# Strengths and Limits — AI Coding Assistants: Use Them Well and Review Them Hard

Source: https://www.geekswithgeeks.com/en/ai-coding-assistants/intro-strengths-limits

> List the tasks assistants handle well and the situations where they struggle.

## Good at the common, weak at the unusual

Assistants shine at **boilerplate**, well-known libraries, small functions, translating between languages, explaining code, and drafting tests. They struggle with **very new or obscure APIs**, large hidden business rules, subtle concurrency and performance issues, and anything depending on context they were not shown. When the context is missing they tend to guess instead of asking.

## A quick fit guide

Use this as a first guess, then adjust from your own experience with your tool and stack.

```text
Usually good                          Be extra careful
-----------------------------------   ---------------------------------
CRUD endpoints, forms, boilerplate    security, auth, cryptography
Unit tests for pure functions         concurrency, race conditions
Regex and SQL drafts (then verify)    performance-critical loops
Explaining unfamiliar code            very new library versions
Format / language conversions         your private business rules
```

## Ask it to say what it is unsure about

Add "list any assumptions you made and anything you are unsure about" to a request. It will not catch everything, but it often surfaces guesses about versions, missing files or unclear requirements that you can then confirm.

**Quiz:** Which task needs the most careful human review?

- [ ] Renaming a variable
- [ ] Formatting a JSON file
- [ ] Adding a code comment
- [x] Authentication and cryptography code

*Answer:* Authentication and cryptography code. Security-critical code fails silently and the cost of a mistake is high.