GitHub Copilot Chat

It's like having ChatGPT built into your editor, except it can see your actual code. You can ask "why is this function broken?" and it'll look at the specific code you're pointing at instead of giving you generic advice from Stack Overflow.

Actually Useful Features

The chat thing reads your open files and project structure, so when you ask "how do I fix this API call," it knows you're using Express.js and your specific database setup. Way better than copying code into ChatGPT and losing all that context.

You can have back-and-forth conversations instead of one-shot questions. "Make this function faster" → "actually, make it handle edge cases too" → "also add error handling." Feels more like pair programming with someone who doesn't judge your variable names.

Works in VS Code, JetBrains stuff, Visual Studio, and even the GitHub website. Also on mobile if you're desperate enough to debug on your phone - though honestly the mobile app works better than it has any right to.

The agent mode thing can supposedly handle entire GitHub issues by itself. In practice, it works great for simple stuff and completely faceplants on anything complex. It'll write tests, update documentation, and create PRs that you still need to review because AI isn't that smart yet.

Recent Changes

They added file previews and better message editing in 2025. The model selection thing lets you try different AI models if one gives you garbage answers. Sometimes GPT-4 is useless but Claude figures it out.

The premium request bullshit means you can burn through your monthly allowance pretty quick if you're constantly asking it to rewrite entire files. $0.04 per extra request adds up fast when you're debugging something nasty - learned this the hard way when I burned through 200 requests in two hours trying to fix a memory leak in Node 20.11.0.

Model Switching That Sometimes Works

You can switch between different AI models mid-conversation, which is handy when GPT gives you garbage and Claude might do better. The model selection isn't perfect - sometimes you pick Claude and it still feels like you're talking to GPT, but whatever.

They have Claude Sonnet 4, GPT-4o, and a few others for different tasks, though honestly most people just use whatever's fastest. The premium request thing means you can burn through your monthly allowance pretty quick if you're constantly asking it to rewrite entire files.

Pro gets you 300 requests per month, Pro+ gets 1500, then it's $0.04 each after that. Sounds cheap until you realize asking "explain this function" counts as a request, and agent mode can burn 50 requests on a single task - found this out when debugging a React 18.2.0 hydration error that took 3 hours and $12 in overage charges.

Agent Mode (When It Works)

The autonomous mode can supposedly handle entire GitHub issues by itself. In practice, it works great for simple stuff and completely faceplants on anything complex. It'll write tests, update documentation, and create PRs that you still need to review because AI isn't that smart yet.

Don't expect miracles - it's like having a junior dev who never gets tired but also never really understands the business requirements. Good for boilerplate generation, terrible for architectural decisions.

Enterprise Stuff

If you work at a big company, they have all the security theater you'd expect: audit logs, content exclusion, policy management. The usual "we need to make sure the lawyers are happy" features.

Content exclusion lets you tell it to ignore certain files or directories. Useful if you don't want it learning from your database passwords or that embarrassing hack you wrote at 3am.

Extensions and MCP

You can hook it up to other tools through extensions and MCP, which is cool if you're into that sort of thing. The GitHub MCP server lets you connect it to GitHub APIs and workflows.

Most people probably won't bother, but it's there if you want to integrate your custom workflow nonsense. Eclipse IDE support means even the Java holdouts can play along - yes, Eclipse still exists. No, I don't know why either.