Chapter C99 · Interactive Course
Editor and cloud coding agents
Pending changes, assistance modes, briefs for detached runs, review surfaces and choosing by review fit.
Core Learning Concept
Choose between editor and cloud workflows using the task’s inputs and review needs. Confirm where execution occurs, what state is available, and how the result will be inspected. Compare the same small task rather than ranking products by novelty. Record the practical reason for the choice.
What You’ll Learn in This Chapter
-
An agent inside your editor
Recognise editor-hosted agent work and where its changes appear.
An editor agent operates inside your development interface, connecting file navigation and diff reviews directly. Context and authority still depend on selected modes. Always inspect actual edits rather than assuming proximity ensures correctness.
-
Copilot has several different modes
Distinguish completion, chat and delegated cloud work.
Copilot includes multiple distinct modes: inline ghost text, interactive editor sidecars, and delegated cloud pull requests. Each uses different context and execution models. Identify the active mode before comparing capabilities or following instructions.
-
A cloud agent needs a written issue
Supply the context a detached run cannot ask for.
A cloud agent cannot rely on a conversation’s unstated local context. A written issue should include the problem, reproduction, acceptance criteria, repository state, and relevant constraints. Provide authorized environment setup separately. The resulting proposal still needs review against those requirements before it is integrated.
-
The review surface changes what you catch
Compare reviewing a diff in an editor and in a pull request.
A review surface determines how you inspect a result: inline editor changes, a diff, command output, or a pull request. Each makes some evidence easier to see. Choose a workflow that exposes the details your task needs and open supporting artifacts when the summary view is insufficient.
-
The same skills transfer
Apply one task contract across two different products.
The review habits transfer across products: define the task, confirm context, inspect actions, examine changes, and verify the outcome. Configuration and interfaces differ, so check the host’s rules. Learn the shared reasoning process without assuming that identical commands or filenames work in every tool.
-
Choose by review fit, not novelty
Pick the surface that matches how you will check the work.
Choose an interface by practical fit: needed files, execution environment, collaboration, review, and recovery. A newer product is not automatically better for your task. Compare a representative small job under clear criteria, including total review effort, and record what evidence would make you change the choice.
-
Apply it: Editor and cloud coding agents
Place editor and cloud agents against a terminal agent, and start one bounded task in each by the review it needs.