From typing code to steering autocomplete.

The first useful layer was not autonomous yet. Copilot lived inside VS Code, completed local code, and made small edits faster, but the human still carried context, planning, testing, and handoff discipline.

VS CodeGitHub CopilotInline completionsHuman-led testing

Before

Manual coding with search/docs as separate tools.

After

Assistant in the editor, still bounded by the current file and developer loop.

Anatomy

Input

Existing editor workflow, local repository, developer intent.

Shift

AI moved from documentation lookup to inline implementation help.

Output

Faster snippets, local edits, and the first habit of asking an assistant before doing repetitive code work.

Limit

Context was shallow. It did not own the task, route work, remember procedures, or verify end-to-end outcomes.

Why this mattered

This created the appetite for a CLI agent that could inspect a repo, run commands, and carry a broader task.