[Paper Review] 2025 State of AI Code Quality
I have just reviewed an interesting report from Qodo: “2025 State of AI Code Quality” that surveyed 609 developers on how AI tools are being used.
The bottom line: AI is mainstream, but trust is still the biggest blocker to realizing its promised efficiency gains. I identify, at least, 3 critical themes that can maximize AI workflow value:
- Context is the Foundation of Trust
The primary complaint about current AI tools isn’t how much code they generate, but how relevant it is.
- The context gap is huge: 65% of developers using AI for refactoring and approximately 60% using it for testing, writing, or reviewing report that the assistant “misses relevant context”.
- The top fix: “Improved contextual understanding” is the #1 requested fix (26% of all votes), rising to about 30% when “customization to team standards” is included.
- Actionable insight: AI must act like a teammate who knows the codebase. A learned, repo-wide context engine is necessary for accuracy, quality, and trust. Manually selecting context is broken; 54% of developers who do this still AI misses relevance.
2. Confidence Drives Adoption (The Hallucination Hurdle)
If AI output isn’t accurate, adoption stalls and engineers waste time reviewing everything.
- The confidence metric: Developers who experience fewer than 20% hallucinations are 2.5x more likely to merge code without reviewing it (24% vs 9% of others).
- The red zone: A massive 76% of developers fall into the “high hallucinations, low confidence” group. This means they use AI, but don’t trust the results, leading to manual review, delays, and limited ROI.
- Confidence and morale: High-confidence engineers are 1.3x more likely to say AI makes their job more enjoyable (46% vs. 35% of those with low confidence).
3. Automated Review is the Quality Multiplier
Speed alone doesn’t guarantee quality; automated review converts raw velocity into durable code quality.
- Productivity & quality synergy: When teams report “considerable” productivity gains, 70% also report better code quality — a 3.5x jump over stagnant teams.
- The AI review benefit: With AI review integrated, quality improvements soar to 81% for fast teams (compared to 55% for equally fast teams without review).
- Even without a speed boost, teams using AI review see 2x the quality gains (36% vs 17%). This continuous, opinionated review is the force-multiplier we need.
In summary, AI tools are generating a quarter or more of our code. To close the trust gap and truly transform our development process, we must focus less on raw speed and more on AI tools that provide deep, automated contextual awareness and robust, continuous quality review. This creates a “confidence flywheel” that reinforces accuracy, quality, and trust.