# How They Work — AI Coding Assistants: Use Them Well and Review Them Hard

Source: https://www.geekswithgeeks.com/en/ai-coding-assistants/intro-how-work

> Explain that assistants predict plausible code from context and do not understand or run it by themselves.

## Plausible, not proven

A coding assistant is a **large language model** trained on lots of code and text. Given the context you provide (open files, your prompt, project notes) it predicts the most likely helpful continuation. That makes it fast and often right, but it does not run the code in its head, and it can produce something that **looks correct and is not**. Agents reduce this by running tests, but only when tests exist.

## Confidence is not correctness

The tone of an answer says nothing about its accuracy. A wrong answer is written just as fluently as a right one, so trust comes from tests, documentation and reading, not from how sure the assistant sounds.

**Quiz:** Why can an assistant give code that looks right but is wrong?

- [x] It predicts plausible code and does not verify it by running it
- [ ] It always copies from the internet
- [ ] It refuses to use tests
- [ ] It cannot read code

*Answer:* It predicts plausible code and does not verify it by running it. Plausibility is what it optimises; correctness has to be checked separately.