Vibe Coding Is The Future

Key Concepts

Vibe coding, AI-generated code, LLMs (Large Language Models), Product Engineer, Systems Engineer, Cursor, Windsurf, Code Debugging, Technical Assessments, Generative AI, Zero to One, One to N, Deliberate Practice, Taste, AI Coding Natives.

Vibe Coding and its Impact

The discussion centers around "Vibe coding," a term coined by Andrej Karpathy, describing a new coding paradigm where developers heavily rely on AI tools to generate code. The panelists discuss the results of a survey conducted among Y Combinator founders regarding their experiences with Vibe coding, its impact on their workflows, and the future of software engineering.

• Definition: Vibe coding involves embracing AI tools and focusing on the overall product vision rather than getting bogged down in the intricacies of code writing.
• Survey Findings:
  • Founders are experiencing significant speedups in coding, with some reporting 10x to 100x improvements.
  • The role of software engineers is shifting towards product engineering, emphasizing human taste and product understanding.
  • Founders are becoming less attached to their code, making it easier to scrap and rewrite when necessary.
  • Parallel coding using multiple AI tools is becoming common.
• Example Quotes:
  • "The role of software engineer will transition to product engineer. Human taste is now more important than ever as coding tools make everyone a 10x engineer." - Founder of Outlet
  • "I don't write code much, I just think and review." - Obby from Asra
  • "I am far less attached to my code now... since I can code three times as fast, it's easy for me to scrap and rewrite if I need to." - RB from Copycat
  • "I write everything with cursor, sometimes I even have two windows of cursor open in parallel." - Yoav from cix
  • "Coding has changed... 6 to 1 months ago 10x speed up, one month ago to now is 100x speed up... I'm no longer an engineer, I'm a product person." - Founder of a train Loop

Shifting Roles: Product vs. Systems Engineers

The discussion highlights a divergence in engineering roles, with a growing emphasis on product engineering and systems engineering.

• Product Engineers: Focus on understanding user needs, translating them into code, and iterating based on feedback. They are essentially product managers who can code.
• Systems Engineers: Focus on infrastructure, architecture, and solving complex technical problems at scale.
• Importance of User Interaction: The ability to communicate with users and incorporate their feedback is crucial for product engineers.
• Debugging: Current AI tools are not effective at debugging, requiring human intervention to identify and resolve errors.
• Embracing Rewriting: The ease of generating code with AI encourages rewriting code from scratch rather than debugging, a practice previously uncommon for human developers.

Tools and Models

The panelists discuss the popular tools and models used in Vibe coding.

• IDEs:
  • Cursor: The leading IDE, but requires users to specify the files to be considered.
  • Windsurf: A fast follower to Cursor, indexes the entire codebase, making it easier to use.
  • NotableDev: Used for small features but lacks a comprehensive understanding of the codebase.
  • Chat GPT: Used for reasoning and debugging due to its more powerful models.
• Models:
  • Claude Sonet 3.5: Still widely used.
  • Claude 1.0, 1.0 Pro, and 3.0: Gaining traction as reasoning models.
  • GPT-4: Not widely used for coding.
  • Deep Seek R1: Emerging as a viable contender.
  • Gemini: Used for its long context window, allowing developers to input the entire codebase for bug fixing.

AI Code Generation Statistics

The survey revealed a significant percentage of code being generated by AI.

• AI-Generated Code Percentage: 25% of the surveyed founders estimated that over 95% of their codebase was AI-generated.
• Impact on Non-Traditional Backgrounds: Founders with non-traditional computer science backgrounds are leveraging AI to become highly productive programmers.

The "Zero to One" vs. "One to N" Stages

The discussion distinguishes between the "zero to one" (initial product development) and "one to N" (scaling) stages of a startup.

• Zero to One: Vibe coding is highly effective for rapidly building and iterating on a product to achieve product-market fit. Speed is the primary concern.
• One to N: Scaling requires a different skillset, including systems engineering, architecture, and optimization. The initial code written using Vibe coding may not be suitable for handling large-scale usage.
• Historical Examples:
  • Facebook: Successfully scaled by hiring systems engineers and building custom tools.
  • Twitter: Faced scalability challenges due to its architecture and choice of technologies (Ruby on Rails).

Technical Assessments in the Age of AI

The panelists discuss how technical assessments for engineers need to evolve in the era of AI.

• Traditional Assessments: Whiteboarding and algorithmic problems are becoming less relevant.
• New Assessment Criteria:
  • Ability to use AI tools effectively.
  • Code review skills.
  • Debugging skills.
  • Taste and judgment in evaluating AI-generated code.
  • System design skills.
• Triplebyte's Approach: The need to adapt technical screens to account for AI proficiency.
• Importance of Deliberate Practice: While AI can lower the barrier to entry, becoming an exceptional engineer still requires deliberate practice and a deep understanding of systems.

The Role of Classical Training

The discussion explores the relevance of classical computer science training in the age of AI.

• Benefits of Classical Training:
  • Understanding fundamental concepts.
  • Ability to debug and optimize code.
  • Ability to identify and correct errors in AI-generated code.
  • Ability to design and architect complex systems.
• Developing Taste: Even without classical training, developers can develop taste through practice and exposure to good code.
• The Picasso Analogy: Just as Picasso mastered traditional painting before developing his abstract style, engineers need a solid foundation to excel in the age of AI.

Conclusion

Vibe coding is a significant shift in software development, driven by the increasing capabilities of AI tools. While it empowers developers to build products faster and more efficiently, it also necessitates a change in engineering roles and technical assessments. The ability to leverage AI effectively, combined with strong product sense, debugging skills, and a deep understanding of systems, will be crucial for success in the future of software engineering. The panelists emphasize that Vibe coding is not a fad but a fundamental change that is here to stay.