The article "Exit Vibe Coding, Enter Vibe Reviewing!", written by Alexandre Mogère, Chapter Lead at Carrefour France's Software Factory, shares his experience experimenting with AI agents to automate code audits, noting a significant reduction in time (cut by a factor of 3) but also difficulties encountered. He insists that AI is not a "magic wand" and warns against the misleading claims of "AI evangelists" lacking practical experience. The central idea: apply a rigorous approach, similar to "Vibe Coding" (where AI generates the code), but to the code review process — dubbed "Vibe Reviewing".
Iterative learning journey
Mogère details an iterative learning journey, moving from initial disillusionment to the development of a functional methodology. He explains that "Vibe Coding" often oversimplifies the complexities of shipping a production-ready application, which requires quality, maintainability, security, performance, and visual consistency. While "Vibe Coding" can boost productivity with human validation, an application entirely generated by AI without strict human oversight is not viable in the long run. This realization led him to explore a rigorous AI-assisted approach for code reviews.
Multi-agent methodological evolution
The article describes several rounds of experimentation, from an initial two-phase process leading to AI "hallucinations" to a more refined multi-agent cross-validation system. He discovered that simplifying the audit plan could backfire, and that a systematic, methodical approach was more effective. The key innovation was using a static site generator (VitePress) to turn markdown audit reports into interactive documentation, with search, inline editing, and progress tracking. This approach "gamified" the audit experience and made the results more accessible and useful.
Documented agent as arbiter
He also explored using a documented agent as an arbiter in disagreements between human reviewers, leveraging the agent's ability to research and provide evidence-backed recommendations. The latest iteration focused on standardizing the full process with templates and pre-instructions ensuring reproducibility and easing adoption by teams.
Effectiveness and limitations
The author concludes that the iterative method, with technical safeguards and a focus on interactive documentation, is highly effective. He notes limitations, however, such as the need for human technical expertise to catch AI errors and the agent's lack of business context. The article insists: the audit report should not be a static document but a living tool that evolves into a migration roadmap, guiding backlog management, milestone definition, and even onboarding new developers.
Structural transformation
This methodology represents a transformation in how code quality is assessed and maintained, moving from reactive manual reviews to a proactive, systematized, AI-augmented approach. Multi-agent cross-validation is particularly innovative, creating checks and balances that prevent a single AI's potential hallucinations from becoming accepted truth. Transforming audit reports via a static site generator addresses the common problem of documentation obsolescence, turning dry reports into engaging, searchable, updatable resources.
Pragmatic AI integration
The final message is pragmatic: AI is powerful for accelerating certain aspects of development and review, but requires thoughtful integration, human oversight, and a systematic methodology to produce real value. Positioning "Vibe Reviewing" as the professional counterpart to the more casual "Vibe Coding" reflects organizations' growing maturity in their approach to AI-assisted development workflows.