
AI's uneven impact on delivery

Making eBay ship 10x faster – with Randy Shoup
Topic 4 of 4
Published Nov 13, 2026 · Clip 41:02–50:51 (9:50)
Randy shares findings from CircleCI's quarterly State of Software Delivery reports, based on millions of pipelines from hundreds of thousands of organizations. The top 5% of teams went from 8x to 9x to 10.5x faster than the median over three quarters, while the median and bottom 75% showed no change. He explains via the theory of constraints: AI accelerates code production, but only teams without downstream bottlenecks (smooth pipelines, modular architecture, good culture) benefit; the best teams succeed in CI after one or two tries while median teams need 3-5. His advice: use AI first to solidify the basics - re-architecting, migrations, and pipeline hardening - which LLMs excel at, before trying to accelerate feature work.