Why Linus Torvalds, Bjarne Stroustrup, and Other Veteran Programmers Are Skeptical of AI‑Written Code

He argues that AI may change how programmers work, but it does not change the fundamental realities of software development—such as understanding systems, designing architecture, and making judgment calls that automated systems struggle to handle.

Why the productivity gains may be smaller than advertised

Java performance expert and author Vlad Mihalcea has emphasized a different point: even when AI generates working code, the net productivity gains may be modest once the full development process is considered.

Writing code is only one part of software engineering. Developers still must:

• verify correctness
• debug edge cases
• integrate code into larger systems
• ensure maintainability and security

If AI output is "almost right," engineers still spend time validating and fixing it. Mihalcea argues that after accounting for this extra work, the real productivity improvements may be marginal.

Industry surveys support this concern. In the 2025 Stack Overflow developer survey, 66% of developers said they spend more time fixing AI‑generated code that is nearly correct but not quite, and "almost‑right" answers were the most frequently reported frustration.

Some research has even found experienced developers completing tasks about 19% slower when using AI tools, despite believing they were working faster.

The corporate narrative: AI engineers and automation

While many veteran programmers emphasize limitations, technology executives often highlight AI’s potential to transform software development.

Several companies are heavily investing in AI coding tools and automation systems. Meta CEO Mark Zuckerberg has suggested that AI could soon perform work comparable to a mid‑level software engineer, potentially writing significant portions of code inside the company.

More broadly, leaders across the industry have predicted that AI might eventually generate most software. Some executives have suggested AI could write the majority of code within a few years, fueling speculation about automation of junior developer roles.

Amazon is also investing heavily in AI development and infrastructure, with CEO Andy Jassy arguing that artificial intelligence will reinvent many digital experiences and reshape how companies build technology products.

The real disagreement: what “productivity” means

The debate is less about whether AI helps developers and more about how its impact should be measured.

Technology companies often focus on output speed: how quickly tools can generate code or help small teams ship features.

Veteran engineers, however, evaluate productivity across the full lifecycle of software:

• system design
• correctness and reliability
• security
• debugging and testing
• long‑term maintenance

When these factors are included, rapid code generation alone may not translate into equally large productivity gains.

The likely middle ground

Most experienced developers do not argue that AI is useless. Instead, they see it as a powerful assistant—especially for boilerplate code, prototypes, and documentation.

But they caution that software engineering remains far more than typing code. Architecture, judgment, debugging, and system understanding remain deeply human tasks.

For now, the emerging consensus among veteran programmers is that AI is changing how developers work—but not eliminating the need for experienced engineers anytime soon.