AI Coding Assistant — AI Glossary

Reference details

Related terms

• Code Completion
• Code Generation
• Coding Agent

How it works

1. The assistant reads editor state as context: open files, the current selection, recent edits, and often terminal output and version-control history.
2. Requests are routed by shape — inline completion to a small fast model, an explain-or-refactor request to a stronger one — because the latency budgets differ by an order of magnitude.
3. For repository-scale questions it retrieves relevant files rather than sending everything, since context is finite and indiscriminate stuffing degrades the answer.
4. Edits are proposed as reviewable diffs rather than applied silently, keeping accept and reject the developer's decision.
5. Where the product offers an agent mode, the same context plumbing drives a multi-step loop — the difference is who holds control, not what the tool can see.

Example: Context access is the differentiator, not the model

Two assistants call the same model. One sends the current file; the other sends the current file, the interface it implements, and the failing test. Asked to fix the test, the first proposes a change consistent with the file and inconsistent with the interface, because it could not see the interface. The second proposes one that satisfies both. Identical model, identical prompt template, different context — which is why assistants built on the same frontier model produce visibly different results, and why multi-file context is the feature worth comparing.

Commonly confused with

Treated as interchangeable with its own completion feature. Completion is one surface; a current assistant also does explanation, refactoring, test generation, and in agent mode a multi-step edit loop — and these are evaluated differently, so a product that completes well can refactor poorly. Against a coding agent the line is control: an assistant keeps the developer in the loop per suggestion, while an agent runs many steps before returning. Most products now ship both modes behind one name, which is why 'is it any good' needs to be asked per mode rather than per tool.

When to use it

Reach for it when:

• Routine implementation where the developer stays in the loop and reviews each change.
• Working in unfamiliar code, where explanation of existing behaviour is the actual need.
• Mechanical refactors that are tedious to type and easy to verify by eye.

Reach for something else when:

• As a substitute for understanding code you are accountable for.
• Where the editor context cannot reach what the change depends on, such as behaviour defined in another service.
• Multi-step work with a machine-checkable goal, which agent mode handles better than suggestion-by-suggestion.

Continue your research

Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to AI Coding Assistant — AI Glossary.

• Browse models
• Review providers
• Inspect benchmarks
• Explore workflows
• Learn AI concepts
• Read decision guides
• Review trust and methodology
• Start a decision path