AI can write the code.
But can you explain why that code should exist?
The better AI coding tools get, the more important software design, debugging, and decision-making become.
Are AI coding tools making developers better engineers or just faster coders? 👀
🚨 Developers, honest question:
If Claude Code and Codex went down for the next 7 days…
Could you still debug, design, and ship your work?
AI makes us faster.
But if removing it makes us helpless, we’re not becoming better engineers, we’re becoming dependent. 🧠
Building a real-world app with this stack:
- React + TypeScript
- Spring Boot
- PostgreSQL
- Spring Security + JWT
- Flyway
- TanStack Query
- Zustand
- Zod + React Hook Form
- AI coding tools
Learning architecture by building, breaking, and fixing. 🚀
Building a real app taught me something tutorials didn't:
Architecture isn't about creating more folders.
It's about knowing where a change belongs.
I'm learning this the hard way while building a React + Spring Boot application.
Is Claude getting dumber day by day?
Lately, I’ve noticed it:
• More confident wrong answers
• Forgetting context
• Struggling with tasks it handled easily before
Maybe it’s just me or are you seeing it too?
Claude users, what’s your experience?
Your AI coding results are often limited by one thing: context.
Bad context -> generic solution.
Good context -> solution that actually fits your architecture.
Your README, conventions, existing code, constraints and tests matter.
Paid ₹1,704 for this SSD in Oct 2022.
Just checked the price today…
₹6,199. 😭
That’s a ₹4,495 price hike for the same SSD.
Tech prices are getting crazy.
Most developers are using AI backwards.
My workflow is:
Understand the problem
Plan the solution
Give AI the context
Let it implement
Review every important change
Test it
Refactor
The goal isn't to make AI write more code.
It's to make good engineering decisions faster.
We need to take AI safety seriously.
AI is moving incredibly fast, but are we moving too fast?
Building powerful AI is exciting, but we also need to think about the risks and responsibilities that come with it.
I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. More thoughts below.
We went from:
"Google it."
to:
"Ask ChatGPT."
to:
"Let Claude/Codex build it."
The next developer advantage won't be typing code faster.
It'll be knowing what should be built, and whether the AI built it correctly.
Coding is changing.
Engineering isn't.
AI coding tools aren't replacing developers.
They're exposing who actually understands the code.
Claude Code
Codex
Cursor
Gemini CLI
Copilot.
The tool matters less than how you use it.
Great developers know what to ask, what to reject, and what to verify.
Your React component is probably doing too much.
If one component handles:
- API calls
- form validation
- business logic
- filtering
state management
-rendering
-error handling
it's not really a component anymore.
It's an entire application wearing JSX.
Split responsibilities.
You're probably putting too much into React state.
Ask: "Does this value need to trigger a render when it changes?"
If not, it might not belong in state.
Not everything that changes needs useState.
Unpopular opinion:
Barrel files ("index.ts") files aren't just for convenience.
They're a contract.
A public API for your feature.
Once I enforced them, random cross-imports disappeared, and refactoring stopped being terrifying.
Small habit.
Big payoff.