Six months ago, our Acceleration Whiplash report found engineering organizations struggling to keep up with AI-assisted development. They have since adapted. AI coding is now standard practice, with 79% of developers using it weekly. Delivery has recovered: deployment frequency is up, throughput is strong, and churn is down sharply.

The pipeline is unblocked, but it’s also leaking.

Our latest Speed Trap report looks at what happens after the code gets written. Pull requests are 72% larger. PRs merged without review are up 76%. Time in QA has risen threefold, the steepest degradation of any flow metric we track. Restarts are up 67%, usually because an agent lacked the context to take the right path the first time. Each individual change is safer than it was in April, but there are far more of them, and monthly incidents are up 125%.

Speed at the start of the pipeline does not remove work. It moves the checking and cleanup downstream, to review, QA, SRE, and on-call, where capacity has not grown to match.

Join us for 45 minutes on what the data shows and where to intervene.

What you’ll learn 

• Where AI-scale output is creating strain in the delivery pipeline, and which metrics surface it
• Why agentic review eases that strain but also falls short when it comes to governance
• Why the highest-leverage fix is upstream at authoring, not downstream at review
• How to shift from measuring AI adoption to measuring AI outcomes