Only The developer types code or a comment → the IDE extension sends context to the cloud service → the model returns multiple candidate suggestions → the IDE shows them → the developer chooses or edits one is correct.

This sequence reflects how an IDE integrated assistant operates in practice. The extension observes what you are typing and sends a carefully selected context from the current file and nearby code to a managed backend. The hosted model generates several candidates which the extension displays inline or in a panel. The developer remains in control and can accept one, cycle through alternatives, or modify the text before inserting it.

The assistant downloads and caches the whole project locally → an offline model generates completions → changes are committed automatically without review is incorrect because mainstream assistants rely on a managed service for inference and do not auto commit code. They keep the developer in the loop to review and accept changes.

Developer enters a prompt → the assistant scrapes public repositories on example.com → a single suggestion is created → the change is committed automatically is incorrect because assistants do not scrape public repositories on demand to craft a single suggestion and they do not commit changes without developer approval. They use the current editor and repository context and present suggestions for manual acceptance.

The developer saves the file → Cloud Build runs a pipeline → Vertex AI creates code updates → the updates are pushed back to the repository is incorrect because that describes a continuous integration workflow and not an interactive in editor suggestion loop. Build pipelines create artifacts or run tests and they do not push unsolicited code updates back to the repository during normal editing.

Full AWS Practitioner Certification Question