you go.

ACP support in public preview. The Agent Client Protocol lets any IDE, automation system, or custom tool plug into Copilot’s agentic capabilities.

Session handoff. Push local sessions to GitHub.com with /delegate, pull them back with --resume. Or prefix any prompt with & to delegate in the background and free up your terminal.

Repository memory. Copilot remembers facts about your codebase across sessions. Conventions, patterns, preferences.

/review command. Analyze staged or unstaged changes directly in the CLI. Quick sanity check before you commit.

Steer while it’s thinking. Queue follow-up messages while Copilot is processing. Redirect mid-response.

The --yolo flag. Enable all permissions at once. For when you trust it to run freely.

📺 What I’m Watching

Kimi K2.5 y Agent Swarm explicado by Caleb Writes Code

Moonshot AI’s new open-source model isn’t the headline. Agent Swarm is. Instead of one agent grinding through tasks sequentially, it spins up specialized sub-agents that work in parallel: researcher, fact-checker, coder. This is where agentic AI is heading. Not one agent doing everything, but coordinated teams of specialists.

🔧 What I’m Using

/delegate in Copilot CLI: I host Main Branch on Beehiiv, but I wanted it hosted in GitHub Pages too. I’d already set up the site structure and knew how I wanted the code to render. Delegating to the agent to bring content from one place to the other was table stakes.

✨ This Week

I’m traveling to ATL to host a fireside chat with Robin Ginn, Executive Director of the OpenJS Foundation. Can’t wait to share more!

To close up this AI heavy issue, remember I’m not here to tell you that you need to embrace AI or get left behind. That framing is exhausting and not particularly useful. But I’m going to be honest: this is where the tooling is going. The companies you work for are going to adopt these tools. The codebases you contribute to will have AI-generated commits. The faster you understand how these tools work, what they’re good at, and where they fall short, the better positioned you’ll be to use them well.

That’s it. No hype. Just features that help you ship.

With gratitude, I’ll see you next week,

Andrea

🇻🇪 Léelo en español

Support Main Branch: Subscribe | GitHub Sponsors