AI Spend Up ~28x, Innovation Ratio Up One Point: The Q2 Numbers Every Eng Lead Should Bring to Their Next Planning Meeting

TL;DR: DX's Q2 2026 benchmarks (Aug 14): ~95% developer AI adoption, over half of merged code AI-authored, PRs grown 42→72 lines — while AI spend rose ~28x and innovation ratio barely moved.

DX's Q2 2026 engineering benchmarks readout (published August 14) is the most useful adoption snapshot a lead can steal numbers from right now. Adoption is effectively saturated — ~95% measured through telemetry, so "should we adopt" is over as a question. More than half of merged code is now AI-authored. Average PR size grew from 42 to 72 lines. And the uncomfortable pair: code maintainability is improving while developers' change confidence is declining — two metrics that historically moved together. Meanwhile AI expenditure hit roughly 28x year-ago levels against an innovation ratio up about one percentage point.

This lines up with the spring DORA 2026 report ("The ROI of AI-assisted Software Development"), whose core finding was that returns show up in the delivery flow — review, integration, deployment — not in individual typing speed, and that AI amplifies whatever organizational strengths or dysfunctions you already have.

The playbook implication for a team that's behind: don't buy seats and measure acceptance rates. Instrument the flow (review latency, PR size, change confidence, rework), expect the J-curve dip, and put your effort where both reports point — the review and integration bottleneck that 72-line AI-assisted PRs are now pressing on.

Sources: DX Q2 2026 benchmarks readout, DX AI Impact Report Q1 2026, InfoQ on DORA 2026