Recent conversations about AI coding tools reveal an unsettling trend: the gap between vibe coding and agentic engineering is narrowing. What was once thought to be a clear distinction now seems less defined, raising questions about responsibility and trust in AI-generated code. This shift has implications for developers and organizations alike.

Understanding Vibe Coding and Agentic Engineering

Vibe coding is typically seen as a hands-off approach where users ask for a result without engaging with the underlying code. Often, such users may not even know how to program. They rely on AI tools to generate code, test it, and move on, trusting the output if it works. The focus is on getting a quick result, with little concern for code quality or long-term maintainability.

In contrast, agentic engineering involves professional programmers who use AI as a supportive tool. These developers understand the importance of security, performance, and maintainability. They leverage AI to enhance their work but still review and oversee the code, ensuring it meets high standards. The goal is to build robust, high-quality systems efficiently.

The Convergence and Its Discontents

Initially, these approaches seemed distinct—vibe coding was casual and personal, while agentic engineering was responsible and professional. However, recent experiences suggest that the lines are starting to blur. Developers find themselves trusting AI-generated code more readily, even for critical production tasks, without thorough reviews.

This shift is partly driven by the increasing reliability of AI tools. For example, asking an AI to create a JSON API or handle database queries often results in code that works perfectly the first time. Automated tests, documentation, and performance are often accurate, reducing the need for human oversight. But this creates a new challenge: how much responsibility does a developer have when they don’t review the code?

There is a concern about the normalization of trust. When AI consistently produces correct results, developers may start to rely on it blindly. Over time, this can lead to a false sense of security, increasing the risk of bugs or vulnerabilities slipping through, especially if the AI is not perfect.

Rethinking How We Evaluate AI-Generated Software

In the past, developers could gauge the quality of a project by its documentation, commit history, and testing coverage. A well-maintained GitHub repo with many commits and thorough tests was a sign of care and expertise. But now, it’s possible to generate similar-looking repositories quickly, making it harder to judge quality at a glance.

This raises a new question: what truly indicates a trustworthy AI-produced codebase? The answer may lie in usage rather than documentation. Developers value software that they have actually used daily. Practical experience with a tool or system provides a better sense of its reliability than just reading its code or tests.

As AI tools become more integrated into development workflows, the emphasis shifts from solely reviewing code to assessing how well the software performs in real-world use. Trusting AI-generated code without human oversight can be risky, especially when the code is part of critical infrastructure or sensitive data.

Ultimately, the line between vibe coding and agentic engineering is becoming less clear. While AI can boost productivity and improve code quality, it also challenges traditional notions of accountability. Developers and organizations need to rethink how they evaluate and trust AI-generated software, balancing efficiency with responsibility. The future may require new standards for oversight and trust in AI-assisted development.

Inspired by

• https://simonwillison.net/2026/May/6/vibe-coding-and-agentic-engineering/#atom-everything

Sources

• fandf.co
• heavybit.com
• github.com