If you already cannot manage your technical debt on traditional software, do not even try playing with GenAI. It will not save your information system. On the contrary, it will literally sign your digital death warrant.
The numbers are brutal: 95% of GenAI projects fail. And they do not fail cleanly. They leave behind junk code, ghost applications, and gaping security holes. Now imagine that inside an organization already struggling to maintain its current ecosystem.
Technical debt: the silent poison
Before talking about generative AI, let us talk about what most executives prefer to ignore: the real state of their information system.
Technical debt is the accumulation of technical compromises made to move fast. Code shortcuts. Postponed updates. Tests never written. Obsolete architectures kept on life support. Every shortcut taken five years ago has become a lock that slows down every new initiative.
In the field, I see the same pattern in nearly every organization I work with. Teams spending 60 to 80% of their time maintaining existing systems rather than innovating. Deployments taking weeks instead of hours. Recurring production incidents whose root cause no one can find because the codebase has become a labyrinth.
And it is in this context that organizations want to plug in generative AI. You do not put a nuclear reactor in a wooden shed.
Why GenAI accelerates the fall
Generative AI has a property that many underestimate: it amplifies the state of the system it runs on. If your codebase is clean, well-tested, well-documented, GenAI will save you time. If your codebase is a minefield, GenAI will produce code that integrates perfectly... into the chaos.
Here is what I concretely see in the field when organizations deploy GenAI on top of unmanaged technical debt:
Unmaintainable code proliferation. Developers use Copilot or ChatGPT to generate code at high speed. But that code builds on existing patterns — including the bad ones. Technical debt multiplies at the speed of autocomplete. Within three months, the codebase has grown by 40% and maintainability has dropped by 60%.
Ghost applications. GenAI POCs are launched across every department. Nobody governs them. Nobody secures them. Six months later, you have twenty applications calling paid APIs, storing sensitive data without encryption, and with no technical owner.
Cascading security vulnerabilities. GenAI generates code that works. But works does not mean secure. SQL injection, hardcoded secrets, unauthenticated endpoints — generated code reproduces the anti-patterns in your existing codebase. With a terrifying acceleration factor.
Hallucinations on business data. Connecting an LLM to a poorly structured database full of duplicates, empty fields, and inconsistencies guarantees wrong answers delivered with perfect confidence. AI does not fix your data. It amplifies it.
The prerequisite nobody wants to hear
Before deploying generative AI, you need to clean house. Not because it is elegant. Because it is a matter of survival.
Step 1: Technical debt audit. Measure it. Not by gut feeling. With tools (SonarQube, CodeClimate, manual reviews). Identify the critical zones: where is the debt that blocks, where is the debt that can wait.
Step 2: Targeted remediation plan. You will not pay back everything. Target the areas where you want to deploy AI. Clean those zones first: tests, documentation, architecture, security.
Step 3: GenAI governance. Before letting anyone deploy a GenAI POC, establish clear rules. Who approves use cases? Who reviews generated code? Where is data stored? What is the acceptable monthly cost?
Step 4: Quality pipeline. Every line of AI-generated code must go through the same pipeline as human code: code review, automated tests, static analysis, security scans. No free pass because it came from AI.
Generative AI is an amplifier, not a savior
The message I hammer home with every executive I work with is simple: GenAI is not a magic wand that compensates for years of technical neglect. It is an amplifier. If your organization is healthy, it amplifies performance. If your organization is sick, it amplifies the disease.
The 5% of GenAI projects that succeed share one thing: they are deployed on solid technical foundations, with clear governance and trained teams. Not on a dying information system that someone hoped to save with a chatbot.
The question is not should you do GenAI. The answer is yes, obviously. The question is: are you ready? And if the honest answer is no, then priority number one is not a GenAI POC. It is paying down your technical debt.
The organizations that will win the AI race will not be those that deployed fastest. They will be those that deployed on the strongest foundations.