3 myths about coding with AI that even experienced developers still believe

There's little doubt that AI has augmented countless tasks for developers, but its rapid adoption has also created a number of assumptions about what AI-assisted programming can, and perhaps more importantly, cannot do. Some of these myths are easy to dismiss, while the others.

Here are some myths about AI in coding which are in need of some serious myth-busting. AI coding assistants like Claude Code, Cursor, and IDEs like Google Antigravity have made it easier than ever to offload the menial and mechanical parts of software development.

What Happened

The prevalent assumption is that, if AI can handle much of the code writing, developers will have more time to focus on parts of their work that require their higher order thinking skills. And sure enough, most of it sounds reasonable.

• In a longitudinal study of professional software engineers, the research found that 82% spent less time writing code, yet their work had broadly shifted from creation toward verifying, evaluating, and correcting AI output.

• Although 84% of surveyed engineers continued to report that AI improved their productivity, the share reporting a worsened developer experience in at least one dimension nearly doubled, going from 14% up to 27%.

• A September 2025 research evaluating GitHub Copilot's Code Review feature seems to suggest that those eyes may not be looking in the right places.

Key Details

After all, frontier models are now faster and better at writing code, and with these models aiding the development process, it should naturally translate into greater productivity. It was this very assumption that recent research from the University of Auckland challenged.

• If anything, the research seems to suggest that AI is changing software development in more complex ways than the prevailing narrative surrounding the technology implies.

• That, however, doesn't mean that every perceived benefit survives scrutiny.

• AI-augmented coding seems to have clearly moved beyond the prevalent assumptions surrounding it, and there's no doubt about it seeing as how the industry has been moving at a breakneck pace.

Why It Matters

The study suggests that the relationship between AI assistance and productivity is more complicated than it appears. The decline concentrated in flow state and cognitive load, both of which suggest that writing less code does not necessarily mean doing less work.

• It does, however, evidence that an AI agent can sustain an autonomous software engineering process over a task that the researchers estimate would take a human weeks to complete.

• Instead, the benchmark reported token usage metrics and the tests passed.

What Reports Say

Coverage of the story so far points to:

• Continued reporting by XDA as more details emerge