𝗔𝗜 𝗖𝗮𝗻𝘁 𝗪𝗿𝗶𝘁𝗲 𝗦𝗮𝗳𝗲 𝗖𝗼𝗱𝗲 𝗬𝗲𝘁
Bjarne Stroustrup created C++. He has a strong opinion on AI coding.
He says AI tools fail at writing complex system code. He believes AI creates more bugs and security holes.
Here is why he worries:
- Tracking changes is hard. Human changes stay in one place. AI changes spread across a whole project. You cannot easily find the side effects.
- AI imitates old habits. LLMs train on existing code. They repeat old mistakes and slow performance patterns.
- Validation is a mess. It is hard to verify what an AI wrote.
He does not hate AI. He thinks it helps with technical writing. You still need a human to check the work.
Stroustrup focuses on the top 20% of code. This is the code used for safety and high performance. AI is not ready for this work.
Experts from PVS-Studio agree.
Phillip Khandeliants says problems start when developers copy and paste AI code without checking it. This leads to bad and insecure software.
Oleg Lisiy adds a different view. AI is part of work now. You should not let an AI agent push code directly to production. That is a bad idea for big companies.
He also notes that using AI is like using autocomplete in your editor. You cannot just hit tab and hope for the best.
The goal is to use AI as a tool, not a replacement for thinking.
Optional learning community: https://t.me/GyaanSetuAi