In To vibe or not to vibe, Birgitta Böckeler explores the factors that influence the decision to trust—or verify—code generated by LLMs.
A frequently reported behavior among software developers is the introduction of technical debt when working under pressure. This often stems from the need to prioritize immediate functionality to meet tight deadlines, which ultimately compromises overall code quality.
Despite the promised productivity gains—at least in the initial implementation of features—LLMs are unlikely to reduce developer stress. Instead, project managers may respond to increased efficiency by promising more features and agreeing to even tighter deadlines with customers.
Consequently, developers may feel pressured to trust LLM-generated code without sufficient scrutiny, as thorough human review can delay delivery. Indeed, it is increasingly reported that code review pipelines have become the primary bottleneck in the development lifecycle.
The article proposes a standard risk assessment framework to evaluate the probability of flawed code generation, its potential impact on the product, and its overall detectability. While the responsibility for weighing these variables rests with the developer, high-pressure environments often bias human judgment. Consequently, the reliance on subjective assessment under stress represents a significant systemic risk in LLM-driven development.
From a pedagogical standpoint, rigorous review is non-negotiable. Yet, acknowledging the practical constraints of the industry, we must view this as a significant open problem—one that both the academic and practitioner communities will have to address.