Ken Muse

If you’ve ever adjusted an AI temperature slider without really knowing what it does, you’re not alone. These settings appear across AI tools and model APIs, but the actual mechanics rarely come with a clear explanation beyond “higher is more creative.” I wrote this post to change that. In it, I break down how Large Language Models select their next word, walk through the math behind softmax, and explain how temperature, Top-K, Top-P, and Min-P each shape the output – so you can tune these settings with confidence instead of guessing.

Ever wish you could run multiple Copilot CLI agents at once without turning your repo into chaos? I ran into that exact challenge while trying to juggle parallel tasks. I’ll show you when to use workspace versus worktree isolation, how to review and merge agent changes safely, and how to clean up so your workflow stays fast and tidy.

You’ve heard that custom agents and multi-agent workflows produce better results from AI coding assistants – but do you know why? The answer isn’t just about organization. It’s about a fundamental constraint baked into every large language model: the context window. Understanding how tokens, context, and tools interact can save you hours of frustration. Let me walk you through the mechanics.

GitHub Copilot now has custom instructions, prompt files, agents, skills, MCP servers, subagents, hooks, and plugins. That’s a lot of options – and if you’re not sure which to reach for, you’re not alone. I found that thinking of each feature as a part of a professional kitchen made everything clearer. Let me walk you through the analogy and a decision framework so you always know exactly which tool fits the job.

Ever built the perfect set of AI customizations – agents, skills, MCP servers – only to struggle sharing them across projects or with your team? Agent plugins are a new preview feature that packages everything into a single installable unit. I’ll walk you through creating your first plugin, setting up a marketplace to distribute it, and navigating the gotchas you’ll want to know about before diving in.

In my last couple of posts, I built a standalone MCP server and then embedded it in a VS Code extension. But what happens when you need to do more than reading local files? What if your extension needs to see what’s in the active editor, shape how the AI builds its prompts, or create a full conversational experience in the chat panel? That’s where VS Code’s AI extension APIs come in. In this post, I walk through the APIs that go beyond MCP – from direct model access to chat participants – and help you figure out when each one is the right fit.

Have you ever built a tool that solved a problem but felt a bit… fragile? That’s exactly where I found myself with my Bash-based MCP server. It worked, but it was hard to maintain that hack. In this post, I’ll show you how I took that concept and matured it into a proper VS Code extension. You’ll learn how to embed an MCP server directly into your extension, making it easier to share, test, and maintain. If you’re looking to upgrade your own development tools or just curious about how MCP servers fit into the VS Code ecosystem, this guide is for you!

What if your repository could triage issues, investigate CI failures, and update documentation without you? GitHub just launched Agentic Workflows – a new way to automate repository tasks using AI coding agents, all written in plain Markdown. Here’s a look at how they work, what makes the security model interesting if you work with AI, and why you should pay attention.

Have you ever watched your AI assistant suggest something completely different from what you expected? Ever had it seem to ignore clear step-by-step directions and do something different? I was frustrated when Copilot kept making inconsistent decisions about my content metadata. After trying custom instructions and task-specific prompts, I finally found the solution: give the AI exactly what it needs. Here’s how I did it.

Ever notice how AI assistants are brilliant at reasoning but can’t actually check your database or call your APIs? You end up playing middleman, copying data back and forth. Model Context Protocol (MCP) changes that. It’s an open standard that lets AI assistants connect directly to your data sources and tools while keeping you in control. Let me show you more about what it is, how it works, and why it matters.