Theory
Most AI coding tools share the same core surface area: inline suggestions, a chat panel, and increasingly an agent mode that can run multi-step tasks. This topic uses GitHub Copilot as the reference, but the patterns apply across tools.
Inline suggestions
As you type, the assistant predicts what comes next and shows it in grayed text. Press Tab to accept. This works best for:
- Boilerplate and repetitive patterns
- Completing function bodies when the signature and context are clear
- Writing test cases based on existing ones
The quality improves significantly when the surrounding code is clean, well-named, and follows consistent patterns. The model uses what's already there as a signal.
Chat
The chat panel lets you ask questions and get longer-form responses. Good uses:
- "What does this function do?"
- "How can I improve the readability of this?"
- "Explain this error message."
- "What tests should I write for this component?"
Keep questions focused. A specific question gets a more useful answer than a vague one.
Agent mode
Agent mode (also called Copilot coding agent, or just "agentic mode" depending on the tool) lets the assistant take a goal and execute multiple steps autonomously: reading files, making edits, running commands, and iterating. This is covered in more depth in the cloud agents topic later today.
The key shift
The most important thing to internalize early is that these tools work best when you give them clear context. A well-described task, a file with good naming conventions, and an explicit goal will produce dramatically better results than typing a vague prompt and hoping for the best.
We use Copilot as the reference example here because it is widely available and well documented. But the same basic concepts apply if you use Cursor, Claude Code, or any other tool. Find the equivalent docs for whatever tool your team has adopted.
Practice
Run this in a repository you already know, not a toy project. The point is to feel where the practice helps and where it gets in the way on code that has history.
Work through the GitHub Copilot quickstart guide linked above from start to finish. As you go, note:
- What is the first useful suggestion you accept? What made the context clear enough for the tool to get it right?
- Open the chat panel and ask three different questions about a file you know well. For each answer, assess: was it accurate, and was it useful?
- Try an example where the suggestion is wrong or unhelpful. What was missing from the context?
- If you use a different tool (Cursor, Claude Code, etc.) rather than Copilot, find the equivalent getting-started docs for that tool and work through those instead. The goal is the same: get your first real suggestion, use chat, and reflect on what context helps.