The widespread integration of code generation assistants and artificial intelligence tools within digital transformation projects is profoundly changing performance indicators. While the initial promise was to accelerate the development phase, feedback from the field highlights a major side effect: the emergence of automated technical debt. This phenomenon is now severely impacting large ERP projects, where system standardization and stability are absolute priorities.
The illusion of speed and the proliferation of duplicated code
The widespread availability of automated code generation tools has led to an unprecedented increase in the volume of code produced. However, the speed of generation now far exceeds the human capacity to test, review, and validate changes securely.
Overall code quality analyses demonstrate a measurable degradation of logical structures. According to research conducted by the platform GitClearThe volume of copied and pasted code increased by 48%. For the first time, the volume of duplicated lines exceeded that of intelligently reused code. Furthermore, refactoring operations decreased by 60%. AI produces functional code in the short term but does not improve the overall application architecture.
This observation is reinforced by the overall empirical study entitledDebt Behind the AI Boom: A Large-Scale Empirical Study of AI-Generated Code in the Wild, available on arXiv:2603.28592Analysis of over 300,000 automatically generated commits reveals that 89.3% of introduced anomalies are design flaws (code smells), such as overly broad exception handling or unused variables. Even more concerning, 22.7% of this technical debt persists and becomes deeply entrenched in final production versions.
Team behavior in response to generated code, the concept of GIST debt
The impact of automation is not only technical, but also methodological and behavioral. A second university study, available on arXiv:2601.07786introduces the concept of debtGIST (GenAI-Induced Self-admitted Technical debt, or AI-induced self-admitted technical debt).
The research highlights a specific behavior: the integration of automatic code into production repositories accompanied by explicit comments of doubt written by the developers themselves (e.g.,// TODO: AI-generated code, no idea of its actual relevanceThis phenomenon gives rise to two major risks.
- The knowledge gap,A gap is created between the code produced by the machine and the team's mental model, which leads to postponing the validation phases.
- The dilution of responsibility,Uncertainty about the reliability of the code leads to implicitly delegating verification to future revisions, weakening the actual ownership of the system.
The critical case of ERP projects, the drift of Clean Core
In the field of integrated management software packages, adherence to the principle ofClean CoreMaintaining a standardized core system to enable automatic cloud updates has become the norm. Specific extensions and customizations must be deployed to the edge.
The use of AI assistants to quickly generate extensions or migration scripts leads to hyper-fragmentation of this edge. A technical report published by Lightrun indicates that 43% of AI-generated code blocks require immediate manual debugging in production, and that the majority of organizations face multiple redeployment cycles to stabilize these changes. Details of these hidden costs are analyzed in publications by STEP Software on AI-related technical debt.
At the global governance level, the market is currently experiencing a period of significant tension. The benchmark studyERP Solutions LandscapeForrester Research describes the current period as a "modernization supercycle." Large-scale cloud implementations are creating significant budgetary pressures and increased risks to deliverables, as summarized in the analysis document available on HouseBlend - Forrester Wave ERPAmong the most frequent causes of failure or budget overruns, there is a glaring lack of business process mapping prior to contract signing, combined with a shortage of specialized skills.
The necessary "Data Detox" before the migration
A modern ERP system driven by predictive algorithms or integrated co-pilots can only function if the source data is flawless. Attempting to migrate decades of inconsistent or redundant historical data represents a major technical debt that skews learning models.
The current methodological recommendation is based on the principle of decoupling. By archiving historical data in separate, legally protected silos and migrating only strictly operational data to the new ERP system, the database footprint is minimized. Research published by JiVS and reviewed by ERP Today experts demonstrates that this strategy can reduce IT costs related to storage and maintenance by up to 80%. Case studies on this topic are documented in the archives of ERP Today - Data Migrationas well as on the News Center JiVS.
In conclusion, we need to rethink the indicators of success.
Managing a project based solely on raw velocity (the number of closed tickets or lines of code produced) is a major methodological error. Generation speed should not be confused with secure delivery speed.
The role of project governance is evolving towards the implementation of strict quality barriers, the imposition of scoping and data cleansing phases lasting several weeks before any configuration, and the safeguarding of the functional scope. In the face of automation, maintaining human understanding of the system remains the primary factor for success.