Jellyfish surveyed more than 600 engineering leaders for its 2026 State of Engineering Management report. 64% say their teams have seen velocity gains of 25% or more from AI tools. 84% say developer productivity is their top concern. 60% say they need better data to understand what is going on. 36% report resistance from senior engineers, 31% report tool fragmentation, and only 10% describe their enablement for these tools as strong.
I think the first number and the third one belong in the same sentence. Nearly two thirds of leaders report a large gain, and nearly two thirds say they do not have the data to be sure. That is what self-reported velocity is: what a leader believes after watching the pull request count go up.
Earlier this month I shared a preprint that looked at telemetry from the same kind of teams and found pull requests up 98%, review time up 91%, and delivery flat. Both pictures can be true. Output rose, people noticed, and the number that would tell you whether anything shipped faster is the one most teams are not measuring.
My view is that the 10% with strong enablement is the number to work on. Enablement is not training on the tool. It is the tests, the review capacity, the instructions file, the measurement, the things that turn more output into more delivery. The gain is real where those exist. Where they do not, the velocity is a feeling.
Does your team measure delivery any differently than it did before the AI tools, or just more of the same?
Photo source: https://photos.robertstowe.com/colorado