It’s easy to keep using GitHub Copilot the way you did on day one: accept a suggestion, press Tab, move on.
But GitHub Copilot agent mode, multi-model choice, and bring-your-own-model (BYOM) support have changed what the tool can do. The difference now comes less from the product and more from how you configure, evaluate, and use it.
Official Website: https://github.com/features/copilot
Here’s where to look.
What GitHub Copilot is now (late 2026)
- Inline suggestions and next edit suggestions as you type
- Chat that works with your workspace context
- Agent mode for bigger jobs. Visual Studio’s new Agent (Preview), for example, can be used for feature work, bug fixes and refactoring
- A model picker with a wide range of options. GitHub’s supported-models list includes models from OpenAI, Anthropic, Google, Microsoft, Moonshot AI and xAI
- Bring-your-own-model support, covered next
BYOM: easy to overlook, worth knowing
BYOM lets you run the models your team already trusts inside Copilot’s agent experience.
In Visual Studio, you can connect to a Microsoft Foundry deployment, or use your own key from another supported provider, in Agent Mode, whether or not you’re signed in to GitHub. Current providers are Microsoft Foundry, OpenAI, Anthropic, and Ollama.
It isn’t limited to the IDE:
- Copilot app: you can add a provider in Settings → Model Providers, and its models then appear in the model picker alongside Copilot-hosted models.
- Copilot CLI: BYOK works with remote services and local models like Ollama, vLLM, and Foundry Local, and GitHub authentication is optional.
This helps teams that want consistency, cost control, or private and self-hosted models. Routing through your own endpoint can also keep traffic in your own tenant for stricter data-boundary requirements.
Two cautions:
- It’s still a preview in Visual Studio. Not every model supports every Agent Mode capability yet.
- There’s a breaking change. BYOM now requires the new Agent (Preview). The previous Ask and Agent modes no longer support it, and previously configured Ollama models must be re-added.
Your own endpoint still receives your code context. Confirm that the provider meets your security and compliance requirements before you rely on it.
Where and how to use it
1. In your IDE (VS Code, Visual Studio, JetBrains, and more)
Install the official extension, sign in, and open Chat. Switch between Ask, Edit, and Agent modes, and attach files, PRs, or errors as context. Use the model picker for built-in models or to add your own. In Visual Studio, BYOM works through Agent (Preview), selected from the agent picker in Copilot Chat.
2. In the standalone GitHub Copilot app
The app is a desktop experience for starting agentic development from an issue, pull request, prompt, or previous session. It runs on Windows, macOS, and Linux, and each session gets its own git work tree, branch, and task state. That makes it a good fit for planning larger changes or working across repos. It’s still a technical preview, and Business and Enterprise plans need an admin to enable previews and the Copilot CLI.
3. In the terminal (Copilot CLI)
Use it for terminal-first workflows and scripting. Newer versions include --mode, --autopilot, and --plan flags to start directly in a specific agent mode.
Five habits that make you better with Copilot
1. Review everything. Treat Copilot like a junior pair programmer. Read each change before you accept it.
2. Use agent mode for multi-file work. Keep chat for questions and quick explanations.
3. Match the model to the task. Model choice changes latency, reasoning depth, and credit burn for the same prompt. Try a few models on the same task and note which performs best.
4. Write custom instructions. Copilot reads a repo-wide .github/copilot-instructions.md, path-specific .instructions.md files, and AGENTS.md. The best approach is to start small and iterate based on results.
5. Use BYOM where model control matters. For security-sensitive work, bring a model and endpoint your team controls.
Treat Copilot as a system you improve
Turning Copilot on is the easy part. The bigger gains come from configuring it deliberately and checking what works.
This week, add a .github/copilot-instructions.md file to your main repo with three or four team conventions. Then see how the suggestions change.
And tell us in the comments: what’s one way you’ve changed how you use Copilot in the last few months?
0 Comments