Is AI Accelerating a Hidden Software Crisis? The Rise of Duplicate Code

AI Boosts Productivity

Developers often hear the claim, "AI won't replace developers, but AI‑using developers will replace those who don't," signaling a clear choice: adopt AI or risk obsolescence. Proponents argue AI can free engineers from low‑value, repetitive tasks.

The latest Stack Overflow Developer Survey shows that over 60% of developers already use AI in their daily work, with many planning to increase usage. While productivity gains are the primary motivation, the GitClear report warns that chasing efficiency blindly is like removing a bicycle chain to pedal faster—ultimately counterproductive.

Software change velocity is accelerating; by 2025 the rate of code changes is projected to be nearly twice the pre‑AI level of 2021.

Refactoring Is Declining

Before obsessing over productivity, we must not neglect foundational practices such as refactoring. Continuous optimization of code structure—maintaining low coupling and single‑purpose components—is essential for building reliable systems.

Classic software design books by Kent Beck, Martin Fowler, Robert C. Martin, and others embody timeless wisdom that AI assistants cannot overturn.

GitClear tracks refactoring through "code moves" (core‑logic preservation with design improvements) and classic patterns like "extract method" or "rename variable." Since 2021, refactoring’s share of code changes has plummeted from 24% to under 10%, while copy‑paste duplication has risen from 10% to nearly 15%.

Projections indicate that by 2025 refactoring will account for only 3% of code changes, risking a technical‑debt avalanche that could jeopardize system stability.

Late Lessons

The author recounts a past experience where introducing continuous delivery, automated testing, deployment pipelines, and robust monitoring dramatically improved reliability and rebuilt trust between development and business teams.

After leaving, the team removed database scripts used to reset test data, deeming them cumbersome. For months the integration tests continued to pass—until the test data became polluted, forcing a painful decision between rebuilding the scripts or discarding failing test cases. Under delivery pressure, they chose the latter.

Deleting tests does not cause immediate failure; functionality appears to work for a while. However, hidden defects accumulate, eventually manifesting as chronic, hard‑to‑fix issues—a classic trade‑off of short‑term efficiency over long‑term sustainability.

Don’t Lose Out

AI code assistants create an illusion of a perfect, parabolic development speed curve—like a zero‑gravity aircraft giving a fleeting sense of weightlessness while actually free‑falling.

To let AI truly sustain productivity, we must not let it dominate code‑quality decisions. Accelerated development must be matched with equally vigorous system‑health maintenance; otherwise today’s efficiency frenzy becomes tomorrow’s technical disaster. The essence of software development is not who writes code fastest, but who builds systems that endure the test of time.