ai-mistakes

47 AI Coding Mistake Patterns

A comprehensive catalog of mistakes that AI coding assistants frequently make. Organized by severity.

How to Use This Guide

1. During code review: Check changed code against each applicable pattern
2. Before claiming completion: Verify none of these patterns are present in your work
3. When debugging: Check if the bug matches a known pattern
4. When planning: Design your approach to avoid these patterns from the start

---

CRITICAL (12 patterns) — Must fix immediately. These cause production failures or security holes.

#1 False Completion Claim

Symptom: Saying "done" or "complete" without running verification commands. Prevention: ALWAYS run tests, build, and lint BEFORE claiming done. Show the output. Rule: No completion claim without fresh command output as evidence.

#2 Phantom Code Reference

Symptom: Importing or calling a function/class that doesn't exist. Prevention: grep -rn "function_name" . before using any reference. Check imports resolve. Rule: Every import must resolve to an actual file/module. Every function call must have a definition.

#3 Silent Error Swallowing

Symptom: except: pass, catch(e) {}, ignoring error return values. Prevention: Every error handler must log, re-raise, or explicitly handle the error condition. Rule: No bare except, no empty catch blocks, no ignored errors.

#4 Test That Tests Nothing

Symptom: assert True, assert response is not None, expect(true).toBe(true). Prevention: Every test must assert specific behavior — values, side effects, or state changes. Rule: Each test function must have at least one meaningful assertion about business logic.

#5 Abandoned Test Code

Symptom: Test files created but not included in test runner, or test code left in production. Prevention: Run the test suite and verify your test file appears in the output. Rule: Every test file must be discovered and run by the test runner.