Is AI Creating Technical Debt Faster Than It Can Fix It?

As an IT leader, keeping up on technical debt for hundreds of systems has been nearly impossible. It’s rarely a priority for your business partners, it’s expensive and time-consuming, and, in some cases, it has questionable value. Then AI came along with the idea that you can refactor and modernize an app or your tech stack, quickly and efficiently. But is AI creating more technical debt faster than it’s fixing it?

GenAI: To the rescue, but also the detriment

In the past year, I’ve had many conversations with friends and colleagues about the promise that GenAI will finally address decades of technical debt. Quickly and efficiently refactoring an application – especially a homegrown app – is a real possibility now.

One fellow IT leader shared that her organization’s sales unit recently spent $50k on tokens and took a 16-bit application written for her sales team in 2003, running on a Windows 2000 server environment, and transformed it into a modern, cloud-native architecture. This effort proved you don’t need reams of documentation nor even the original development teams who wrote the app, and the new life is proving useful for their business at a fraction of the cost of moving to another app.

But now the app has sprung to life, and so have various challenges. The code generated by AI has introduced change volumes that they aren’t prepared or staffed to handle. Not all the production code has a clear owner, and the app is about 100,000 lines of code larger than the previous app, she said.

The costs of software development are changing quickly. Code used to be pretty expensive to produce, with many developers spending a lot of time, infrastructure engineers building out the core technology stack in a data center, and business personnel spending a lot of time on requirements. Now, it’s inexpensive to produce that code, but the ongoing cost of owning and maintaining that software is going up.

And since software is easy to produce, citizen developer teams are churning out new apps all the time. But they need to be integrated, tested, security scanned and pass all the same rigor you expect from your IT team. Token budgets are all the rage, but nobody’s giving you more money to support these apps. This is the origin of the new technical debt.

You are not alone

If this is also you, well, you have company. One March 2026 study examining more than 300,000 verified AI-authored commits found that over 15% introduced at least one code-quality, security or reliability issue. Roughly one-quarter of the identified issues were still present in the latest version of the repo, so the issues are piling up as technical debt.

Another study from the Association for Computing Machinery (ACM) found that AI agent-generated changes were more likely to miss opportunities for reuse and introduce redundant code, even though reviewers often responded positively to the submissions. The code looked ok because the AI passed all the diff reviews, linting and automated tests on the surface, while maintainability problems accumulated underneath it.

One final data point: I’ve been building my own application on the side for fun, which is 100% written by multiple AI’s working in concert. Every week when I do my security scans, the AI finds new issues. I asked ChatGPT (5.6 Sol) to analyze the codebase for technical debt and it rated it a 5/10 with a variety of architecture, dependency, and code sprawl issues. And this application is only 4 weeks old. Technical debt is a living and breathing thing, it seems.

What’s a CIO to do?

So if I’m right and this is a vicious cycle that never ends, what can you do to at least shrink the blast radius of that technical debt circle? Here are a few basic ideas that CIO’s should be focusing on with their team:

1) Lead with architecture – Ensure your EA and SA teams have an established, scalable architecture that you’ll hold as the foundation for any technical debt remediations. Make sure you have solid patterns for things like authentication, authorization, data access and observability, to name a few. Every feature should reference those patterns.

2) Control the test plane – AI is great at linting and executing dozens or even hundreds of automated tests in seconds to validate its own work. But some of those tests are surface level or incomplete. It’s not a volume play, but an accuracy play. Make sure your team has a critical eye on the outcomes of those tests and that they don’t grow significantly, quickly.

3) Ensure clear ownership – Every critical system, module and integration needs a clearly defined human owner. That’s both a role (e.g., Solution Architect) and, oftentimes, a primary and secondary name. “Claude” or “GLM5.3” can’t be your ownership model.

4) Set a complexity budget – Put limits around things like file size, dependency counts and database migration complexity. The human owners are accountable for ensuring they’re proactively keeping an eye on this. Exceptions should be well-documented and approved via your governance structure.

5) Fund maintenance explicitly – Nobody likes working on Keep The Lights On (KTLO) activities, but you must reserve capacity for refactoring, dependency upgrades and security remediation. That can’t compete for funding and resources with the new feature and functionality that everyone would prefer to work on.

We need to ask different questions

AI clearly has a ton of value and will continue to change the game when it comes to time and effort related to generating new, valuable software. But you can’t let this come at the expense of more technical debt that accumulates faster. 

CIOs and CTOs need to ask different questions, such as “How much AI-generated software has entered the environment? Who owns it? Can the team explain it? Does it follow the company’s architecture and security standards? What will it cost to operate for the next five years? And for every new application being created, is anything being retired?”

Take steps to control the AI technical debt now, before it controls you.

‍

Todd Mazza is the former CTO of Factory Mutual Insurance Company. He’s also held leadership roles at Rockwell Automation, Workday, AECOM, Levi Strauss & Co., MGM Mirage, and NBCUniversal. Contact Todd: LinkedIn | @ToddMazzaCIO on X