# ia-sdlc-cutting-hype-aijournal-2025-09-15

## Veille

AI in the software development lifecycle - Quality vs speed - Systematic quality assurance - AI Journal

## Titre Article

AI in the SDLC: Cutting Through the Hype

## Date

2025-09-15

## URL

https://aijourn.com/ai-in-the-sdlc-cutting-through-the-hype/

## Keywords

AI in the SDLC, Software Development Lifecycle, Quality Assurance, Code Generation, Security Vulnerabilities, Technical Debt, Testing, Code Review

## Authors

Edgar Kussberg, AIJ Guest Post

## Ton

**Profile:** Professional-Critical | Analytical guest post | Analytical-Prescriptive | Intermediate-Expert

Kussberg adopts the tone of a quality assurance advocate, tempering enthusiasm for AI with rigorous caution. The title "Cutting Through the Hype" signals a skeptical stance. The balanced structure — promises → challenges → strategies — reveals a measured assessment. Research citations (the Stanford University study on developers introducing vulnerabilities) empirically ground the warnings. Technical language (SDLC stages, technical debt, security vulnerabilities) targets engineering leaders. Systematic prescriptions (mandatory unit tests, separate QA tools) provide actionable mitigation measures. The serious, professional tone avoids both dismissive cynicism and naive enthusiasm. Typical of tech-review guest columns introducing a quality-centered, contrarian perspective against prevailing AI optimism.

## Pense-betes

- AI transforms the SDLC, but short-term speed can sacrifice long-term value
- Evolution: code co-writing → generation of complete features by agents
- Correlation between AI adoption and declining delivery stability
- Developers accept AI suggestions without thorough review → breakdown in code ownership
- Limits: code complexity, cognitive complexity, code duplication
- Comprehensive documentation essential for architecture alignment and AI context
- Prioritizing speed harms overall quality: subtle bugs, inefficiencies, maintainability issues
- Eliminating unused code is critical (security vulnerabilities)
- AI models trained on existing code: perpetuate/amplify biases and vulnerabilities
- Stanford study: developers with AI assistants more likely to introduce vulnerabilities and judge them safe
- Excessive reliance on AI → degradation of human skills
- Strategies: mandatory unit tests, rigorous code reviews, specialized tools, third-party library checks
- Separate code generation and quality assurance using different AI tools

## RésuméDe400mots

Edgar Kussberg explores how Artificial Intelligence is radically transforming the Software Development Lifecycle (SDLC), promising efficiency and productivity gains comparable to those of the industrial revolution. However, it is crucial to distinguish hype from reality, since most developers work on complex existing codebases, where errors can have significant consequences. The question is not whether AI tools work, but whether short-term speed comes at the expense of long-term value.

AI is advancing through the various stages of the SDLC, moving from co-writing code to the generation of entire features by coding agents. This evolution makes it imperative to ensure that AI-generated code meets high standards of quality and security from the earliest stages of the development process. The growing adoption of AI, including coding assistants and autonomous agents, has shown a correlation with a decline in delivery stability, underscoring the need for guardrails to avoid compromising code stability, security, or performance.

Despite the potential for increased productivity, AI introduces management challenges. Developers feel more productive, but often accept AI suggestions without thorough review, which can lead to a breakdown in code ownership and future maintainability. To address this, teams must set clear limits on code complexity, minimize cognitive complexity, and maintain strict standards regarding code duplication. Comprehensive documentation is also essential so that AI-generated code aligns with the overall architecture and to provide relevant context to AI systems.

Prioritizing speed can harm overall code quality. AI can produce code that works in the short term while introducing subtle bugs, inefficiencies, or maintainability issues that accumulate over time. Eliminating unused code is a critical habit, as AI tools can generate superfluous references and dependencies, creating security vulnerabilities.

Furthermore, AI models, often trained on existing open-source codebases, can perpetuate or amplify biases and vulnerabilities. A study from Stanford University showed that developers using AI assistants were more likely to introduce security vulnerabilities and to judge them as safe. Excessive reliance on AI can also lead to a degradation of human skills.

To ensure quality, robust testing strategies are necessary, including mandatory unit tests independent of the code generation process. Rigorous code reviews are non-negotiable, and specialized tools are needed to identify and triage complex bugs, security vulnerabilities, and third-party library licensing issues.

In conclusion, AI is a powerful tool that amplifies human capabilities, but it does not replace human judgment and accountability. It is essential to separate code generation from its quality assurance by using different AI tools to avoid bias.

## GrapheDeConnaissance

- Edgar Kussberg —publie→ AI in the SDLC: Cutting Through the Hype (DOCUMENT, 0.98)
- Edgar Kussberg —travaille_chez→ Sonar (ORGANISATION, 0.97)
- IA —s_applique_à→ SDLC (METHODOLOGIE, 0.97)
- adoption —réduit→ stabilité livraison (CONCEPT, 0.95)
- rapport DORA 2024 —affirme_que→ une augmentation de 25% de l'adoption IA corrèle avec une baisse de 7,2% de la stabilité de livraison (AFFIRMATION, 0.95)
- assistants de code —réduit→ sécurité du code produit (CONCEPT, 0.93)
- Stanford University —publie→ étude sur assistants IA et vulnérabilités sécurité (DOCUMENT, 0.95)
- dépendance excessive —réduit→ compétences développeurs (CONCEPT, 0.9)
- code inutilisé —permet→ vulnérabilités sécurité (CONCEPT, 0.88)
- revue de code rigoureuse —réduit→ dette technique (CONCEPT, 0.85)
- Edgar Kussberg —recommande→ séparer la génération de code IA de l'assurance qualité IA via des outils différents (AFFIRMATION, 0.92)
- modèles génératifs —est_basé_sur→ bases de code open-source (CONCEPT, 0.9)
- biais et vulnérabilités —observé_dans→ modèles génératifs (TECHNOLOGIE, 0.87)
- tests unitaires obligatoires —améliore→ qualité du code généré (CONCEPT, 0.85)

---
Canonical: https://www.thekb.eu/en/fiches/ia-sdlc-cutting-hype-aijournal-2025-09-15/