CI/CD speed run: Diagnosing and fixing pipeline bottlenecks

Your CI is slow. The bottlenecks were always there - you just cared less when humans were the rate limiter. Now AI-assisted development is pushing more code through CI than ever, PR volume is up, queue depth is up, and the slow build you tolerated for two years is suddenly blocking the whole org from shipping.
We'll start with a synthetic repo that looks uncomfortably like what you've inherited - 3-tier app, slow tests, messy CI, Kubernetes deploys via Argo CD - and walk the diagnostic process: map the flow, find the worst bottleneck, fix it, move on. We'll hit the fixes that actually move the needle (Docker caching done right, the QEMU trap, sharding vs parallelism, commit-to-pod deploys) and skip the ones that just feel productive.
You'll leave with a fast pipeline, a process for finding the next bottleneck, and the receipts to justify the work back home.