AI in engineering: Q2 2026 benchmarks & research readout

The episode begins with an overview of the latest AI Impact Report by DX, emphasizing its focus on how AI is altering software engineering practices across the industry. The hosts, Brian Houck and Justin Reock, discuss the motivation behind the report, especially considering the near-universal adoption of AI in development processes. They highlight that previous methods of measuring AI impact were insufficient due to near 100% adoption rates, prompting a shift in their analytical approach.

The discussion moves to AI adoption levels, noting that almost everyone is using AI in some form. The key differentiator now is how mature and sophisticated the usage is—whether developers are leveraging high-impact, complex applications of AI or just basic integrations. Justin introduces the concept of 'AI-native developers' to measure this maturity and stresses the importance of understanding the depth of AI integration in workflows.

The hosts analyze various productivity metrics influenced by AI, such as time savings, PR throughput, and deployment frequency. They observe positive trends, including increased developer time savings (about an hour per quarter) and a significant portion of merge code being AI-authored (over 50%). Despite these gains, they question how effectively these translate into real outcomes like shipping value and quality improvements, warning against over-reliance on these indicators alone.

A significant focus is given to potential drawbacks of AI in software development. These include increased PR sizes, which may impair review quality, and the rising change failure rates, suggesting more volatility and potential quality issues. They discuss how bigger PRs and reduced reviewer engagement could contribute to technical debt and lower confidence in code changes, raising alarms about long-term maintainability.

The conversation explores how perceptions of code maintainability are evolving. While AI potentially makes code more maintainable technically, developers report decreased confidence in that code, reflecting a possible shift in the sense of ownership and trust. Justin mentions ongoing research into how AI impacts various dimensions of code quality, suggesting that traditional standards may need reevaluation.

Data shows that larger companies perceive an increase in perceived code quality, likely due to structured rollouts and policies. However, skepticism remains about whether this reflects actual improvements or organizational bias, as larger codebases and dependencies tend to decrease clarity and trust independently.

The episode examines how AI influences change failure rates and system volatility. Although failure rates remain high, the amplitude of fluctuations has increased, indicating higher volatility. The hosts warn that AI can amplify both good and bad outcomes, stressing the need for comprehensive, multi-metric assessments to understand true system health.

The DXI, a composite measure of various developer experience drivers, shows a slight decline, contrary to its usual upward trend. Drivers like documentation, incremental delivery, and overall confidence are highlighted, with concerns that increased AI usage might be diluting some aspects of quality and ownership. Justin emphasizes the importance of redefining core concepts like maintainability in the context of AI-driven development.

Finally, the speakers discuss the increasing costs associated with AI adoption—spending has risen dramatically—yet this is not clearly translating into increased innovation or value delivery. The episode underscores the importance of mapping value streams and addressing organizational inefficiencies such as unnecessary meetings, which can negate the benefits of AI investments. They suggest that focusing on process improvements alongside AI deployment is critical for realizing true gains.