"Typing is not the bottleneck" – illustrated (blog.robbowley.net)

“Typing is not the bottleneck” — a Jira cumulative flow chart from a real team illustrates why: only about 30% of work time sits in the actual value-creation stage (Development); the other 70% is eaten by manual, non-value activities like code reviews, feature/regression testing, and idle queues waiting for review, test, or release. That pattern is common, so boosting developer output—whether by hiring more engineers or using GenAI to write code faster—just pumps more work into downstream stages, increasing waiting, testing load, and release risk, and often slowing overall delivery rather than speeding it. For the AI/ML community this has a clear implication: GenAI-assisted coding alone won’t deliver productivity gains unless teams eliminate downstream bottlenecks. The recommended remedy is Continuous Delivery and related practices—small, safe, frequent releases (daily or multiple times per day), automated testing, CI/CD pipelines, feature flags, trunk-based development, and improved observability—to reduce batching and deployment overhead. In short, accelerate the entire pipeline, not just typing: turn the ratio so most time is in fast-feedback development and automated delivery, otherwise faster code generation simply creates bigger, more fragile queues.