State of AI Impact in Engineering: Q2 Report

Engineering organizations have entered a critical transition phase, shifting their focus from measuring baseline AI adoption to evaluating concrete return on investment. With AI utilization now above 90% across the industry, the question engineering leaders face is no longer “are we using AI?” but “is it paying off?”

This quarter’s data shows a system in tension: code output is climbing, cost is climbing faster, and the human experience of building software has not kept pace with either. While AI tools are accelerating individual output, the surge in AI-generated code is testing the limits of existing infrastructure and human review gates.

This report measures 500+ organizations against the DX Core 4 (Speed, Effectiveness, Quality, Impact) and the DX AI Measurement Framework (Utilization, Impact, Cost). Four distinct themes have arisen from the data this quarter:

1. Speed gains are real but uneven

• Median weekly TrueThroughput rose 37% over four quarters (1.42 to 1.94 PRs/eng/week), and deployment frequency is up across most segments.
• However, gains are concentrated among small organizations (less than 100 engineers) and tech-sector teams, pulling away from larger and traditional-industry teams, and the gap is widening.

2. AI is improving some aspects of developers’ work while creating new bottlenecks in others

• Documentation quality, code maintainability, and production debugging all improved, and represent the clearest, least-contested wins in this report.
• Incremental delivery, local iteration speed, and review turnaround all declined, and the average Developer Experience Index (DXI) slipped from 67 to 65. PR size nearly doubled over the same period, consistent with AI-driven inflation rather than disciplined, testable code.
• AI is making individual tasks faster, but the surrounding system is absorbing the slack rather than converting it into new value. This is the same “AI efficiency paradox” now being documented industry-wide: individual output surges while human gates become the new bottleneck.

3. Code is easier to read but harder to trust

• Code maintainability improved, but change confidence fell in the same period. These metrics are historically correlated, but are now in tension. AI can help engineers understand and change the code in front of them, but faith in the output and trust in the code is slipping.
• Change failure rate volatility increased, with many organizations now reaching +/-3 percentage points against a 4% industry benchmark. AI has not created this volatility, but it has amplified it.
• Perceived software quality tells a counterintuitive story: Traditional industries and Financial Services, the slowest-moving segments on throughput, report the highest perceived quality. Speed and perceived quality are not moving together, and leaders should not assume they will.

4. Spend is outrunning the return it’s meant to justify

• Median quarterly AI spend grew from ~$1.5K to ~$44K in a year, roughly a 28x increase in the Tech sector alone, while the innovation ratio (time spent on creating new features vs. maintenance) has stayed essentially flat over the same four quarters.
• Time savings are real and growing, now over 6 hours a week on average, but saved hours are not yet visibly converting into new value creation at the portfolio level.
• Smaller organizations pay more per seat for AI tools, since they lack the negotiating leverage larger companies get from volume licensing. Yet they’re getting more output per dollar spent than larger companies. This challenges the common assumption that scaling up AI spend at the enterprise level will automatically improve ROI. The data suggests the opposite: bigger AI budgets don’t guarantee better returns, and may even come with diminishing ones.

The bottom line for leaders

AI has accelerated code-writing, but is not yet translating into faster felt delivery, higher innovation capacity, or spend that has proven its return. The organizations best positioned going into the remainder of the year are treating AI adoption as the starting line, not the finish line. Leaders should pair every throughput or spend metric with a quality and experience counterweight (change confidence, DXI, innovation ratio, etc), and invest as much in the surrounding environment and platform as in the AI tooling itself.