Lesson 1 of 8 · AI for Developers: Code Faster with AI Assistants
Chat Assistants, In-Editor Assistants and Coding Agents
Understand the three kinds of AI coding tools, what each is good at, and how to combine them in a normal working day.
"Use AI for coding" can mean three quite different tools. Each has a different strength, and most productive developers use more than one. Picking the right one for the task matters more than picking the "best" model.
1. Chat assistants
Examples: ChatGPT, Claude, Gemini, Microsoft Copilot chat.
You paste code or describe a problem in a browser or desktop app and get an answer.
Good for:
- Explaining code, errors and concepts
- Designing an approach before you write anything
- Writing a self-contained function, query or script
- Reviewing a diff or a class you paste in
Limits: it only sees what you paste. It does not know the rest of your solution, your conventions or which package versions you use unless you tell it.
2. In-editor assistants
Examples: GitHub Copilot in Visual Studio or VS Code, and similar assistants built into JetBrains Rider and other IDEs.
They suggest completions as you type and usually have a chat panel that can see the open file and sometimes related files.
Good for:
- Boilerplate: DTOs, mappings, guard clauses, repetitive tests
- Finishing a line or block when the pattern is obvious
- Quick questions about the code you are looking at
Limits: completions are quick guesses. They are often right for routine code and confidently wrong for business logic. Read before you press Tab.
3. Coding agents
Examples: Claude Code, GitHub Copilot's agent mode, Cursor's agent and similar tools.
You give a task ("add pagination to the orders API with tests"), and the agent reads files across the repository, edits several of them and can run commands such as builds and tests.
Good for:
- Multi-file changes that follow existing patterns
- Running the build and tests, then fixing what fails
- Tedious but well-defined work: renames, migrations, adding a field end to end
Limits: an agent can change a lot quickly. Without clear instructions and review it can also make a lot of wrong changes quickly. Always work on a branch and review the diff.
Choosing quickly
| Task | Best fit |
|---|---|
| "Why does this throw?" | Chat |
| Write a DTO and its mapping | In-editor |
| Understand a new module | Chat or agent (it can read the repo) |
| Add a feature touching 6 files, with tests | Agent, on a branch |
| Design a caching strategy | Chat, then you decide |
| One tricky LINQ query | Chat, with the entity classes pasted in |
Ground rules before you start
- Company policy first. Many companies allow only approved tools for proprietary code. Check before pasting anything.
- Never paste secrets: connection strings, API keys, tokens or customer data (lesson 8).
- Use git for everything. Commit before an AI change so you can see and undo exactly what changed.
- Review like a pull request from a new colleague: helpful, fast, and in need of checking.
Practice task
Take one small, real task from your backlog. Do it three ways: ask a chat assistant for the code, write it with in-editor suggestions, and (if available) give it to an agent on a separate branch. Compare time taken, code quality and how much you had to fix.
Key takeaways
- Chat assistants explain and design; in-editor assistants speed up routine typing; agents make multi-file changes.
- Choose by task, not by hype.
- Follow company policy, never paste secrets, and use git for every AI change.
- Review AI code the way you would review a new colleague's pull request.
Finished this lesson?Mark it complete to track your progress on this device.