Michael Zuercher is CEO and co-founder of Prismatic. With over two decades in tech, he is an entrepreneur behind many startups.getty​AI is helping developers generate code faster than ever before, but it doesn’t eliminate the constraints of software development. Instead, it shifts them downstream, creating bottlenecks in testing, validation and deployment. When these processes don’t keep pace, the result is faster accumulation of technical debt, not faster software delivery.​Fortunately, while AI can add to technical debt if not used properly, it can also help development teams address existing technical debt. ​Technical debt—the accumulation of shortcuts and outdated decisions in a codebase—often persists because the perceived cost of fixing it is too high. Historically, that assumption has been valid, but today, AI is challenging it. ​AI Changes The Cost Of RefactoringI recently spoke with the CEO of another company who said AI was fundamentally changing their refactoring plans. A project they initially estimated would require 10 people over the course of two years is now being reconsidered as something three engineers could complete in six months. ​I’ve also spoken to an investor who said that several of their large, late-stage portfolio companies are considering restructuring their engineering teams and allocating budget to hire AI-first consultants who would handle a product rewrite, while a smaller, more focused in-house team maintains the current platform. Whether or not this will be effective is yet to be determined, at least from my perspective, but these examples show that AI is changing not just how code is written, but how companies think about their technical debt. ​How Legacy Companies Are Using AI To RefactorAI can help companies of all sizes refactor their code—whether they’re a new company or a legacy enterprise with a complex, long-standing codebase that needs reworking. GitHub, for example, has discussed how it used Copilot to rethink its internal tooling for helping customers with large-scale migrations. Instead of building custom applications for each new customer, the team refactored the application into a modular system that could be reused and extended for new customers. ​Amazon has also spoken about internally leveraging its CodeGuru tool, which uses AI to find and fix inefficiencies in Java or Python code. Back in 2020, one of the teams that worked with enterprise e-book publishers found that they were increasingly experiencing performance issues in the system they used to analyze publishing data. When dealing with especially large datasets, the system began to struggle, resulting in increasing failures and timeouts. ​The developers didn’t find any obvious issues when investigating the codebase themselves, but when they checked it with CodeGuru, they discovered that a legacy transform was using over 75% of CPU time, and they were able to go in and address that. ​Refactoring With The Benefit of HindsightThese examples showcase how refactoring isn’t just about cleaning up code, but also about revisiting earlier decisions in light of new scale, requirements and capabilities. AI-assisted refactoring tools allow business leaders to take a step back and ask a question that was previously too expensive to entertain: If we were building this today, would we do it the same way? ​The team at GitHub might not have realized how much of a hassle their old method of helping with migrations would become as their customer base grew. Similarly, Amazon’s data transform wasn’t giving them problems when they set it up, but as they started working with larger datasets, issues started occurring. ​With the benefit of hindsight, these companies were able to revamp their processes based on what they learned was and wasn’t working. ​The same thinking can be applied to address integration technical debt as well. Often, companies build integrations incrementally, solving for immediate needs rather than planning for the long term. Over time, this can lead to hard-to-maintain integration layers. ​AI-assisted refactoring can be used to revisit those systems, standardize the way integrations are designed and move toward a more modular approach. ​When To Use AI-Assisted Refactoring​Before undertaking a major AI refactoring project, make sure that what you want to rewrite is actually an ideal candidate for it. AI is most effective when:• The intent is well-defined.• The work is pattern-driven.• Results can be easily validated.​It can also help development teams explore and understand legacy systems, making it easier to identify issues before making changes. ​It is less effective when: ​• The problems require excellent judgment rather than just execution.• Requirements are ambiguous.• The underlying issue is product or design-related​.In these situations, AI can assist, but it shouldn’t replace human engineering judgment. ​What AI Doesn't Do: Eliminate Refactoring RiskIt’s also important to proceed with caution when considering this sort of project. The history of software development is filled with failed rewrites. AI speeds up refactoring, but it doesn’t reduce the risks of rewrites or architectural complexity.​For example, a company may begin a major rewrite but struggle to achieve feature parity with the current platform, which can result in customer losses. ​Ensuring the best outcome for any rewrite requires strong test coverage and clear criteria for measuring success—and these are especially crucial when AI is in the picture. ​Where AI Creates Real ValueUltimately, AI creates opportunity, but only if it’s applied to the right problems and managed responsibly. ​Used thoughtfully, AI-assisted refactoring can unlock projects that once felt out of reach. If you’ve been putting off a major refactoring because of time or resource constraints, it may be worth revisiting those assumptions in light of these new capabilities. ​The key is to make sure you’re solving the right problems, not just solving them faster. Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?