Vibe coding in 2026 — Are AI coding tools actually making us faster?

Vibe coding burst onto the scene and quickly became not just a buzzword but a new paradigm in software development. With the Collins Dictionary naming it the Word of the Year for 2025, the concept has embedded itself in our daily routines. It’s 2026, and an astounding 84% of developers now lean on AI tools like Claude Code and GitHub Copilot to streamline their coding process. Yet, a surprising study from METR suggests that, despite the hype, we might actually be coding 19% slower. This raises an important question: Are AI coding assistants boosting our productivity, or is it just a well-marketed myth?

What is vibe coding and why it went mainstream in 2026

Vibe coding, at its core, is about harnessing AI to make the development process more intuitive and efficient. It’s an approach that prioritizes flow state and leveraging AI to handle the mundane, allowing developers to focus on creative problem-solving. The mainstream adoption can be attributed to several factors:

• Increasing sophistication of AI models: The models have become remarkably good at understanding context and generating relevant code.
• Integration in popular IDEs: Seamless integration in tools we use daily made it accessible to everyone.
• Cultural shift towards efficiency: The tech industry’s relentless pursuit of doing more with less has made the promise of AI-enhanced coding irresistible.

The productivity paradox: perceived vs measured speed gains

While the narrative around vibe coding and AI tools has largely been positive, a nuanced view emerges when we dissect the METR study. Developers often report feeling faster, citing the effortless generation of code blocks and solutions. However, when measured, the speed gains are not as pronounced, and in some cases, developers are actually slower. This discrepancy likely stems from a heightened sense of progress when AI swiftly suggests code, even if the overall project timeline doesn’t shorten.

Breaking down the METR study: why experienced devs got slower

The METR study’s findings are a wake-up call. It highlighted that experienced developers, in particular, tend to be slower when using AI coding tools due to several factors:

• Over-reliance on suggestions: Developers might wait for or sift through AI suggestions instead of coding from memory or intuition.
• Increased review time: Code generated by AI requires thorough review to ensure it aligns with best practices and doesn’t introduce bugs.

Where AI coding tools genuinely excel

Despite the METR study’s findings, it’s undeniable that AI coding tools have their strengths:

• Boilerplate code: They can generate repetitive patterns and boilerplate code with ease.
• Writing tests: Generating test cases based on function signatures saves valuable time.
• Explaining code: They can provide quick explanations of complex code snippets, enhancing understanding.

The hidden costs: context switching, review burden, and skill atrophy

The drawbacks of relying too heavily on AI include:

• Context switching: Frequently interacting with AI suggestions can disrupt a developer’s flow state.
• Review burden: Every piece of AI-generated code needs to be reviewed, which adds time and cognitive load.
• Skill atrophy: There’s a risk that developers might lose touch with coding fundamentals by relying too much on AI.

A framework for effective AI-assisted development

To harness AI coding tools effectively, consider this framework:

1. Use AI for exploration: Let AI assist you in exploring solutions or generating boilerplate.
2. Manually code core logic: Engage deeply with the problem by coding the core logic yourself.
3. Review with a critical eye: Always review AI-generated code as if it came from an unfamiliar colleague.

Security concerns: CVE awareness and vulnerable dependencies

One area where AI tools need improvement is security. While they can generate functional code, they might not always be up-to-date with the latest CVEs (Common Vulnerabilities and Exposures) or might suggest using libraries with known vulnerabilities.

Finding your personal productivity balance

The key takeaway should be that AI coding tools are a supplement, not a replacement. The best approach is a balanced one, leveraging AI for speed in certain areas while maintaining and developing your coding skills in others.

Example of AI-generated code that looks right but has subtle bugs

// AI-generated code snippet - subtle bug included
function calculateDiscount(prices: number[], discount: number): number[] {
  return prices.map(price => price - discount);
}

The subtle bug here is that the function doesn’t check if the resulting price is negative, which could be a logical error depending on the context.

Prompt patterns that consistently produce better results

Effective prompts for AI tools:

• Be specific about the task and the context.
• Include examples of input and desired output.
• Ask for best practices or security considerations explicitly.

Code review checklist for AI-generated code

• Does the code handle edge cases?
• Are there any security vulnerabilities?
• Is the code efficient and maintainable?
• Does it adhere to coding standards?

In conclusion, while the allure of boosting productivity with AI coding tools is strong, it’s crucial to approach them with a balanced perspective. By understanding their strengths and limitations, we can make informed decisions about when and how to use them to truly augment our capabilities.

Until next time, happy coding 👨‍💻
– Patricio Marroquin 💜