Developers gain productivity from AI coding tools but lack trust

A recent survey by Qodo, an AI coding quality platform, highlights the mixed feelings developers have about AI-assisted coding tools. While 82% of respondents use these tools weekly and 78% report productivity gains, a significant portion remains wary of their output.

Key Findings:

• Uneven Gains: Only a small group of "10Xers"—experienced developers—see massive productivity boosts. Most report moderate gains, while others struggle to leverage AI effectively.
• Trust Issues: 76% of developers refuse to ship AI-generated code without manual review, citing frequent hallucinations (e.g., syntax errors or nonexistent packages).
• Code Quality: About 60% said AI improved code quality, but 20% reported degradation.

The Review Bottleneck

Despite AI's potential, manual reviews undercut efficiency. However, the survey found AI excels at code reviews: 81% of users reported quality improvements when using AI for reviews, compared to 55% with manual reviews.

"Models like Gemini 2.5 Pro are excellent judges of code quality," said Itamar Friedman, Qodo CEO. Read the full report here.

Reducing Hallucinations

Friedman shared tips to minimize AI errors:

1. Context matters: Feed the AI detailed specs, examples, and coding styles.
2. Automate context augmentation: Mimic Google's search relevance tactics.
3. Start fresh: If the AI veers off track, reset rather than correct.

Top Developer Requests

1. Improved contextual understanding (26%)
2. Fewer hallucinations (24%)
3. Better code quality (15%)

Organizational Challenges

Companies must ensure AI tools comply with policies when ingesting context. As Friedman noted, "Garbage in, garbage out" applies—precision in inputs yields better outputs.

For more on AI in development, check Qodo's public benchmark.