Anti-Patterns - What Doesn't Work

Learn from our mistakes.

Anti-Pattern 1: Blind Acceptance

Mistake: "AI is smart, I'll just accept everything."

Reality: AI makes plausible mistakes.

Example from real project:

// AI generated
async function deleteUser(userId: string) {
  await db.users.delete({ id: userId });
  // Looks fine, ships to production
}

// Problem: No cascading deletes
// User deleted, but their orders, preferences, sessions remain
// Database integrity violated

Lesson: Always ask "What could go wrong?" before merging.

Anti-Pattern 2: Over-Engineering

Mistake: Asking AI to build "scalable, production-ready, enterprise-grade" solution for simple problem.

What happens:

• AI generates 500 lines of abstractions
• Multiple design patterns stacked
• Harder to maintain than simple solution

Example:

Bad prompt: "Create enterprise-grade user service with repository pattern,
CQRS, event sourcing, and microservices architecture"

For: A simple CRUD app with 100 users

Good prompt: "Create simple UserService with:
- findById, create, update, delete
- Basic validation
- Follow patterns from @src/orders/OrderService.ts"

Lesson: Start simple. Add complexity when needed, not preemptively.

Anti-Pattern 3: Skipping the Understanding Phase

Mistake: Generate code → copy/paste → commit → merge

Why it fails:

• Can't debug when it breaks
• Can't explain to teammates
• Can't modify when requirements change

Real case from Client A: Developer used AI to implement payment processing. Code worked in development. Failed in production due to timezone handling. Developer couldn't debug because didn't understand the code.

Better approach:

You: "I need to implement payment processing. First, explain:
1. What are the key security considerations?
2. What error cases should I handle?
3. What testing strategy do you recommend?

Then we'll implement together, step by step."

Lesson: AI is a collaborator, not a replacement for thinking.

Anti-Pattern 4: Context Overload

Mistake: Dumping entire codebase into context.

Example:

"Here's all 50 files in @src/, analyze everything and fix all bugs"

Why it fails:

• AI gets confused with too much context
• Misses important details
• Suggestions are generic
• Slow and expensive

Better approach:

"Focus on @src/orders/OrderService.ts

Specific issue: calculateTotal() returns wrong amount when discount applied

Related files:
@src/orders/models/Order.ts
@src/promotions/DiscountCalculator.ts

Help me debug."

Lesson: Be surgical with context. Quality over quantity.

Anti-Pattern 5: Using AI for Everything

Things AI does poorly:

• Architectural decisions (use AI for analysis, human for decision)
• Security threat modeling (AI helps, humans must verify)
• Business logic validation (AI doesn't know your business)
• Political/interpersonal code review (use human judgment)

Example:

Good:

"Should we use microservices or monolith for [specific requirements]?
Analyze trade-offs."

[AI provides analysis]

[Human makes decision based on business context AI doesn't have]

Bad:

"Should we use microservices or monolith? Decide for us."

[AI picks one]

[Team commits without considering business context]

Lesson: AI augments decisions, doesn't make them.