Are We Overusing AI? When Classic Software Works Better

Are we overusing AI in software engineering?

AI is everywhere today — from architecture diagrams to internal tools and chatbots. But not every problem actually needs an LLM layer.

In this video, I share a practical engineering perspective on when AI makes sense — and when classic software works better. Inspired by Andrej Karpathy’s breakdown of Software 1.0, 2.0, and 3.0, I walk through real-world examples from backend and production systems to explain why mixing deterministic and probabilistic systems incorrectly can increase complexity instead of reducing it.

You’ll learn:

- How Software 1.0 (deterministic) and Software 3.0 (probabilistic) differ in practice
- Why adding AI everywhere can hurt reliability and predictability
- Real examples where AI is overused (and simpler solutions exist)
- Real examples where AI actually adds value (monitoring, exploration, triage)
- A simple mental model to decide: AI vs classic code

This video isn’t anti-AI — it’s about using the right tool for the right job, especially in backend and production systems.

📌 For engineers, architects, and anyone building systems at scale.

👉 Let me know in the comments: Where have you seen AI being overused — or used really well?

#aiengineering #softwarearchitecture #backendengineering #llm #systemdesign #productthinking

🔗 Resources
- Software is changing (again)
https://www.youtube.com/watch?v=LCEmiRjPEtQ
- How to determine unused index in PostgreSQL
https://youtu.be/YMhk6DniOWQ

---

⌚️ Timestamps:

0:00 Introduction — Are We Overusing AI?
0:57 My Perspective: Where AI Helps vs Where It Hurts
2:07 Andrej Karpathy’s View: Software 1.0, 2.0, and 3.0
2:36 Deterministic vs Probabilistic Systems (and Why It Matters)
4:16 Example #1: Chatbots and Hidden Product Gaps
6:33 Example #2: Unused Indexes — Do We Really Need an LLM Layer?
7:40 When AI Makes Sense: Monitoring Workflow Before AI
9:20 When AI Makes Sense: Monitoring Workflow After AI
10:57 AI and Repetitive Work — What’s the Real Signal?
11:40 Key Takeaways & Final Thoughts

Видео Are We Overusing AI? When Classic Software Works Better канала Anas Anjaria