GitHub Copilot App: How to Write Your First Prompt

GitHub has published a beginner's walkthrough for its standalone Copilot app, explaining how to connect a project, write plain-English prompts, choose between AI models, use voice input, and run sessions remotely. The guide is aimed at developers who have not used the app before and are unsure what to type first. The core message: start with a small, concrete request, connect your repo or local folder, and refine from there rather than trying to write the perfect prompt upfront.

What happened

GitHub published a how-to guide for first-time users of the GitHub Copilot app, covering the key decisions you face before and during a session. The guide is light on new feature announcements but clear on how the app’s existing pieces fit together.

How a Copilot session actually works

Before the model can do anything useful, it needs context. The app lets you connect a session to a GitHub repository or a folder on your local machine. Once that link is in place, Copilot can read the relevant files and work on changes inside your codebase.

From there, you write a prompt in plain English. GitHub’s example: “Add a most-funded sort option to the games list.” That single sentence is enough to start. The model finds the relevant code, attempts the change, and you iterate from there.

What options can you change mid-session?

Three controls are worth knowing about, though none need to be configured before your first prompt:

• Model selector. Different models suit different tasks. The default model handles most work fine; switch to a more capable one when the task involves complex reasoning or the first result misses the mark.
• Agent selector. Agents can be set up for specific types of work. You pick the one that fits the task from the session title.
• Remote sessions. Instead of running everything on your local machine, a remote session keeps your work accessible from the web. Start on your laptop, close it, and resume from another device without losing your place.

Voice input: useful or gimmick?

The app includes built-in voice input. Speech converts to text in the prompt box, where you can edit it before sending. According to GitHub, this is particularly handy when you are thinking through a problem out loud or have a longer request to describe. Crucially, nothing is sent until you review and confirm the transcribed text, so speaking imprecisely does not immediately fire off a bad prompt.

Why it matters

Most of the friction around AI coding tools comes from not knowing how to start, not from the tools themselves. GitHub’s guide strips away that friction by making clear that the first prompt does not need to be comprehensive. The iterative framing is honest: you describe a goal, see what the model does, and correct it. That matches how experienced developers actually use these tools.

The remote session feature is quietly significant for teams. If you have developers working across machines or time zones, being able to hand off a session mid-task without losing context reduces a real coordination cost. Pair that with model switching and you have a setup closer to a proper agentic workflow than a simple chat interface.

For context on how Copilot has been expanding across environments, our earlier coverage of GitHub Copilot’s JetBrains update with persistent memory and Ollama support shows the broader direction GitHub is pushing.

Our take

The guide is useful, but it is essentially onboarding documentation dressed as a blog post. What it does well is set accurate expectations: prompting is iterative, you will not get it right first time, and that is fine. That is the right message for business owners or non-engineering team members dipping into Copilot for the first time.

The model selector is the feature most people will overlook and then later wish they had used. If a task involves reading across a large codebase, planning a refactor, or writing something with nuanced logic, switching to a more capable model before you start will save you correction cycles. The default is not always the right tool.

If your team is exploring how AI tools fit into your development workflow, our AI integration services cover exactly this kind of tooling assessment and setup.

What to do about it

1. Open the GitHub Copilot app and connect it to an existing repo or local folder you already know well.
2. Write one short, concrete prompt describing a real change you want, then send it without overthinking the wording.
3. Review the result, then follow up with a clarifying prompt rather than rewriting the original from scratch.
4. If the output is consistently shallow or misses context, switch to a more capable model from the model selector.
5. Try a remote session if you work across multiple devices or want to share session access with a colleague.

Pick the smallest useful task on your current project and use that as your first Copilot session. You will learn more in ten minutes of doing than in an hour of reading about it.

Source: GitHub Blog