AI Won't Fix Broken Systems (www.aviator.co)

A recent DORA report finds AI adoption in software engineering is now nearly universal, with roughly 90% of teams using AI and over 80% saying it increased their productivity — but those gains are often perceived rather than realized. Studies (e.g., METR) even show developers feeling 20% faster while objective measures found a 19% slowdown. Significantly, faster individual coding doesn’t automatically translate to better delivery: flaky CI/CD pipelines, poor testing, fragmented toolchains and weak governance mean AI can amplify existing chaos, not fix it. The report also flags rising delivery instability and little evidence that AI reduces friction or burnout. Technically, the report identifies seven capabilities that predict positive AI impact: a clear AI strategy, healthy data ecosystem, strong version control, small-batch workflows, user-centered design, high-quality internal platforms, and tight team-system alignment. It also defines “AI engineering waste” — prompt latency, context loss, toolchain fragmentation and heavy validation overhead — which erode gains unless addressed. The takeaway for engineering leaders: treat AI adoption as a systems-design problem. Redesign workflows to match new throughput, invest in platforms and VSM, tighten governance and version-control practices, and fix cultural and process bottlenecks so AI becomes a multiplier of real, sustainable value.