Vibe Coding Is The WORST IDEA Of 2025

Summary

In this video, Dave Farley critiques the concept of 'vibe coding' introduced by Andre Karpathy. He argues that while AI-assisted programming may seem appealing, it misses the core principles of effective software development. Farley emphasizes that true programming is about problem decomposition, clear communication, and precise instructions. He highlights the importance of modular design, automated testing, and continuous delivery for maintaining quality in software. AI-generated code, he warns, presents challenges in maintaining precision, trust, and the ability to evolve systems incrementally, making automated testing even more crucial.

Takeaways

• 😀 Vibe coding, introduced by Andre Karpathy, refers to a vague and imprecise approach to programming with AI assistance that lacks structure and precision.
• 😀 Programming languages are tools designed to help organize thoughts, structure problem-solving, and communicate ideas clearly, rather than simply writing code.
• 😀 Writing code is not the hard part of programming; understanding the problem, structuring the solution, and translating it to machine instructions are the main challenges.
• 😀 Programming isn't about knowing specific languages or frameworks but about using the right tools and having a solid understanding of design and problem-solving.
• 😀 Good programming requires an understanding of modularity, abstraction, cohesion, and separation of concerns to manage complexity effectively.
• 😀 The imprecision of natural language, such as in AI programming, complicates specifying and verifying what a system should do.
• 😀 AI-generated code often ignores the process of making incremental changes and correcting errors over time, which leads to problems in software evolution.
• 😀 Automated testing and continuous integration are crucial for ensuring that AI-generated code remains reliable and functional as changes are made.
• 😀 The real issue with AI programming is the lack of version control, which makes it hard to ensure consistency and accuracy when regenerating code from scratch.
• 😀 Vibe coding fails to address critical issues in software engineering, including maintaining code quality, ensuring reliable changes, and enabling continuous improvement.
• 😀 Software engineering is evolving, and any new approach to programming, including AI-driven tools, must address the need for clear specifications, automated testing, and incremental development to be successful.

Q & A

What is 'vibe coding' and who introduced the term?

'Vibe coding' is a term introduced by Andre Karpathy, a former head of AI at Tesla and one of OpenAI's founders. It refers to a style of programming with AI assistance, where coding becomes more intuitive and less focused on traditional syntax, often characterized by quick, trial-and-error methods like copy-pasting, running code, and using natural language prompts.

Why does Dave Farley criticize the idea of 'vibe coding'?

Dave Farley criticizes vibe coding because he sees it as a shallow and naive approach to software development. He believes it overlooks the need for precision and structured thinking in programming, which are essential for building reliable and maintainable systems. He suggests that vibe coding fails to address key challenges in software development, like understanding the problem in enough detail and ensuring that systems are easy to modify over time.

What are the three main goals of programming languages according to Dave Farley?

According to Dave Farley, the three main goals of programming languages are: 1) to help organize our thinking about a problem, 2) to communicate our understanding to other humans, and 3) to specify instructions for a computer to follow. These goals aim to make software development more structured and effective.

What does Dave Farley believe is the hardest part of programming?

Dave Farley believes the hardest part of programming is understanding the problem in enough detail that a solution can be effectively crafted. This requires breaking down the problem and organizing it in a way that ensures the solution will be both effective and maintainable.

How does Dave Farley view the role of programming languages in software development?

Dave Farley views programming languages as tools that help structure our thinking and allow us to decompose complex problems into manageable steps. He believes that programming languages are not inherently difficult but are designed to provide precision and clarity, which are crucial for developing reliable software.

Why does Farley believe AI-generated code might lead to issues?

Farley suggests that AI-generated code could lead to problems because large language models (LLMs) might not have a strong understanding of what constitutes good code. LLMs are trained on a wide range of code, including bad and ugly examples, and might not create efficient or high-quality solutions. This lack of discernment could result in poor programming practices.

What is the 'vagueness and imprecision of natural language' issue in programming?

The vagueness and imprecision of natural language refer to the challenges of translating human intentions into a programming language. Natural language is often ambiguous, and this can cause misunderstandings when trying to define system behaviors. Farley highlights this as a major challenge in programming, especially when using AI to assist in coding.

What does Farley suggest as the solution for improving AI-driven programming?

Farley suggests that one solution to improving AI-driven programming is the use of executable specifications. This means clearly defining the system’s behavior in precise, repeatable terms so that both humans and AI can verify that the system works as expected. This approach emphasizes clarity and accuracy in specifying the system’s intended outcomes.

What is the role of automated testing in AI-driven programming according to Farley?

Automated testing plays a crucial role in AI-driven programming because it ensures that AI-generated code continues to work as intended after every change. Since AI may regenerate code from scratch, it’s vital to test the system rigorously to confirm that no unintended side effects have occurred and that the code remains functional.

How does Farley view the evolution of software engineering with AI assistance?

Farley believes that AI is transforming software engineering, but not necessarily in the way some might expect. While AI can assist with coding, it still faces significant challenges, particularly in maintaining code quality and supporting incremental development. Farley argues that to be truly effective, AI programming tools must address issues of code precision, testing, and maintainability.