How AI Is Quietly Reshaping Software Development: A Veteran Engineer’s Perspective

Artificial intelligence is rapidly changing the way software is built, tested, and maintained, but not by simply replacing developers with robots; instead, AI is becoming a subtle, pervasive part of the engineering toolkit.

The author, originally trained in chemistry and physics, switched to programming in the 1990s, learning languages such as Visual Basic, Delphi, C/C++, and C# on Windows. Early career learning relied on face‑to‑face courses, books, newsgroups, peer interaction, and hands‑on coding.

In recent years the author has observed AI tools appear in daily workflows: code assistants, test‑generation utilities, infrastructure management, and even project planning. These tools are now embedded in modern IDEs, making it hard for developers to avoid them.

Developers are grouped into three categories based on AI usage:

Light or non‑users (the “Luddites”), who risk missing a transformative wave.

Moderate users (the “Lucky”), who review AI output for quality while gaining productivity.

Heavy or blind users (the “Crazy”), who may create maintenance and security hazards by over‑relying on AI.

The author sees himself between the first two groups, using AI sparingly but thoughtfully.

Concerns are raised about IDE‑integrated AI: the tools are confident, often providing plausible but incomplete answers, which can discourage deep problem‑solving and creative solution design. Additionally, the concentration of AI power in a few large companies raises privacy and copyright issues.

**Example 1 – Registry lookup in Delphi**: The author asked Copilot how to read the PDF‑file‑association program from the Windows registry. Copilot suggested the path HKCU\Software\Classes\.pdf, but the code would not compile in Delphi because details were missing. The full registry hierarchy is:

HKEY_CLASSES_ROOT (combined view)
  HKEY_CURRENT_USER\Software\Classes (user‑specific)
  HKEY_LOCAL_MACHINE\Software\Classes (system‑wide)

AI failed to mention that the association can be overridden by user preferences (e.g., the “Open with” setting in Windows Explorer) and gave no guidance for C#.

**Example 2 – Suppressing WebBrowser warnings in C#**: The author asked how to silence script‑error dialogs when loading a URI in a C# WebBrowser control. Copilot correctly pointed to the ScriptErrorsSuppressed property, demonstrating AI’s strength on straightforward API questions.

A conversation with an experienced test manager revealed that AI often requires extensive prompt engineering (“prompt engineering” or “query crafting”) to produce usable unit‑test code, and integrating such code into large codebases may demand a complete rewrite of the prompts, making AI a potential waste of time for seasoned testers.

When considering AI‑generated code for new projects, copy‑pasting snippets can be effective, but for large, complex, long‑lived codebases the advice is to proceed with caution: AI suggestions are frequently incomplete or outright wrong.

In summary, the author’s experience with AI is mixed. Simple, direct queries can yield helpful answers that integrate smoothly, yet relying on AI for substantial, maintainable code introduces significant risk. Independent thinking and manual problem‑solving remain essential to avoid over‑dependence on tools that may disappear tomorrow.