Vibe Coding Is a CEO's Dream, but Experienced Engineers Will Win the AI Race

Matt Cullerton, a software engineer with over 20 years of experience, shares his walking take on vibe coding and why he thinks it favors experienced developers over weekend hobbyists. Recorded while walking his dog, he breaks down the gap between what AI-generated code delivers and what it takes to build sustainable software products.

Every CEO's Dream Come True

The appeal of vibe coding is obvious. Every CEO Matt has ever worked with has asked the same question: "What takes those damn engineers so long?" They ask for something and hear it will take weeks. Now, tools like ChatGPT, Lovable, and Bolt let anyone type a prompt and get working software in seconds. That instant gratification is real, and the progress happening in this space is impressive.

Matt also points to the arms race on the IDE front, citing Cognition Labs' acquisition of Windsurf as a sign that AI-powered developer tools are accelerating fast. The tooling landscape is moving quickly, and the competition is heating up.

Why Knowing How to Build Still Matters

The core problem with vibe coding is that the LLM makes all the architectural decisions when you do not give it direction. When someone prompts "build this tool with X, Y, and Z features," the model chooses the language, framework, libraries, and patterns on its own. It draws from a massive corpus of knowledge spanning many years, which means it might pick libraries that are no longer maintained, or use outdated patterns like class-based React components instead of function-based ones.

A weekend vibe coder may know what they want to build, but they do not know how to build it. Matt sees that as the fundamental distinction between products that are vibe-coded on a weekend and products that have sustainability.

Experienced Engineers Get the Biggest Advantage

Matt is not scared by AI coding tools. He sees them as the next step in a long history of developer productivity gains, similar to IntelliSense and other tools that have stacked up over the years. The key insight is that people who already know how to build software will benefit the most from vibe coding. They can direct the LLM to build in smaller, well-defined chunks. They can specify the architecture, the libraries, and the patterns they want. They can guide the model to build the way they would build, just faster.

The result is that CEOs will get their wish. Building something will not be instantaneous, but it will not take two weeks either. Matt is currently building three products using Cursor and other LLM tools, and he has his team working on becoming AI-first engineers.

Key Takeaways

Vibe coding is satisfying but fragile. Typing a prompt and getting instant output feels great, but the resulting code often lacks the architectural discipline needed for long-term maintenance.
LLMs make bad architects when left unsupervised. Without direction, they pick outdated libraries, deprecated patterns, and inconsistent approaches across a codebase.
The "how" still matters. Knowing how to break work into small chunks, choose the right stack, and guide the model is what separates throwaway demos from real products.
AI tools amplify existing skill. Experienced developers use AI to move faster without sacrificing quality. The efficiency gains are real, but they go to the people who already understand the craft.
This is early. Matt expects all of this to get better over time, but the advantage will continue to favor developers who can direct the tools with intention.

Resources

Cursor - AI-powered code editor Matt uses for building products
Lovable - AI app builder mentioned as an example of instant code generation
Cognition Labs (Devin) - The AI lab that acquired Windsurf, fueling the IDE arms race