Cursor just solved vibecoding. But it's very scary.
In a Nutshell
Cursor's new demo feature lets users evaluate AI-generated code via videos of working apps, bypassing code review entirely—admitting they ship unreviewed code. This "vibe coding" approach is dangerous and unethical, especially for critical apps like cryptocurrency wallets, as no one can verify security or safety without human review. Society must decide if we accept unreviewed AI code in production, where code review remains the essential bottleneck to prevent risks in authentication, data storage, and beyond.
These notes were generated by AI and may contain inaccuracies.
Cursor has all but admitted the rumor is true that they don't look at the code they ship. They launched a new feature that takes you away from the code and shows what the feature actually looks like.
Vibe code an app without ever looking at the code. It's a serious application like a cryptocurrency wallet. A customer asks, "Is your app trustworthy?" You've never looked at the code, and an LLM fundamentally can't tell the truth. You would be lying if you said it was safe because you don't know. Nobody knows because no one's looked at the code.
The only honest, ethical, moral answer is, "I don't know whether this application is safe or not." Therefore, it is immoral, unethical, and dishonest to ship code that no human has ever reviewed. Cursor now shows demos, not diffs.
Give it some instructions. Later it came back with a video. It uses mouse and keyboard like a human. After implementing code front and back end, it runs through testing to ensure it works and records it so you can see it works. You can watch this video in a few seconds, evaluate the work product without looking at any code, and know it has a working feature.
This is the future Cursor is pushing: code review is a thing of the past. These are questions we have to figure out.
Would love more than anyone to not review code anymore. Working on project shape, supposed to be working on it now, trying to build it for 6 months. Somewhat conscientious, conscientious enough to not want to ship code that can't review. When agents code, can't keep up—they move at a thousand miles per hour.
Cursor is building this because it's impossible to keep up with agents. People complain codeex is slow, but you want it a little slow because Opus sometimes moves so fast you can't keep up. Eventually decide whether to move fast or review code.
Every developer knows reviewing code is more difficult and takes longer than writing it. If code review is a must, LLMs are not a speed up. If humans review every line, no speed up in LLMs—the bottleneck is human review. To go fast with automated software development, won't be able to look at the code. This is the future companies like Cursor are pushing, and it's dangerous.
Be very careful and decide as a society: do we allow unreviewed code in the wild? Depends on scope—banking, finance, or photo editing—but code is code. Every application has critical security layers like authentication or storing user data. All data is sensitive. If a user asks if their data is safe and you've never looked at the code, you can't answer honestly. Can't ask Claude because LLMs can't tell the truth—they are incapable.
Love to hear your stance, open to being persuaded, but unreviewed code is immoral, unethical, and very dangerous. Thanks for watching.
Keep Mo Bitar in your library
Save the videos and channels worth coming back to, and find them again in one place.
