Why AI Coding Boosts Some Developers Tenfold but Leaves Others Behind

According to a survey by AI coding company Qodo, developers appreciate productivity gains from AI coding tools but do not fully trust their output.

The study of 609 developers across various organizations found that 82% use AI tools weekly and 78% say they improve efficiency, yet confidence gaps reduce some benefits.

Qodo’s CEO Itamar Friedman notes that while AI coding delivers net gains, the distribution is uneven: a small group of “10‑x programmers” reap huge benefits, most see moderate gains, and some fall behind.

The survey shows 60% of developers report AI improves or slightly improves code quality, while about 20% say it worsens it. However, 76% will not ship AI‑suggested code without manual review, preferring to rewrite or verify suggestions.

Developers who use AI for code review report higher quality improvements (81%) compared to manual reviewers (55%). Friedman adds that models like Gemini 2.5 Pro can assess code quality more accurately than traditional metrics.

About Trust

Around three‑quarters of respondents frequently encounter “hallucinations” such as syntax errors or calls to non‑existent packages; only about a quarter report rare hallucinations.

Friedman suggests mitigating hallucinations by prompting the AI to check repository structure and documentation before assigning tasks, providing clear specifications, and using the AI to generate tests that must pass before execution.

Developers prioritize improving contextual understanding (26%) over reducing hallucinations (24%) and enhancing code quality (15%). Providing detailed context to the model—product requirements, specs, examples, and coding style—significantly impacts output quality.

Organizations must ensure any content fed into AI models complies with company policies.